第16章 CNNを使った時系列のフィルタリング

In [8]:

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import tensorflow as tf
from tensorflow.keras import Model
from tensorflow.keras.losses import MeanSquaredError
from tensorflow.keras.metrics import MeanAbsoluteError
from tensorflow.keras.callbacks import EarlyStopping
from tensorflow.keras.optimizers import Adam
from tensorflow.keras import Sequential
from tensorflow.keras.layers import Dense, LSTM, Conv1D

import pandas as pd import numpy as np import matplotlib.pyplot as plt import tensorflow as tf from tensorflow.keras import Model from tensorflow.keras.losses import MeanSquaredError from tensorflow.keras.metrics import MeanAbsoluteError from tensorflow.keras.callbacks import EarlyStopping from tensorflow.keras.optimizers import Adam from tensorflow.keras import Sequential from tensorflow.keras.layers import Dense, LSTM, Conv1D

In [2]:

url_train = 'https://raw.githubusercontent.com/marcopeix/TimeSeriesForecastingInPython/master/data/train.csv'
url_val = 'https://raw.githubusercontent.com/marcopeix/TimeSeriesForecastingInPython/master/data/val.csv'
url_test = 'https://raw.githubusercontent.com/marcopeix/TimeSeriesForecastingInPython/master/data/test.csv'

df_train = pd.read_csv(url_train, index_col=0)
df_val = pd.read_csv(url_val, index_col=0)
df_test = pd.read_csv(url_test, index_col=0)

url_train = 'https://raw.githubusercontent.com/marcopeix/TimeSeriesForecastingInPython/master/data/train.csv' url_val = 'https://raw.githubusercontent.com/marcopeix/TimeSeriesForecastingInPython/master/data/val.csv' url_test = 'https://raw.githubusercontent.com/marcopeix/TimeSeriesForecastingInPython/master/data/test.csv' df_train = pd.read_csv(url_train, index_col=0) df_val = pd.read_csv(url_val, index_col=0) df_test = pd.read_csv(url_test, index_col=0)

In [3]:

class DataWindow:
    def __init__(self, input_width, label_width, shift, df_train, df_val, df_test, label_columns=None):
        # window size
        self.input_width = input_width
        self.label_width = label_width
        self.shift = shift
        self.total_window_size = input_width + shift
        
        # データ
        self.df_train = df_train
        self.df_val = df_val
        self.df_test = df_test
        
        # ラベル
        self.label_columns = label_columns
        if label_columns is not None:
            self.label_columns_indices = {name: i for i, name in enumerate(label_columns)}
        self.column_indices = {name: i for i, name in enumerate(self.df_train.columns)}
        
        # スライス
        self.input_slice = slice(0, input_width)
        self.input_indices = np.arange(self.total_window_size)[self.input_slice]
        
        # ラベル開始位置
        self.label_start = self.total_window_size - self.label_width
        self.labels_slice = slice(self.label_start, None)
        self.label_indices = np.arange(self.total_window_size)[self.labels_slice]
    
    def split_to_inputs_labels(self, features):
        inputs = features[:, self.input_slice, :]
        labels = features[:, self.labels_slice, :]
        if self.label_columns is not None:
            labels = tf.stack([labels[:, :, self.column_indices[name]] for name in self.label_columns], axis=-1)
        inputs.set_shape([None, self.input_width, None])
        labels.set_shape([None, self.label_width, None])
        return inputs, labels
    
    def plot(self, plot_col: str, model=None, max_subplots=3):
        inputs, labels = self.sample_batch
        plt.figure(figsize=(12, 8))
        plot_col_index = self.column_indices[plot_col]
        n_max = min(max_subplots, len(inputs))
        
        for n in range(n_max):
            plt.subplot(n_max, 1, n+1)
            plt.ylabel(f'{plot_col} [scaled]')
            plt.plot(self.input_indices, inputs[n, :, plot_col_index], label='Inputs', marker='.', zorder=-10)
            
            if self.label_columns:
                label_col_index = self.label_columns_indices.get(plot_col, None)
            else:
                label_col_index = plot_col_index
            
            if label_col_index is None:
                continue
            
            plt.scatter(self.label_indices, labels[n, :, label_col_index], edgecolors='k', label='Labels', c='tab:green', s=64)
            
            if model is not None:
                predictions = model(inputs)
                plt.scatter(self.label_indices, predictions[n, :, label_col_index], marker='X', edgecolors='k', label='Predictions', c='tab:red', s=64)
            
            if n == 0:
                plt.legend()
        plt.xlabel('Time (h)')
    
    def make_dataset(self, data):
        data = np.array(data, dtype=np.float32)
        ds = tf.keras.utils.timeseries_dataset_from_array(
            data=data,
            targets=None,
            sequence_length=self.total_window_size,
            sequence_stride=1,
            shuffle=True,
            batch_size=32,
        )
        ds = ds.map(self.split_to_inputs_labels)
        return ds
    
    @property
    def train(self):
        return self.make_dataset(self.df_train)
    
    @property
    def val(self):
        return self.make_dataset(self.df_val)
    
    @property
    def test(self):
        return self.make_dataset(self.df_test)
    
    @property
    def sample_batch(self):
        """Get and cache an example batch of `inputs, labels` for plotting."""
        result = getattr(self, '_sample_batch', None)
        if result is None:
            result = next(iter(self.train))
            self._sample_batch = result
        return result

class DataWindow: def __init__(self, input_width, label_width, shift, df_train, df_val, df_test, label_columns=None): # window size self.input_width = input_width self.label_width = label_width self.shift = shift self.total_window_size = input_width + shift # データ self.df_train = df_train self.df_val = df_val self.df_test = df_test # ラベル self.label_columns = label_columns if label_columns is not None: self.label_columns_indices = {name: i for i, name in enumerate(label_columns)} self.column_indices = {name: i for i, name in enumerate(self.df_train.columns)} # スライス self.input_slice = slice(0, input_width) self.input_indices = np.arange(self.total_window_size)[self.input_slice] # ラベル開始位置 self.label_start = self.total_window_size - self.label_width self.labels_slice = slice(self.label_start, None) self.label_indices = np.arange(self.total_window_size)[self.labels_slice] def split_to_inputs_labels(self, features): inputs = features[:, self.input_slice, :] labels = features[:, self.labels_slice, :] if self.label_columns is not None: labels = tf.stack([labels[:, :, self.column_indices[name]] for name in self.label_columns], axis=-1) inputs.set_shape([None, self.input_width, None]) labels.set_shape([None, self.label_width, None]) return inputs, labels def plot(self, plot_col: str, model=None, max_subplots=3): inputs, labels = self.sample_batch plt.figure(figsize=(12, 8)) plot_col_index = self.column_indices[plot_col] n_max = min(max_subplots, len(inputs)) for n in range(n_max): plt.subplot(n_max, 1, n+1) plt.ylabel(f'{plot_col} [scaled]') plt.plot(self.input_indices, inputs[n, :, plot_col_index], label='Inputs', marker='.', zorder=-10) if self.label_columns: label_col_index = self.label_columns_indices.get(plot_col, None) else: label_col_index = plot_col_index if label_col_index is None: continue plt.scatter(self.label_indices, labels[n, :, label_col_index], edgecolors='k', label='Labels', c='tab:green', s=64) if model is not None: predictions = model(inputs) plt.scatter(self.label_indices, predictions[n, :, label_col_index], marker='X', edgecolors='k', label='Predictions', c='tab:red', s=64) if n == 0: plt.legend() plt.xlabel('Time (h)') def make_dataset(self, data): data = np.array(data, dtype=np.float32) ds = tf.keras.utils.timeseries_dataset_from_array( data=data, targets=None, sequence_length=self.total_window_size, sequence_stride=1, shuffle=True, batch_size=32, ) ds = ds.map(self.split_to_inputs_labels) return ds @property def train(self): return self.make_dataset(self.df_train) @property def val(self): return self.make_dataset(self.df_val) @property def test(self): return self.make_dataset(self.df_test) @property def sample_batch(self): """Get and cache an example batch of `inputs, labels` for plotting.""" result = getattr(self, '_sample_batch', None) if result is None: result = next(iter(self.train)) self._sample_batch = result return result

In [4]:

# for training
def compile_and_fit(model, window, patience=3, max_epochs=50):
    early_stopping = EarlyStopping(
        monitor='val_loss',
        patience=patience,
        mode='min'
    )
    model.compile(
        loss=MeanSquaredError(),
        optimizer=Adam(),
        metrics=[MeanAbsoluteError()]
    )
    history = model.fit(
        window.train,
        epochs=max_epochs,
        validation_data=window.val,
        callbacks=[early_stopping]
    )
    return history

# for training def compile_and_fit(model, window, patience=3, max_epochs=50): early_stopping = EarlyStopping( monitor='val_loss', patience=patience, mode='min' ) model.compile( loss=MeanSquaredError(), optimizer=Adam(), metrics=[MeanAbsoluteError()] ) history = model.fit( window.train, epochs=max_epochs, validation_data=window.val, callbacks=[early_stopping] ) return history

In [6]:

# models
class Baseline(Model):
    def __init__(self, label_index=None):
        super().__init__()
        self.label_index = label_index
        
    def call(self, inputs):
        if self.label_index is None:
            return inputs
        elif isinstance(self.label_index, list):
            tensors = []
            for index in self.label_index:
                res = inputs[:, :, index]
                res = res[:, :, tf.newaxis]
                tensors.append(res)
            return tf.concat(tensors, axis=-1)
        else:
            res = inputs[:, :, self.label_index]
            return res[:, :, tf.newaxis]

class MultiStepLastBaseline(Model):
    def __init__(self, label_index=None):
        super().__init__()
        self.label_index = label_index
    
    def call(self, inputs):
        if self.label_index is None:
            return tf.tile(inputs[:, -1:, :], [1, 24, 1])
        return tf.tile(inputs[:, -1:, self.label_index:], [1, 24, 1])
    
class RepeatBaseline(Model):
    def __init__(self, label_index=None):
        super().__init__()
        self.label_index = label_index
    
    def call(self, inputs):
        return inputs[:, :, self.label_index:]

# 線形モデル
linear = Sequential([Dense(units=1)])
ms_linear = Sequential([Dense(units=1, kernel_initializer=tf.initializers.zeros)])
mo_linear = Sequential([Dense(units=2)])
# DNN
dense = Sequential([
    Dense(units=64, activation='relu'),
    Dense(units=64, activation='relu'),
    Dense(units=1)
])
ms_dense = Sequential([
    Dense(units=64, activation='relu'),
    Dense(units=64, activation='relu'),
    Dense(units=1, kernel_initializer=tf.initializers.zeros)
])
mo_dense = Sequential([
    Dense(units=64, activation='relu'),
    Dense(units=64, activation='relu'),
    Dense(units=2)
])
# LSTM
lstm_model = Sequential([
    LSTM(32, return_sequences=True),
    Dense(units=1)
])
ms_lstm_model = Sequential([
    LSTM(32, return_sequences=True),
    Dense(units=1, kernel_initializer=tf.initializers.zeros)
])
mo_lstm_model = Sequential([
    LSTM(32, return_sequences=True),
    Dense(units=2)
])

# models class Baseline(Model): def __init__(self, label_index=None): super().__init__() self.label_index = label_index def call(self, inputs): if self.label_index is None: return inputs elif isinstance(self.label_index, list): tensors = [] for index in self.label_index: res = inputs[:, :, index] res = res[:, :, tf.newaxis] tensors.append(res) return tf.concat(tensors, axis=-1) else: res = inputs[:, :, self.label_index] return res[:, :, tf.newaxis] class MultiStepLastBaseline(Model): def __init__(self, label_index=None): super().__init__() self.label_index = label_index def call(self, inputs): if self.label_index is None: return tf.tile(inputs[:, -1:, :], [1, 24, 1]) return tf.tile(inputs[:, -1:, self.label_index:], [1, 24, 1]) class RepeatBaseline(Model): def __init__(self, label_index=None): super().__init__() self.label_index = label_index def call(self, inputs): return inputs[:, :, self.label_index:] # 線形モデル linear = Sequential([Dense(units=1)]) ms_linear = Sequential([Dense(units=1, kernel_initializer=tf.initializers.zeros)]) mo_linear = Sequential([Dense(units=2)]) # DNN dense = Sequential([ Dense(units=64, activation='relu'), Dense(units=64, activation='relu'), Dense(units=1) ]) ms_dense = Sequential([ Dense(units=64, activation='relu'), Dense(units=64, activation='relu'), Dense(units=1, kernel_initializer=tf.initializers.zeros) ]) mo_dense = Sequential([ Dense(units=64, activation='relu'), Dense(units=64, activation='relu'), Dense(units=2) ]) # LSTM lstm_model = Sequential([ LSTM(32, return_sequences=True), Dense(units=1) ]) ms_lstm_model = Sequential([ LSTM(32, return_sequences=True), Dense(units=1, kernel_initializer=tf.initializers.zeros) ]) mo_lstm_model = Sequential([ LSTM(32, return_sequences=True), Dense(units=2) ])

In [7]:

# シングルステップ
single_step_window = DataWindow(input_width=1, label_width=1, shift=1, df_train=df_train, df_val=df_val, df_test=df_test, label_columns=['traffic_volume'])
wide_window = DataWindow(input_width=24, label_width=24, shift=1, df_train=df_train, df_val=df_val, df_test=df_test, label_columns=['traffic_volume'])
column_indices = {name: i for i, name in enumerate(df_train.columns)}
baseline_last = Baseline(column_indices['traffic_volume'])
baseline_last.compile(loss=MeanSquaredError(), metrics=[MeanAbsoluteError()])

val_performance = {}
test_performance = {}

# マルチステップ
multi_window = DataWindow(input_width=24, label_width=24, shift=24, df_train=df_train, df_val=df_val, df_test=df_test, label_columns=['traffic_volume'])
ms_baseline_last = MultiStepLastBaseline(label_index=column_indices['traffic_volume'])
ms_baseline_last.compile(loss=MeanSquaredError(), metrics=[MeanAbsoluteError()])
ms_baseline_repeat = RepeatBaseline(label_index=column_indices['traffic_volume'])
ms_baseline_repeat.compile(loss=MeanSquaredError(), metrics=[MeanAbsoluteError()])

ms_val_performance = {}
ms_test_performance = {}

# 多変数アウトプット
col_names = ['temp', 'traffic_volume']
mo_single_step_window = DataWindow(input_width=1, label_width=1, shift=1, df_train=df_train, df_val=df_val, df_test=df_test, label_columns=col_names)
mo_wide_window = DataWindow(input_width=24, label_width=24, shift=1, df_train=df_train, df_val=df_val, df_test=df_test, label_columns=col_names)
mo_baseline_last = Baseline(label_index=[column_indices[col] for col in col_names])
mo_baseline_last.compile(loss=MeanSquaredError(), metrics=[MeanAbsoluteError()])

mo_val_performance = {}
mo_test_performance = {}

## ベースライン
val_performance['Baseline - Last'] = baseline_last.evaluate(single_step_window.val)
test_performance['Baseline - Last'] = baseline_last.evaluate(single_step_window.test, verbose=0)
ms_val_performance['Baseline - Last'] = ms_baseline_last.evaluate(multi_window.val)
ms_test_performance['Baseline - Last'] = ms_baseline_last.evaluate(multi_window.test)
ms_val_performance['Baseline - Repeat'] = ms_baseline_repeat.evaluate(multi_window.val)
ms_test_performance['Baseline - Repeat'] = ms_baseline_repeat.evaluate(multi_window.test)
mo_val_performance['Baseline - Last'] = mo_baseline_last.evaluate(mo_single_step_window.val)
mo_test_performance['Baseline - Last'] = mo_baseline_last.evaluate(mo_single_step_window.val)

## 線形モデル
history = compile_and_fit(linear, single_step_window)
val_performance['Linear'] = linear.evaluate(single_step_window.val)
test_performance['Linear'] = linear.evaluate(single_step_window.test, verbose=0)
history = compile_and_fit(ms_linear, multi_window)
ms_val_performance['Linear'] = ms_linear.evaluate(multi_window.val)
ms_test_performance['Linear'] = ms_linear.evaluate(multi_window.test)
history = compile_and_fit(mo_linear, mo_single_step_window)
mo_val_performance['Linear'] = mo_linear.evaluate(mo_single_step_window.val)
mo_test_performance['Linear'] = mo_linear.evaluate(mo_single_step_window.test)

## DNN
history = compile_and_fit(dense, single_step_window)
val_performance['Dense'] = dense.evaluate(single_step_window.val)
test_performance['Dense'] = dense.evaluate(single_step_window.test, verbose=0)
history = compile_and_fit(ms_dense, multi_window)
ms_val_performance['Dense'] = ms_dense.evaluate(multi_window.val)
ms_test_performance['Dense'] = ms_dense.evaluate(multi_window.test)
history = compile_and_fit(mo_dense, mo_single_step_window)
mo_val_performance['Dense'] = mo_dense.evaluate(mo_single_step_window.val)
mo_test_performance['Dense'] = mo_dense.evaluate(mo_single_step_window.test)

## LSTM
history = compile_and_fit(lstm_model, wide_window)
val_performance['LSTM'] = lstm_model.evaluate(wide_window.val)
test_performance['LSTM'] = lstm_model.evaluate(wide_window.test, verbose=0)
history = compile_and_fit(ms_lstm_model, multi_window)
ms_val_performance['LSTM'] = ms_lstm_model.evaluate(multi_window.val)
ms_test_performance['LSTM'] = ms_lstm_model.evaluate(multi_window.test)
history = compile_and_fit(mo_lstm_model, mo_wide_window)
mo_val_performance['LSTM'] = mo_lstm_model.evaluate(mo_wide_window.val)
mo_test_performance['LSTM'] = mo_lstm_model.evaluate(mo_wide_window.test)

# シングルステップ single_step_window = DataWindow(input_width=1, label_width=1, shift=1, df_train=df_train, df_val=df_val, df_test=df_test, label_columns=['traffic_volume']) wide_window = DataWindow(input_width=24, label_width=24, shift=1, df_train=df_train, df_val=df_val, df_test=df_test, label_columns=['traffic_volume']) column_indices = {name: i for i, name in enumerate(df_train.columns)} baseline_last = Baseline(column_indices['traffic_volume']) baseline_last.compile(loss=MeanSquaredError(), metrics=[MeanAbsoluteError()]) val_performance = {} test_performance = {} # マルチステップ multi_window = DataWindow(input_width=24, label_width=24, shift=24, df_train=df_train, df_val=df_val, df_test=df_test, label_columns=['traffic_volume']) ms_baseline_last = MultiStepLastBaseline(label_index=column_indices['traffic_volume']) ms_baseline_last.compile(loss=MeanSquaredError(), metrics=[MeanAbsoluteError()]) ms_baseline_repeat = RepeatBaseline(label_index=column_indices['traffic_volume']) ms_baseline_repeat.compile(loss=MeanSquaredError(), metrics=[MeanAbsoluteError()]) ms_val_performance = {} ms_test_performance = {} # 多変数アウトプット col_names = ['temp', 'traffic_volume'] mo_single_step_window = DataWindow(input_width=1, label_width=1, shift=1, df_train=df_train, df_val=df_val, df_test=df_test, label_columns=col_names) mo_wide_window = DataWindow(input_width=24, label_width=24, shift=1, df_train=df_train, df_val=df_val, df_test=df_test, label_columns=col_names) mo_baseline_last = Baseline(label_index=[column_indices[col] for col in col_names]) mo_baseline_last.compile(loss=MeanSquaredError(), metrics=[MeanAbsoluteError()]) mo_val_performance = {} mo_test_performance = {} ## ベースライン val_performance['Baseline - Last'] = baseline_last.evaluate(single_step_window.val) test_performance['Baseline - Last'] = baseline_last.evaluate(single_step_window.test, verbose=0) ms_val_performance['Baseline - Last'] = ms_baseline_last.evaluate(multi_window.val) ms_test_performance['Baseline - Last'] = ms_baseline_last.evaluate(multi_window.test) ms_val_performance['Baseline - Repeat'] = ms_baseline_repeat.evaluate(multi_window.val) ms_test_performance['Baseline - Repeat'] = ms_baseline_repeat.evaluate(multi_window.test) mo_val_performance['Baseline - Last'] = mo_baseline_last.evaluate(mo_single_step_window.val) mo_test_performance['Baseline - Last'] = mo_baseline_last.evaluate(mo_single_step_window.val) ## 線形モデル history = compile_and_fit(linear, single_step_window) val_performance['Linear'] = linear.evaluate(single_step_window.val) test_performance['Linear'] = linear.evaluate(single_step_window.test, verbose=0) history = compile_and_fit(ms_linear, multi_window) ms_val_performance['Linear'] = ms_linear.evaluate(multi_window.val) ms_test_performance['Linear'] = ms_linear.evaluate(multi_window.test) history = compile_and_fit(mo_linear, mo_single_step_window) mo_val_performance['Linear'] = mo_linear.evaluate(mo_single_step_window.val) mo_test_performance['Linear'] = mo_linear.evaluate(mo_single_step_window.test) ## DNN history = compile_and_fit(dense, single_step_window) val_performance['Dense'] = dense.evaluate(single_step_window.val) test_performance['Dense'] = dense.evaluate(single_step_window.test, verbose=0) history = compile_and_fit(ms_dense, multi_window) ms_val_performance['Dense'] = ms_dense.evaluate(multi_window.val) ms_test_performance['Dense'] = ms_dense.evaluate(multi_window.test) history = compile_and_fit(mo_dense, mo_single_step_window) mo_val_performance['Dense'] = mo_dense.evaluate(mo_single_step_window.val) mo_test_performance['Dense'] = mo_dense.evaluate(mo_single_step_window.test) ## LSTM history = compile_and_fit(lstm_model, wide_window) val_performance['LSTM'] = lstm_model.evaluate(wide_window.val) test_performance['LSTM'] = lstm_model.evaluate(wide_window.test, verbose=0) history = compile_and_fit(ms_lstm_model, multi_window) ms_val_performance['LSTM'] = ms_lstm_model.evaluate(multi_window.val) ms_test_performance['LSTM'] = ms_lstm_model.evaluate(multi_window.test) history = compile_and_fit(mo_lstm_model, mo_wide_window) mo_val_performance['LSTM'] = mo_lstm_model.evaluate(mo_wide_window.val) mo_test_performance['LSTM'] = mo_lstm_model.evaluate(mo_wide_window.test)

110/110 [==============================] - 0s 3ms/step - loss: 0.0133 - mean_absolute_error: 0.0831
109/109 [==============================] - 0s 2ms/step - loss: 0.1875 - mean_absolute_error: 0.3522
54/54 [==============================] - 0s 2ms/step - loss: 0.1814 - mean_absolute_error: 0.3473
109/109 [==============================] - 0s 2ms/step - loss: 0.2065 - mean_absolute_error: 0.3473
54/54 [==============================] - 0s 2ms/step - loss: 0.2018 - mean_absolute_error: 0.3413
110/110 [==============================] - 0s 2ms/step - loss: 0.0069 - mean_absolute_error: 0.0482
110/110 [==============================] - 0s 2ms/step - loss: 0.0069 - mean_absolute_error: 0.0482
Epoch 1/50
384/384 [==============================] - 1s 3ms/step - loss: 0.2975 - mean_absolute_error: 0.4443 - val_loss: 0.0746 - val_mean_absolute_error: 0.2244
Epoch 2/50
384/384 [==============================] - 1s 3ms/step - loss: 0.0617 - mean_absolute_error: 0.2041 - val_loss: 0.0539 - val_mean_absolute_error: 0.1910
Epoch 3/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0412 - mean_absolute_error: 0.1653 - val_loss: 0.0386 - val_mean_absolute_error: 0.1619
Epoch 4/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0304 - mean_absolute_error: 0.1407 - val_loss: 0.0300 - val_mean_absolute_error: 0.1417
Epoch 5/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0243 - mean_absolute_error: 0.1244 - val_loss: 0.0242 - val_mean_absolute_error: 0.1260
Epoch 6/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0200 - mean_absolute_error: 0.1115 - val_loss: 0.0199 - val_mean_absolute_error: 0.1131
Epoch 7/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0166 - mean_absolute_error: 0.1000 - val_loss: 0.0163 - val_mean_absolute_error: 0.1011
Epoch 8/50
384/384 [==============================] - 3s 7ms/step - loss: 0.0140 - mean_absolute_error: 0.0902 - val_loss: 0.0137 - val_mean_absolute_error: 0.0907
Epoch 9/50
384/384 [==============================] - 3s 8ms/step - loss: 0.0121 - mean_absolute_error: 0.0822 - val_loss: 0.0119 - val_mean_absolute_error: 0.0831
Epoch 10/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0109 - mean_absolute_error: 0.0767 - val_loss: 0.0107 - val_mean_absolute_error: 0.0768
Epoch 11/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0102 - mean_absolute_error: 0.0732 - val_loss: 0.0100 - val_mean_absolute_error: 0.0722
Epoch 12/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0098 - mean_absolute_error: 0.0712 - val_loss: 0.0096 - val_mean_absolute_error: 0.0694
Epoch 13/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0095 - mean_absolute_error: 0.0700 - val_loss: 0.0094 - val_mean_absolute_error: 0.0679
Epoch 14/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0094 - mean_absolute_error: 0.0692 - val_loss: 0.0092 - val_mean_absolute_error: 0.0666
Epoch 15/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0094 - mean_absolute_error: 0.0687 - val_loss: 0.0092 - val_mean_absolute_error: 0.0665
Epoch 16/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0093 - mean_absolute_error: 0.0684 - val_loss: 0.0091 - val_mean_absolute_error: 0.0653
Epoch 17/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0093 - mean_absolute_error: 0.0682 - val_loss: 0.0091 - val_mean_absolute_error: 0.0654
Epoch 18/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0093 - mean_absolute_error: 0.0680 - val_loss: 0.0091 - val_mean_absolute_error: 0.0661
Epoch 19/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0093 - mean_absolute_error: 0.0680 - val_loss: 0.0091 - val_mean_absolute_error: 0.0655
Epoch 20/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0093 - mean_absolute_error: 0.0678 - val_loss: 0.0091 - val_mean_absolute_error: 0.0649
Epoch 21/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0093 - mean_absolute_error: 0.0677 - val_loss: 0.0091 - val_mean_absolute_error: 0.0648
Epoch 22/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0093 - mean_absolute_error: 0.0677 - val_loss: 0.0091 - val_mean_absolute_error: 0.0659
Epoch 23/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0093 - mean_absolute_error: 0.0677 - val_loss: 0.0090 - val_mean_absolute_error: 0.0656
Epoch 24/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0093 - mean_absolute_error: 0.0677 - val_loss: 0.0091 - val_mean_absolute_error: 0.0662
Epoch 25/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0093 - mean_absolute_error: 0.0676 - val_loss: 0.0091 - val_mean_absolute_error: 0.0661
Epoch 26/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0093 - mean_absolute_error: 0.0678 - val_loss: 0.0091 - val_mean_absolute_error: 0.0651
110/110 [==============================] - 0s 3ms/step - loss: 0.0091 - mean_absolute_error: 0.0651
Epoch 1/50
383/383 [==============================] - 2s 4ms/step - loss: 0.0898 - mean_absolute_error: 0.2464 - val_loss: 0.0458 - val_mean_absolute_error: 0.1822
Epoch 2/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0298 - mean_absolute_error: 0.1368 - val_loss: 0.0261 - val_mean_absolute_error: 0.1254
Epoch 3/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0201 - mean_absolute_error: 0.1044 - val_loss: 0.0212 - val_mean_absolute_error: 0.1074
Epoch 4/50
383/383 [==============================] - 2s 6ms/step - loss: 0.0177 - mean_absolute_error: 0.0954 - val_loss: 0.0196 - val_mean_absolute_error: 0.1001
Epoch 5/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0168 - mean_absolute_error: 0.0914 - val_loss: 0.0188 - val_mean_absolute_error: 0.0952
Epoch 6/50
383/383 [==============================] - 2s 6ms/step - loss: 0.0164 - mean_absolute_error: 0.0887 - val_loss: 0.0184 - val_mean_absolute_error: 0.0922
Epoch 7/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0162 - mean_absolute_error: 0.0868 - val_loss: 0.0183 - val_mean_absolute_error: 0.0902
Epoch 8/50
383/383 [==============================] - 3s 8ms/step - loss: 0.0161 - mean_absolute_error: 0.0857 - val_loss: 0.0182 - val_mean_absolute_error: 0.0892
Epoch 9/50
383/383 [==============================] - 2s 6ms/step - loss: 0.0161 - mean_absolute_error: 0.0850 - val_loss: 0.0182 - val_mean_absolute_error: 0.0886
Epoch 10/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0161 - mean_absolute_error: 0.0846 - val_loss: 0.0182 - val_mean_absolute_error: 0.0882
Epoch 11/50
383/383 [==============================] - 3s 6ms/step - loss: 0.0161 - mean_absolute_error: 0.0843 - val_loss: 0.0182 - val_mean_absolute_error: 0.0881
Epoch 12/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0161 - mean_absolute_error: 0.0841 - val_loss: 0.0182 - val_mean_absolute_error: 0.0882
Epoch 13/50
383/383 [==============================] - 3s 6ms/step - loss: 0.0161 - mean_absolute_error: 0.0840 - val_loss: 0.0182 - val_mean_absolute_error: 0.0879
Epoch 14/50
383/383 [==============================] - 3s 7ms/step - loss: 0.0161 - mean_absolute_error: 0.0839 - val_loss: 0.0182 - val_mean_absolute_error: 0.0879
Epoch 15/50
383/383 [==============================] - 2s 6ms/step - loss: 0.0161 - mean_absolute_error: 0.0838 - val_loss: 0.0182 - val_mean_absolute_error: 0.0879
109/109 [==============================] - 0s 3ms/step - loss: 0.0182 - mean_absolute_error: 0.0879
54/54 [==============================] - 0s 3ms/step - loss: 0.0142 - mean_absolute_error: 0.0753
Epoch 1/50
384/384 [==============================] - 2s 4ms/step - loss: 0.2482 - mean_absolute_error: 0.3754 - val_loss: 0.0894 - val_mean_absolute_error: 0.2307
Epoch 2/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0494 - mean_absolute_error: 0.1644 - val_loss: 0.0418 - val_mean_absolute_error: 0.1526
Epoch 3/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0359 - mean_absolute_error: 0.1355 - val_loss: 0.0332 - val_mean_absolute_error: 0.1363
Epoch 4/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0292 - mean_absolute_error: 0.1238 - val_loss: 0.0271 - val_mean_absolute_error: 0.1251
Epoch 5/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0239 - mean_absolute_error: 0.1140 - val_loss: 0.0222 - val_mean_absolute_error: 0.1156
Epoch 6/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0194 - mean_absolute_error: 0.1045 - val_loss: 0.0179 - val_mean_absolute_error: 0.1047
Epoch 7/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0157 - mean_absolute_error: 0.0950 - val_loss: 0.0143 - val_mean_absolute_error: 0.0954
Epoch 8/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0126 - mean_absolute_error: 0.0857 - val_loss: 0.0112 - val_mean_absolute_error: 0.0842
Epoch 9/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0101 - mean_absolute_error: 0.0765 - val_loss: 0.0089 - val_mean_absolute_error: 0.0749
Epoch 10/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0081 - mean_absolute_error: 0.0680 - val_loss: 0.0071 - val_mean_absolute_error: 0.0652
Epoch 11/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0068 - mean_absolute_error: 0.0602 - val_loss: 0.0060 - val_mean_absolute_error: 0.0581
Epoch 12/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0059 - mean_absolute_error: 0.0535 - val_loss: 0.0053 - val_mean_absolute_error: 0.0512
Epoch 13/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0054 - mean_absolute_error: 0.0483 - val_loss: 0.0049 - val_mean_absolute_error: 0.0461
Epoch 14/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0051 - mean_absolute_error: 0.0446 - val_loss: 0.0048 - val_mean_absolute_error: 0.0430
Epoch 15/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0050 - mean_absolute_error: 0.0422 - val_loss: 0.0047 - val_mean_absolute_error: 0.0409
Epoch 16/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0050 - mean_absolute_error: 0.0408 - val_loss: 0.0047 - val_mean_absolute_error: 0.0398
Epoch 17/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0049 - mean_absolute_error: 0.0402 - val_loss: 0.0047 - val_mean_absolute_error: 0.0398
Epoch 18/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0049 - mean_absolute_error: 0.0398 - val_loss: 0.0047 - val_mean_absolute_error: 0.0391
Epoch 19/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0049 - mean_absolute_error: 0.0396 - val_loss: 0.0046 - val_mean_absolute_error: 0.0385
Epoch 20/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0049 - mean_absolute_error: 0.0395 - val_loss: 0.0046 - val_mean_absolute_error: 0.0386
Epoch 21/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0049 - mean_absolute_error: 0.0395 - val_loss: 0.0047 - val_mean_absolute_error: 0.0389
Epoch 22/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0049 - mean_absolute_error: 0.0395 - val_loss: 0.0047 - val_mean_absolute_error: 0.0402
110/110 [==============================] - 0s 3ms/step - loss: 0.0047 - mean_absolute_error: 0.0402
55/55 [==============================] - 0s 3ms/step - loss: 0.0044 - mean_absolute_error: 0.0384
Epoch 1/50
384/384 [==============================] - 3s 6ms/step - loss: 0.0198 - mean_absolute_error: 0.0841 - val_loss: 0.0054 - val_mean_absolute_error: 0.0540
Epoch 2/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0051 - mean_absolute_error: 0.0518 - val_loss: 0.0041 - val_mean_absolute_error: 0.0472
Epoch 3/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0043 - mean_absolute_error: 0.0478 - val_loss: 0.0035 - val_mean_absolute_error: 0.0434
Epoch 4/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0038 - mean_absolute_error: 0.0450 - val_loss: 0.0032 - val_mean_absolute_error: 0.0410
Epoch 5/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0036 - mean_absolute_error: 0.0434 - val_loss: 0.0035 - val_mean_absolute_error: 0.0452
Epoch 6/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0034 - mean_absolute_error: 0.0423 - val_loss: 0.0027 - val_mean_absolute_error: 0.0380
Epoch 7/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0032 - mean_absolute_error: 0.0411 - val_loss: 0.0034 - val_mean_absolute_error: 0.0463
Epoch 8/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0031 - mean_absolute_error: 0.0402 - val_loss: 0.0026 - val_mean_absolute_error: 0.0371
Epoch 9/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0029 - mean_absolute_error: 0.0390 - val_loss: 0.0026 - val_mean_absolute_error: 0.0376
Epoch 10/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0029 - mean_absolute_error: 0.0386 - val_loss: 0.0024 - val_mean_absolute_error: 0.0364
Epoch 11/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0028 - mean_absolute_error: 0.0376 - val_loss: 0.0030 - val_mean_absolute_error: 0.0426
Epoch 12/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0026 - mean_absolute_error: 0.0363 - val_loss: 0.0021 - val_mean_absolute_error: 0.0331
Epoch 13/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0026 - mean_absolute_error: 0.0361 - val_loss: 0.0022 - val_mean_absolute_error: 0.0338
Epoch 14/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0026 - mean_absolute_error: 0.0358 - val_loss: 0.0024 - val_mean_absolute_error: 0.0350
Epoch 15/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0025 - mean_absolute_error: 0.0349 - val_loss: 0.0023 - val_mean_absolute_error: 0.0339
110/110 [==============================] - 0s 3ms/step - loss: 0.0023 - mean_absolute_error: 0.0339
Epoch 1/50
383/383 [==============================] - 3s 5ms/step - loss: 0.0308 - mean_absolute_error: 0.1119 - val_loss: 0.0166 - val_mean_absolute_error: 0.0857
Epoch 2/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0140 - mean_absolute_error: 0.0785 - val_loss: 0.0156 - val_mean_absolute_error: 0.0842
Epoch 3/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0135 - mean_absolute_error: 0.0777 - val_loss: 0.0152 - val_mean_absolute_error: 0.0805
Epoch 4/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0133 - mean_absolute_error: 0.0769 - val_loss: 0.0151 - val_mean_absolute_error: 0.0809
Epoch 5/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0131 - mean_absolute_error: 0.0765 - val_loss: 0.0148 - val_mean_absolute_error: 0.0808
Epoch 6/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0130 - mean_absolute_error: 0.0762 - val_loss: 0.0150 - val_mean_absolute_error: 0.0803
Epoch 7/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0130 - mean_absolute_error: 0.0761 - val_loss: 0.0147 - val_mean_absolute_error: 0.0795
Epoch 8/50
383/383 [==============================] - 2s 6ms/step - loss: 0.0129 - mean_absolute_error: 0.0756 - val_loss: 0.0148 - val_mean_absolute_error: 0.0820
Epoch 9/50
383/383 [==============================] - 2s 6ms/step - loss: 0.0128 - mean_absolute_error: 0.0753 - val_loss: 0.0148 - val_mean_absolute_error: 0.0808
Epoch 10/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0128 - mean_absolute_error: 0.0752 - val_loss: 0.0146 - val_mean_absolute_error: 0.0773
Epoch 11/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0128 - mean_absolute_error: 0.0753 - val_loss: 0.0147 - val_mean_absolute_error: 0.0787
Epoch 12/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0127 - mean_absolute_error: 0.0748 - val_loss: 0.0145 - val_mean_absolute_error: 0.0783
Epoch 13/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0127 - mean_absolute_error: 0.0747 - val_loss: 0.0147 - val_mean_absolute_error: 0.0769
Epoch 14/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0127 - mean_absolute_error: 0.0747 - val_loss: 0.0145 - val_mean_absolute_error: 0.0789
Epoch 15/50
383/383 [==============================] - 3s 7ms/step - loss: 0.0126 - mean_absolute_error: 0.0747 - val_loss: 0.0144 - val_mean_absolute_error: 0.0790
Epoch 16/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0126 - mean_absolute_error: 0.0744 - val_loss: 0.0144 - val_mean_absolute_error: 0.0792
Epoch 17/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0125 - mean_absolute_error: 0.0742 - val_loss: 0.0148 - val_mean_absolute_error: 0.0785
Epoch 18/50
383/383 [==============================] - 2s 4ms/step - loss: 0.0125 - mean_absolute_error: 0.0741 - val_loss: 0.0144 - val_mean_absolute_error: 0.0770
Epoch 19/50
383/383 [==============================] - 2s 4ms/step - loss: 0.0124 - mean_absolute_error: 0.0739 - val_loss: 0.0144 - val_mean_absolute_error: 0.0779
Epoch 20/50
383/383 [==============================] - 2s 4ms/step - loss: 0.0124 - mean_absolute_error: 0.0739 - val_loss: 0.0144 - val_mean_absolute_error: 0.0809
Epoch 21/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0124 - mean_absolute_error: 0.0738 - val_loss: 0.0144 - val_mean_absolute_error: 0.0779
Epoch 22/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0123 - mean_absolute_error: 0.0735 - val_loss: 0.0144 - val_mean_absolute_error: 0.0801
Epoch 23/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0122 - mean_absolute_error: 0.0732 - val_loss: 0.0148 - val_mean_absolute_error: 0.0776
Epoch 24/50
383/383 [==============================] - 2s 4ms/step - loss: 0.0122 - mean_absolute_error: 0.0732 - val_loss: 0.0142 - val_mean_absolute_error: 0.0771
Epoch 25/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0122 - mean_absolute_error: 0.0730 - val_loss: 0.0142 - val_mean_absolute_error: 0.0777
Epoch 26/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0121 - mean_absolute_error: 0.0728 - val_loss: 0.0142 - val_mean_absolute_error: 0.0764
Epoch 27/50
383/383 [==============================] - 2s 4ms/step - loss: 0.0121 - mean_absolute_error: 0.0729 - val_loss: 0.0144 - val_mean_absolute_error: 0.0765
Epoch 28/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0121 - mean_absolute_error: 0.0726 - val_loss: 0.0142 - val_mean_absolute_error: 0.0780
Epoch 29/50
383/383 [==============================] - 2s 4ms/step - loss: 0.0120 - mean_absolute_error: 0.0724 - val_loss: 0.0144 - val_mean_absolute_error: 0.0761
Epoch 30/50
383/383 [==============================] - 2s 4ms/step - loss: 0.0120 - mean_absolute_error: 0.0722 - val_loss: 0.0140 - val_mean_absolute_error: 0.0767
Epoch 31/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0119 - mean_absolute_error: 0.0719 - val_loss: 0.0141 - val_mean_absolute_error: 0.0764
Epoch 32/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0119 - mean_absolute_error: 0.0719 - val_loss: 0.0141 - val_mean_absolute_error: 0.0758
Epoch 33/50
383/383 [==============================] - 2s 4ms/step - loss: 0.0119 - mean_absolute_error: 0.0718 - val_loss: 0.0142 - val_mean_absolute_error: 0.0759
109/109 [==============================] - 0s 3ms/step - loss: 0.0142 - mean_absolute_error: 0.0759
54/54 [==============================] - 0s 3ms/step - loss: 0.0093 - mean_absolute_error: 0.0600
Epoch 1/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0100 - mean_absolute_error: 0.0581 - val_loss: 0.0031 - val_mean_absolute_error: 0.0402
Epoch 2/50
384/384 [==============================] - 1s 4ms/step - loss: 0.0029 - mean_absolute_error: 0.0343 - val_loss: 0.0024 - val_mean_absolute_error: 0.0316
Epoch 3/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0024 - mean_absolute_error: 0.0313 - val_loss: 0.0023 - val_mean_absolute_error: 0.0319
Epoch 4/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0022 - mean_absolute_error: 0.0292 - val_loss: 0.0017 - val_mean_absolute_error: 0.0273
Epoch 5/50
384/384 [==============================] - 1s 4ms/step - loss: 0.0021 - mean_absolute_error: 0.0287 - val_loss: 0.0016 - val_mean_absolute_error: 0.0254
Epoch 6/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0020 - mean_absolute_error: 0.0274 - val_loss: 0.0017 - val_mean_absolute_error: 0.0271
Epoch 7/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0020 - mean_absolute_error: 0.0277 - val_loss: 0.0016 - val_mean_absolute_error: 0.0256
Epoch 8/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0020 - mean_absolute_error: 0.0273 - val_loss: 0.0022 - val_mean_absolute_error: 0.0315
Epoch 9/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0019 - mean_absolute_error: 0.0268 - val_loss: 0.0014 - val_mean_absolute_error: 0.0242
Epoch 10/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0019 - mean_absolute_error: 0.0269 - val_loss: 0.0014 - val_mean_absolute_error: 0.0260
Epoch 11/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0019 - mean_absolute_error: 0.0266 - val_loss: 0.0015 - val_mean_absolute_error: 0.0241
Epoch 12/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0018 - mean_absolute_error: 0.0261 - val_loss: 0.0013 - val_mean_absolute_error: 0.0234
Epoch 13/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0018 - mean_absolute_error: 0.0254 - val_loss: 0.0014 - val_mean_absolute_error: 0.0248
Epoch 14/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0018 - mean_absolute_error: 0.0254 - val_loss: 0.0014 - val_mean_absolute_error: 0.0258
Epoch 15/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0017 - mean_absolute_error: 0.0250 - val_loss: 0.0013 - val_mean_absolute_error: 0.0241
110/110 [==============================] - 0s 2ms/step - loss: 0.0013 - mean_absolute_error: 0.0241
55/55 [==============================] - 0s 2ms/step - loss: 9.2933e-04 - mean_absolute_error: 0.0206
Epoch 1/50
384/384 [==============================] - 8s 18ms/step - loss: 0.0318 - mean_absolute_error: 0.1260 - val_loss: 0.0111 - val_mean_absolute_error: 0.0743
Epoch 2/50
384/384 [==============================] - 8s 20ms/step - loss: 0.0072 - mean_absolute_error: 0.0594 - val_loss: 0.0048 - val_mean_absolute_error: 0.0489
Epoch 3/50
384/384 [==============================] - 8s 20ms/step - loss: 0.0044 - mean_absolute_error: 0.0463 - val_loss: 0.0035 - val_mean_absolute_error: 0.0425
Epoch 4/50
384/384 [==============================] - 7s 19ms/step - loss: 0.0037 - mean_absolute_error: 0.0425 - val_loss: 0.0033 - val_mean_absolute_error: 0.0408
Epoch 5/50
384/384 [==============================] - 7s 18ms/step - loss: 0.0034 - mean_absolute_error: 0.0404 - val_loss: 0.0030 - val_mean_absolute_error: 0.0389
Epoch 6/50
384/384 [==============================] - 7s 18ms/step - loss: 0.0032 - mean_absolute_error: 0.0390 - val_loss: 0.0027 - val_mean_absolute_error: 0.0364
Epoch 7/50
384/384 [==============================] - 7s 18ms/step - loss: 0.0030 - mean_absolute_error: 0.0379 - val_loss: 0.0027 - val_mean_absolute_error: 0.0362
Epoch 8/50
384/384 [==============================] - 7s 18ms/step - loss: 0.0030 - mean_absolute_error: 0.0373 - val_loss: 0.0025 - val_mean_absolute_error: 0.0352
Epoch 9/50
384/384 [==============================] - 8s 21ms/step - loss: 0.0029 - mean_absolute_error: 0.0366 - val_loss: 0.0025 - val_mean_absolute_error: 0.0347
Epoch 10/50
384/384 [==============================] - 9s 22ms/step - loss: 0.0028 - mean_absolute_error: 0.0361 - val_loss: 0.0025 - val_mean_absolute_error: 0.0356
Epoch 11/50
384/384 [==============================] - 10s 26ms/step - loss: 0.0027 - mean_absolute_error: 0.0355 - val_loss: 0.0024 - val_mean_absolute_error: 0.0343
Epoch 12/50
384/384 [==============================] - 8s 19ms/step - loss: 0.0027 - mean_absolute_error: 0.0351 - val_loss: 0.0023 - val_mean_absolute_error: 0.0333
Epoch 13/50
384/384 [==============================] - 7s 17ms/step - loss: 0.0026 - mean_absolute_error: 0.0347 - val_loss: 0.0023 - val_mean_absolute_error: 0.0332
Epoch 14/50
384/384 [==============================] - 7s 17ms/step - loss: 0.0026 - mean_absolute_error: 0.0344 - val_loss: 0.0023 - val_mean_absolute_error: 0.0335
Epoch 15/50
384/384 [==============================] - 7s 18ms/step - loss: 0.0025 - mean_absolute_error: 0.0340 - val_loss: 0.0021 - val_mean_absolute_error: 0.0322
Epoch 16/50
384/384 [==============================] - 7s 17ms/step - loss: 0.0025 - mean_absolute_error: 0.0338 - val_loss: 0.0022 - val_mean_absolute_error: 0.0336
Epoch 17/50
384/384 [==============================] - 6s 17ms/step - loss: 0.0024 - mean_absolute_error: 0.0335 - val_loss: 0.0023 - val_mean_absolute_error: 0.0331
Epoch 18/50
384/384 [==============================] - 7s 18ms/step - loss: 0.0024 - mean_absolute_error: 0.0333 - val_loss: 0.0022 - val_mean_absolute_error: 0.0341
109/109 [==============================] - 1s 6ms/step - loss: 0.0022 - mean_absolute_error: 0.0341
Epoch 1/50
383/383 [==============================] - 8s 18ms/step - loss: 0.0464 - mean_absolute_error: 0.1531 - val_loss: 0.0205 - val_mean_absolute_error: 0.0995
Epoch 2/50
383/383 [==============================] - 7s 17ms/step - loss: 0.0166 - mean_absolute_error: 0.0891 - val_loss: 0.0177 - val_mean_absolute_error: 0.0897
Epoch 3/50
383/383 [==============================] - 7s 17ms/step - loss: 0.0150 - mean_absolute_error: 0.0823 - val_loss: 0.0166 - val_mean_absolute_error: 0.0856
Epoch 4/50
383/383 [==============================] - 7s 17ms/step - loss: 0.0143 - mean_absolute_error: 0.0798 - val_loss: 0.0160 - val_mean_absolute_error: 0.0822
Epoch 5/50
383/383 [==============================] - 7s 18ms/step - loss: 0.0138 - mean_absolute_error: 0.0780 - val_loss: 0.0155 - val_mean_absolute_error: 0.0808
Epoch 6/50
383/383 [==============================] - 7s 17ms/step - loss: 0.0134 - mean_absolute_error: 0.0765 - val_loss: 0.0151 - val_mean_absolute_error: 0.0813
Epoch 7/50
383/383 [==============================] - 7s 17ms/step - loss: 0.0132 - mean_absolute_error: 0.0757 - val_loss: 0.0149 - val_mean_absolute_error: 0.0784
Epoch 8/50
383/383 [==============================] - 7s 18ms/step - loss: 0.0131 - mean_absolute_error: 0.0751 - val_loss: 0.0150 - val_mean_absolute_error: 0.0796
Epoch 9/50
383/383 [==============================] - 6s 17ms/step - loss: 0.0131 - mean_absolute_error: 0.0749 - val_loss: 0.0145 - val_mean_absolute_error: 0.0789
Epoch 10/50
383/383 [==============================] - 7s 18ms/step - loss: 0.0129 - mean_absolute_error: 0.0744 - val_loss: 0.0147 - val_mean_absolute_error: 0.0790
Epoch 11/50
383/383 [==============================] - 8s 20ms/step - loss: 0.0128 - mean_absolute_error: 0.0738 - val_loss: 0.0148 - val_mean_absolute_error: 0.0796
Epoch 12/50
383/383 [==============================] - 8s 22ms/step - loss: 0.0127 - mean_absolute_error: 0.0735 - val_loss: 0.0145 - val_mean_absolute_error: 0.0761
109/109 [==============================] - 1s 9ms/step - loss: 0.0145 - mean_absolute_error: 0.0761
54/54 [==============================] - 0s 7ms/step - loss: 0.0119 - mean_absolute_error: 0.0646
Epoch 1/50
384/384 [==============================] - 16s 37ms/step - loss: 0.0325 - mean_absolute_error: 0.1182 - val_loss: 0.0116 - val_mean_absolute_error: 0.0698
Epoch 2/50
384/384 [==============================] - 12s 30ms/step - loss: 0.0071 - mean_absolute_error: 0.0542 - val_loss: 0.0048 - val_mean_absolute_error: 0.0443
Epoch 3/50
384/384 [==============================] - 8s 20ms/step - loss: 0.0040 - mean_absolute_error: 0.0400 - val_loss: 0.0032 - val_mean_absolute_error: 0.0368
Epoch 4/50
384/384 [==============================] - 8s 20ms/step - loss: 0.0032 - mean_absolute_error: 0.0355 - val_loss: 0.0027 - val_mean_absolute_error: 0.0336
Epoch 5/50
384/384 [==============================] - 7s 19ms/step - loss: 0.0029 - mean_absolute_error: 0.0334 - val_loss: 0.0023 - val_mean_absolute_error: 0.0309
Epoch 6/50
384/384 [==============================] - 8s 19ms/step - loss: 0.0026 - mean_absolute_error: 0.0317 - val_loss: 0.0022 - val_mean_absolute_error: 0.0297
Epoch 7/50
384/384 [==============================] - 7s 19ms/step - loss: 0.0024 - mean_absolute_error: 0.0302 - val_loss: 0.0020 - val_mean_absolute_error: 0.0287
Epoch 8/50
384/384 [==============================] - 8s 20ms/step - loss: 0.0023 - mean_absolute_error: 0.0288 - val_loss: 0.0019 - val_mean_absolute_error: 0.0274
Epoch 9/50
384/384 [==============================] - 7s 19ms/step - loss: 0.0021 - mean_absolute_error: 0.0277 - val_loss: 0.0017 - val_mean_absolute_error: 0.0265
Epoch 10/50
384/384 [==============================] - 8s 21ms/step - loss: 0.0020 - mean_absolute_error: 0.0268 - val_loss: 0.0016 - val_mean_absolute_error: 0.0251
Epoch 11/50
384/384 [==============================] - 12s 30ms/step - loss: 0.0019 - mean_absolute_error: 0.0264 - val_loss: 0.0017 - val_mean_absolute_error: 0.0259
Epoch 12/50
384/384 [==============================] - 10s 25ms/step - loss: 0.0018 - mean_absolute_error: 0.0257 - val_loss: 0.0016 - val_mean_absolute_error: 0.0257
Epoch 13/50
384/384 [==============================] - 9s 23ms/step - loss: 0.0018 - mean_absolute_error: 0.0253 - val_loss: 0.0014 - val_mean_absolute_error: 0.0238
Epoch 14/50
384/384 [==============================] - 9s 24ms/step - loss: 0.0017 - mean_absolute_error: 0.0248 - val_loss: 0.0014 - val_mean_absolute_error: 0.0239
Epoch 15/50
384/384 [==============================] - 11s 27ms/step - loss: 0.0017 - mean_absolute_error: 0.0245 - val_loss: 0.0015 - val_mean_absolute_error: 0.0246
Epoch 16/50
384/384 [==============================] - 9s 23ms/step - loss: 0.0017 - mean_absolute_error: 0.0243 - val_loss: 0.0015 - val_mean_absolute_error: 0.0244
Epoch 17/50
384/384 [==============================] - 11s 29ms/step - loss: 0.0016 - mean_absolute_error: 0.0240 - val_loss: 0.0013 - val_mean_absolute_error: 0.0228
Epoch 18/50
384/384 [==============================] - 12s 31ms/step - loss: 0.0016 - mean_absolute_error: 0.0238 - val_loss: 0.0013 - val_mean_absolute_error: 0.0225
Epoch 19/50
384/384 [==============================] - 11s 30ms/step - loss: 0.0016 - mean_absolute_error: 0.0235 - val_loss: 0.0013 - val_mean_absolute_error: 0.0224
Epoch 20/50
384/384 [==============================] - 13s 33ms/step - loss: 0.0015 - mean_absolute_error: 0.0233 - val_loss: 0.0013 - val_mean_absolute_error: 0.0227
Epoch 21/50
384/384 [==============================] - 9s 24ms/step - loss: 0.0015 - mean_absolute_error: 0.0231 - val_loss: 0.0012 - val_mean_absolute_error: 0.0219
Epoch 22/50
384/384 [==============================] - 7s 19ms/step - loss: 0.0015 - mean_absolute_error: 0.0230 - val_loss: 0.0012 - val_mean_absolute_error: 0.0219
Epoch 23/50
384/384 [==============================] - 8s 20ms/step - loss: 0.0015 - mean_absolute_error: 0.0229 - val_loss: 0.0014 - val_mean_absolute_error: 0.0233
Epoch 24/50
384/384 [==============================] - 8s 20ms/step - loss: 0.0015 - mean_absolute_error: 0.0227 - val_loss: 0.0012 - val_mean_absolute_error: 0.0218
Epoch 25/50
384/384 [==============================] - 8s 20ms/step - loss: 0.0015 - mean_absolute_error: 0.0226 - val_loss: 0.0012 - val_mean_absolute_error: 0.0228
Epoch 26/50
384/384 [==============================] - 8s 20ms/step - loss: 0.0014 - mean_absolute_error: 0.0225 - val_loss: 0.0012 - val_mean_absolute_error: 0.0212
Epoch 27/50
384/384 [==============================] - 8s 20ms/step - loss: 0.0014 - mean_absolute_error: 0.0223 - val_loss: 0.0012 - val_mean_absolute_error: 0.0216
Epoch 28/50
384/384 [==============================] - 8s 20ms/step - loss: 0.0014 - mean_absolute_error: 0.0223 - val_loss: 0.0011 - val_mean_absolute_error: 0.0210
Epoch 29/50
384/384 [==============================] - 8s 20ms/step - loss: 0.0014 - mean_absolute_error: 0.0221 - val_loss: 0.0011 - val_mean_absolute_error: 0.0212
Epoch 30/50
384/384 [==============================] - 7s 19ms/step - loss: 0.0014 - mean_absolute_error: 0.0220 - val_loss: 0.0011 - val_mean_absolute_error: 0.0213
Epoch 31/50
384/384 [==============================] - 9s 24ms/step - loss: 0.0014 - mean_absolute_error: 0.0219 - val_loss: 0.0011 - val_mean_absolute_error: 0.0212
Epoch 32/50
384/384 [==============================] - 9s 23ms/step - loss: 0.0014 - mean_absolute_error: 0.0218 - val_loss: 0.0011 - val_mean_absolute_error: 0.0208
Epoch 33/50
384/384 [==============================] - 8s 21ms/step - loss: 0.0013 - mean_absolute_error: 0.0218 - val_loss: 0.0011 - val_mean_absolute_error: 0.0205
Epoch 34/50
384/384 [==============================] - 8s 20ms/step - loss: 0.0013 - mean_absolute_error: 0.0216 - val_loss: 0.0011 - val_mean_absolute_error: 0.0208
Epoch 35/50
384/384 [==============================] - 9s 23ms/step - loss: 0.0013 - mean_absolute_error: 0.0215 - val_loss: 0.0012 - val_mean_absolute_error: 0.0213
Epoch 36/50
384/384 [==============================] - 8s 20ms/step - loss: 0.0013 - mean_absolute_error: 0.0215 - val_loss: 0.0010 - val_mean_absolute_error: 0.0201
Epoch 37/50
384/384 [==============================] - 8s 20ms/step - loss: 0.0013 - mean_absolute_error: 0.0213 - val_loss: 0.0011 - val_mean_absolute_error: 0.0206
Epoch 38/50
384/384 [==============================] - 8s 20ms/step - loss: 0.0013 - mean_absolute_error: 0.0213 - val_loss: 0.0010 - val_mean_absolute_error: 0.0201
Epoch 39/50
384/384 [==============================] - 8s 20ms/step - loss: 0.0013 - mean_absolute_error: 0.0211 - val_loss: 9.9319e-04 - val_mean_absolute_error: 0.0198
Epoch 40/50
384/384 [==============================] - 8s 21ms/step - loss: 0.0013 - mean_absolute_error: 0.0210 - val_loss: 9.8922e-04 - val_mean_absolute_error: 0.0201
Epoch 41/50
384/384 [==============================] - 8s 20ms/step - loss: 0.0013 - mean_absolute_error: 0.0210 - val_loss: 0.0011 - val_mean_absolute_error: 0.0210
Epoch 42/50
384/384 [==============================] - 8s 20ms/step - loss: 0.0012 - mean_absolute_error: 0.0209 - val_loss: 0.0010 - val_mean_absolute_error: 0.0201
Epoch 43/50
384/384 [==============================] - 8s 20ms/step - loss: 0.0012 - mean_absolute_error: 0.0208 - val_loss: 0.0011 - val_mean_absolute_error: 0.0207
109/109 [==============================] - 1s 9ms/step - loss: 0.0011 - mean_absolute_error: 0.0207
55/55 [==============================] - 1s 8ms/step - loss: 7.9657e-04 - mean_absolute_error: 0.0176

畳み込みニューラルネットワーク(CNN)を調べる¶

CNNを実装する¶

In [10]:

KERNEL_WIDTH = 3
conv_window = DataWindow(input_width=KERNEL_WIDTH, label_width=1, shift=1, df_train=df_train, df_val=df_val, df_test=df_test, label_columns=['traffic_volume'])

cnn_model = Sequential([
    Conv1D(filters=32, kernel_size=(KERNEL_WIDTH,), activation='relu'),
    Dense(units=32, activation='relu'),
    Dense(units=1)
])
history = compile_and_fit(cnn_model, conv_window)
val_performance['CNN'] = cnn_model.evaluate(conv_window.val)
test_performance['CNN'] = cnn_model.evaluate(conv_window.test)

KERNEL_WIDTH = 3 conv_window = DataWindow(input_width=KERNEL_WIDTH, label_width=1, shift=1, df_train=df_train, df_val=df_val, df_test=df_test, label_columns=['traffic_volume']) cnn_model = Sequential([ Conv1D(filters=32, kernel_size=(KERNEL_WIDTH,), activation='relu'), Dense(units=32, activation='relu'), Dense(units=1) ]) history = compile_and_fit(cnn_model, conv_window) val_performance['CNN'] = cnn_model.evaluate(conv_window.val) test_performance['CNN'] = cnn_model.evaluate(conv_window.test)

Epoch 1/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0360 - mean_absolute_error: 0.1166 - val_loss: 0.0072 - val_mean_absolute_error: 0.0614
Epoch 2/50
384/384 [==============================] - 1s 4ms/step - loss: 0.0064 - mean_absolute_error: 0.0587 - val_loss: 0.0050 - val_mean_absolute_error: 0.0527
Epoch 3/50
384/384 [==============================] - 1s 4ms/step - loss: 0.0051 - mean_absolute_error: 0.0525 - val_loss: 0.0042 - val_mean_absolute_error: 0.0494
Epoch 4/50
384/384 [==============================] - 1s 4ms/step - loss: 0.0043 - mean_absolute_error: 0.0482 - val_loss: 0.0041 - val_mean_absolute_error: 0.0500
Epoch 5/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0039 - mean_absolute_error: 0.0459 - val_loss: 0.0033 - val_mean_absolute_error: 0.0439
Epoch 6/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0036 - mean_absolute_error: 0.0438 - val_loss: 0.0033 - val_mean_absolute_error: 0.0439
Epoch 7/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0033 - mean_absolute_error: 0.0421 - val_loss: 0.0036 - val_mean_absolute_error: 0.0461
Epoch 8/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0032 - mean_absolute_error: 0.0409 - val_loss: 0.0027 - val_mean_absolute_error: 0.0393
Epoch 9/50
384/384 [==============================] - 2s 6ms/step - loss: 0.0030 - mean_absolute_error: 0.0396 - val_loss: 0.0025 - val_mean_absolute_error: 0.0367
Epoch 10/50
384/384 [==============================] - 2s 6ms/step - loss: 0.0029 - mean_absolute_error: 0.0386 - val_loss: 0.0026 - val_mean_absolute_error: 0.0375
Epoch 11/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0029 - mean_absolute_error: 0.0385 - val_loss: 0.0023 - val_mean_absolute_error: 0.0354
Epoch 12/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0028 - mean_absolute_error: 0.0384 - val_loss: 0.0022 - val_mean_absolute_error: 0.0349
Epoch 13/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0027 - mean_absolute_error: 0.0374 - val_loss: 0.0023 - val_mean_absolute_error: 0.0358
Epoch 14/50
384/384 [==============================] - 2s 6ms/step - loss: 0.0027 - mean_absolute_error: 0.0372 - val_loss: 0.0021 - val_mean_absolute_error: 0.0340
Epoch 15/50
384/384 [==============================] - 2s 6ms/step - loss: 0.0026 - mean_absolute_error: 0.0364 - val_loss: 0.0021 - val_mean_absolute_error: 0.0336
Epoch 16/50
384/384 [==============================] - 2s 6ms/step - loss: 0.0026 - mean_absolute_error: 0.0359 - val_loss: 0.0023 - val_mean_absolute_error: 0.0357
Epoch 17/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0026 - mean_absolute_error: 0.0360 - val_loss: 0.0021 - val_mean_absolute_error: 0.0336
Epoch 18/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0026 - mean_absolute_error: 0.0358 - val_loss: 0.0021 - val_mean_absolute_error: 0.0345
110/110 [==============================] - 0s 3ms/step - loss: 0.0021 - mean_absolute_error: 0.0345
55/55 [==============================] - 0s 4ms/step - loss: 0.0017 - mean_absolute_error: 0.0308

In [11]:

cnn_lstm_model = Sequential([
    Conv1D(filters=32, kernel_size=(KERNEL_WIDTH,), activation='relu'),
    LSTM(32, return_sequences=True),
    LSTM(32, return_sequences=True),
    Dense(units=1)
])
history = compile_and_fit(cnn_lstm_model, conv_window)
val_performance['CNN + LSTM'] = cnn_lstm_model.evaluate(conv_window.val)
test_performance['CNN + LSTM'] = cnn_lstm_model.evaluate(conv_window.test)

cnn_lstm_model = Sequential([ Conv1D(filters=32, kernel_size=(KERNEL_WIDTH,), activation='relu'), LSTM(32, return_sequences=True), LSTM(32, return_sequences=True), Dense(units=1) ]) history = compile_and_fit(cnn_lstm_model, conv_window) val_performance['CNN + LSTM'] = cnn_lstm_model.evaluate(conv_window.val) test_performance['CNN + LSTM'] = cnn_lstm_model.evaluate(conv_window.test)

Epoch 1/50
384/384 [==============================] - 6s 8ms/step - loss: 0.0306 - mean_absolute_error: 0.1142 - val_loss: 0.0070 - val_mean_absolute_error: 0.0596
Epoch 2/50
384/384 [==============================] - 3s 7ms/step - loss: 0.0063 - mean_absolute_error: 0.0583 - val_loss: 0.0050 - val_mean_absolute_error: 0.0522
Epoch 3/50
384/384 [==============================] - 4s 11ms/step - loss: 0.0049 - mean_absolute_error: 0.0515 - val_loss: 0.0039 - val_mean_absolute_error: 0.0466
Epoch 4/50
384/384 [==============================] - 4s 10ms/step - loss: 0.0041 - mean_absolute_error: 0.0469 - val_loss: 0.0039 - val_mean_absolute_error: 0.0478
Epoch 5/50
384/384 [==============================] - 4s 10ms/step - loss: 0.0038 - mean_absolute_error: 0.0446 - val_loss: 0.0031 - val_mean_absolute_error: 0.0417
Epoch 6/50
384/384 [==============================] - 4s 11ms/step - loss: 0.0035 - mean_absolute_error: 0.0427 - val_loss: 0.0033 - val_mean_absolute_error: 0.0429
Epoch 7/50
384/384 [==============================] - 4s 10ms/step - loss: 0.0033 - mean_absolute_error: 0.0412 - val_loss: 0.0027 - val_mean_absolute_error: 0.0392
Epoch 8/50
384/384 [==============================] - 4s 10ms/step - loss: 0.0031 - mean_absolute_error: 0.0402 - val_loss: 0.0026 - val_mean_absolute_error: 0.0377
Epoch 9/50
384/384 [==============================] - 4s 10ms/step - loss: 0.0031 - mean_absolute_error: 0.0398 - val_loss: 0.0026 - val_mean_absolute_error: 0.0374
Epoch 10/50
384/384 [==============================] - 4s 10ms/step - loss: 0.0029 - mean_absolute_error: 0.0382 - val_loss: 0.0026 - val_mean_absolute_error: 0.0382
Epoch 11/50
384/384 [==============================] - 4s 11ms/step - loss: 0.0028 - mean_absolute_error: 0.0378 - val_loss: 0.0030 - val_mean_absolute_error: 0.0419
110/110 [==============================] - 1s 5ms/step - loss: 0.0030 - mean_absolute_error: 0.0419
55/55 [==============================] - 0s 5ms/step - loss: 0.0027 - mean_absolute_error: 0.0408

In [12]:

KERNEL_WIDTH = 3
LABEL_WIDTH = 24
INPUT_WIDTH = LABEL_WIDTH + KERNEL_WIDTH - 1
multi_conv_window = DataWindow(input_width=INPUT_WIDTH, label_width=LABEL_WIDTH, shift=24, df_train=df_train, df_val=df_val, df_test=df_test, label_columns=['traffic_volume'])

ms_cnn_model = Sequential([
    Conv1D(filters=32, kernel_size=(KERNEL_WIDTH,), activation='relu'),
    Dense(units=32, activation='relu'),
    Dense(units=1, kernel_initializer=tf.initializers.zeros)
])
history = compile_and_fit(ms_cnn_model, multi_conv_window)
ms_val_performance['CNN'] = ms_cnn_model.evaluate(multi_conv_window.val)
ms_test_performance['CNN'] = ms_cnn_model.evaluate(multi_conv_window.test)

KERNEL_WIDTH = 3 LABEL_WIDTH = 24 INPUT_WIDTH = LABEL_WIDTH + KERNEL_WIDTH - 1 multi_conv_window = DataWindow(input_width=INPUT_WIDTH, label_width=LABEL_WIDTH, shift=24, df_train=df_train, df_val=df_val, df_test=df_test, label_columns=['traffic_volume']) ms_cnn_model = Sequential([ Conv1D(filters=32, kernel_size=(KERNEL_WIDTH,), activation='relu'), Dense(units=32, activation='relu'), Dense(units=1, kernel_initializer=tf.initializers.zeros) ]) history = compile_and_fit(ms_cnn_model, multi_conv_window) ms_val_performance['CNN'] = ms_cnn_model.evaluate(multi_conv_window.val) ms_test_performance['CNN'] = ms_cnn_model.evaluate(multi_conv_window.test)

Epoch 1/50
383/383 [==============================] - 3s 5ms/step - loss: 0.0358 - mean_absolute_error: 0.1232 - val_loss: 0.0164 - val_mean_absolute_error: 0.0839
Epoch 2/50
383/383 [==============================] - 2s 4ms/step - loss: 0.0142 - mean_absolute_error: 0.0791 - val_loss: 0.0160 - val_mean_absolute_error: 0.0859
Epoch 3/50
383/383 [==============================] - 2s 6ms/step - loss: 0.0138 - mean_absolute_error: 0.0784 - val_loss: 0.0158 - val_mean_absolute_error: 0.0823
Epoch 4/50
383/383 [==============================] - 2s 6ms/step - loss: 0.0135 - mean_absolute_error: 0.0774 - val_loss: 0.0154 - val_mean_absolute_error: 0.0840
Epoch 5/50
383/383 [==============================] - 2s 6ms/step - loss: 0.0132 - mean_absolute_error: 0.0766 - val_loss: 0.0152 - val_mean_absolute_error: 0.0831
Epoch 6/50
383/383 [==============================] - 2s 6ms/step - loss: 0.0130 - mean_absolute_error: 0.0759 - val_loss: 0.0147 - val_mean_absolute_error: 0.0796
Epoch 7/50
383/383 [==============================] - 2s 6ms/step - loss: 0.0130 - mean_absolute_error: 0.0759 - val_loss: 0.0146 - val_mean_absolute_error: 0.0820
Epoch 8/50
383/383 [==============================] - 3s 7ms/step - loss: 0.0129 - mean_absolute_error: 0.0757 - val_loss: 0.0148 - val_mean_absolute_error: 0.0824
Epoch 9/50
383/383 [==============================] - 3s 9ms/step - loss: 0.0127 - mean_absolute_error: 0.0750 - val_loss: 0.0149 - val_mean_absolute_error: 0.0822
Epoch 10/50
383/383 [==============================] - 3s 7ms/step - loss: 0.0126 - mean_absolute_error: 0.0746 - val_loss: 0.0146 - val_mean_absolute_error: 0.0796
Epoch 11/50
383/383 [==============================] - 3s 7ms/step - loss: 0.0126 - mean_absolute_error: 0.0745 - val_loss: 0.0145 - val_mean_absolute_error: 0.0796
Epoch 12/50
383/383 [==============================] - 3s 7ms/step - loss: 0.0125 - mean_absolute_error: 0.0741 - val_loss: 0.0145 - val_mean_absolute_error: 0.0778
Epoch 13/50
383/383 [==============================] - 3s 7ms/step - loss: 0.0124 - mean_absolute_error: 0.0735 - val_loss: 0.0145 - val_mean_absolute_error: 0.0768
Epoch 14/50
383/383 [==============================] - 3s 6ms/step - loss: 0.0123 - mean_absolute_error: 0.0732 - val_loss: 0.0146 - val_mean_absolute_error: 0.0806
Epoch 15/50
383/383 [==============================] - 2s 6ms/step - loss: 0.0123 - mean_absolute_error: 0.0730 - val_loss: 0.0145 - val_mean_absolute_error: 0.0769
109/109 [==============================] - 1s 5ms/step - loss: 0.0145 - mean_absolute_error: 0.0769
54/54 [==============================] - 0s 4ms/step - loss: 0.0097 - mean_absolute_error: 0.0625

In [13]:

ms_cnn_lstm_model = Sequential([
    Conv1D(filters=32, kernel_size=(KERNEL_WIDTH,), activation='relu'),
    LSTM(32, return_sequences=True),
    Dense(units=1, kernel_initializer=tf.initializers.zeros)
])
history = compile_and_fit(ms_cnn_lstm_model, multi_conv_window)
ms_val_performance['CNN + LSTM'] = ms_cnn_lstm_model.evaluate(multi_conv_window.val)
ms_test_performance['CNN + LSTM'] = ms_cnn_lstm_model.evaluate(multi_conv_window.test)

ms_cnn_lstm_model = Sequential([ Conv1D(filters=32, kernel_size=(KERNEL_WIDTH,), activation='relu'), LSTM(32, return_sequences=True), Dense(units=1, kernel_initializer=tf.initializers.zeros) ]) history = compile_and_fit(ms_cnn_lstm_model, multi_conv_window) ms_val_performance['CNN + LSTM'] = ms_cnn_lstm_model.evaluate(multi_conv_window.val) ms_test_performance['CNN + LSTM'] = ms_cnn_lstm_model.evaluate(multi_conv_window.test)

Epoch 1/50
383/383 [==============================] - 10s 22ms/step - loss: 0.0419 - mean_absolute_error: 0.1452 - val_loss: 0.0174 - val_mean_absolute_error: 0.0904
Epoch 2/50
383/383 [==============================] - 10s 25ms/step - loss: 0.0147 - mean_absolute_error: 0.0823 - val_loss: 0.0161 - val_mean_absolute_error: 0.0859
Epoch 3/50
383/383 [==============================] - 10s 27ms/step - loss: 0.0141 - mean_absolute_error: 0.0801 - val_loss: 0.0158 - val_mean_absolute_error: 0.0866
Epoch 4/50
383/383 [==============================] - 9s 24ms/step - loss: 0.0137 - mean_absolute_error: 0.0787 - val_loss: 0.0155 - val_mean_absolute_error: 0.0855
Epoch 5/50
383/383 [==============================] - 8s 22ms/step - loss: 0.0132 - mean_absolute_error: 0.0770 - val_loss: 0.0149 - val_mean_absolute_error: 0.0810
Epoch 6/50
383/383 [==============================] - 11s 28ms/step - loss: 0.0127 - mean_absolute_error: 0.0752 - val_loss: 0.0151 - val_mean_absolute_error: 0.0766
Epoch 7/50
383/383 [==============================] - 9s 23ms/step - loss: 0.0123 - mean_absolute_error: 0.0733 - val_loss: 0.0136 - val_mean_absolute_error: 0.0763
Epoch 8/50
383/383 [==============================] - 9s 24ms/step - loss: 0.0120 - mean_absolute_error: 0.0722 - val_loss: 0.0134 - val_mean_absolute_error: 0.0750
Epoch 9/50
383/383 [==============================] - 9s 23ms/step - loss: 0.0120 - mean_absolute_error: 0.0721 - val_loss: 0.0132 - val_mean_absolute_error: 0.0769
Epoch 10/50
383/383 [==============================] - 9s 23ms/step - loss: 0.0117 - mean_absolute_error: 0.0711 - val_loss: 0.0131 - val_mean_absolute_error: 0.0743
Epoch 11/50
383/383 [==============================] - 9s 23ms/step - loss: 0.0116 - mean_absolute_error: 0.0705 - val_loss: 0.0129 - val_mean_absolute_error: 0.0742
Epoch 12/50
383/383 [==============================] - 9s 23ms/step - loss: 0.0114 - mean_absolute_error: 0.0699 - val_loss: 0.0138 - val_mean_absolute_error: 0.0805
Epoch 13/50
383/383 [==============================] - 9s 23ms/step - loss: 0.0113 - mean_absolute_error: 0.0695 - val_loss: 0.0127 - val_mean_absolute_error: 0.0736
Epoch 14/50
383/383 [==============================] - 9s 23ms/step - loss: 0.0112 - mean_absolute_error: 0.0690 - val_loss: 0.0129 - val_mean_absolute_error: 0.0742
Epoch 15/50
383/383 [==============================] - 9s 23ms/step - loss: 0.0111 - mean_absolute_error: 0.0687 - val_loss: 0.0125 - val_mean_absolute_error: 0.0730
Epoch 16/50
383/383 [==============================] - 9s 24ms/step - loss: 0.0109 - mean_absolute_error: 0.0678 - val_loss: 0.0126 - val_mean_absolute_error: 0.0743
Epoch 17/50
383/383 [==============================] - 9s 24ms/step - loss: 0.0109 - mean_absolute_error: 0.0678 - val_loss: 0.0131 - val_mean_absolute_error: 0.0719
Epoch 18/50
383/383 [==============================] - 9s 24ms/step - loss: 0.0108 - mean_absolute_error: 0.0674 - val_loss: 0.0131 - val_mean_absolute_error: 0.0774
109/109 [==============================] - 1s 9ms/step - loss: 0.0131 - mean_absolute_error: 0.0774
54/54 [==============================] - 1s 9ms/step - loss: 0.0084 - mean_absolute_error: 0.0599

In [14]:

mo_conv_window = DataWindow(input_width=INPUT_WIDTH, label_width=LABEL_WIDTH, shift=1, df_train=df_train, df_val=df_val, df_test=df_test, label_columns=col_names)

mo_cnn_model = Sequential([
    Conv1D(filters=32, kernel_size=(KERNEL_WIDTH,), activation='relu'),
    Dense(units=32, activation='relu'),
    Dense(units=2)
])
history = compile_and_fit(mo_cnn_model, mo_conv_window)
mo_val_performance['CNN'] = mo_cnn_model.evaluate(mo_conv_window.val)
mo_test_performance['CNN'] = mo_cnn_model.evaluate(mo_conv_window.test)

mo_conv_window = DataWindow(input_width=INPUT_WIDTH, label_width=LABEL_WIDTH, shift=1, df_train=df_train, df_val=df_val, df_test=df_test, label_columns=col_names) mo_cnn_model = Sequential([ Conv1D(filters=32, kernel_size=(KERNEL_WIDTH,), activation='relu'), Dense(units=32, activation='relu'), Dense(units=2) ]) history = compile_and_fit(mo_cnn_model, mo_conv_window) mo_val_performance['CNN'] = mo_cnn_model.evaluate(mo_conv_window.val) mo_test_performance['CNN'] = mo_cnn_model.evaluate(mo_conv_window.test)

Epoch 1/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0101 - mean_absolute_error: 0.0589 - val_loss: 0.0023 - val_mean_absolute_error: 0.0318
Epoch 2/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0020 - mean_absolute_error: 0.0288 - val_loss: 0.0014 - val_mean_absolute_error: 0.0258
Epoch 3/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0016 - mean_absolute_error: 0.0252 - val_loss: 0.0012 - val_mean_absolute_error: 0.0236
Epoch 4/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0015 - mean_absolute_error: 0.0238 - val_loss: 0.0012 - val_mean_absolute_error: 0.0232
Epoch 5/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0014 - mean_absolute_error: 0.0230 - val_loss: 0.0011 - val_mean_absolute_error: 0.0219
Epoch 6/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0013 - mean_absolute_error: 0.0224 - val_loss: 0.0011 - val_mean_absolute_error: 0.0213
Epoch 7/50
384/384 [==============================] - 2s 6ms/step - loss: 0.0013 - mean_absolute_error: 0.0220 - val_loss: 0.0011 - val_mean_absolute_error: 0.0214
Epoch 8/50
384/384 [==============================] - 2s 6ms/step - loss: 0.0013 - mean_absolute_error: 0.0217 - val_loss: 0.0010 - val_mean_absolute_error: 0.0206
Epoch 9/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0013 - mean_absolute_error: 0.0215 - val_loss: 9.9162e-04 - val_mean_absolute_error: 0.0201
Epoch 10/50
384/384 [==============================] - 3s 6ms/step - loss: 0.0012 - mean_absolute_error: 0.0213 - val_loss: 0.0010 - val_mean_absolute_error: 0.0204
Epoch 11/50
384/384 [==============================] - 2s 6ms/step - loss: 0.0012 - mean_absolute_error: 0.0209 - val_loss: 0.0010 - val_mean_absolute_error: 0.0206
Epoch 12/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0012 - mean_absolute_error: 0.0208 - val_loss: 0.0011 - val_mean_absolute_error: 0.0214
109/109 [==============================] - 0s 3ms/step - loss: 0.0011 - mean_absolute_error: 0.0214
55/55 [==============================] - 0s 3ms/step - loss: 8.4113e-04 - mean_absolute_error: 0.0187

In [16]:

mo_cnn_lstm_model = Sequential([
    Conv1D(filters=32, kernel_size=(KERNEL_WIDTH,), activation='relu'),
    LSTM(32, return_sequences=True),
    Dense(units=2)
])
history = compile_and_fit(mo_cnn_lstm_model, mo_conv_window)
mo_val_performance['CNN + LSTM'] = mo_cnn_lstm_model.evaluate(mo_conv_window.val)
mo_test_performance['CNN + LSTM'] = mo_cnn_lstm_model.evaluate(mo_conv_window.test)

mo_cnn_lstm_model = Sequential([ Conv1D(filters=32, kernel_size=(KERNEL_WIDTH,), activation='relu'), LSTM(32, return_sequences=True), Dense(units=2) ]) history = compile_and_fit(mo_cnn_lstm_model, mo_conv_window) mo_val_performance['CNN + LSTM'] = mo_cnn_lstm_model.evaluate(mo_conv_window.val) mo_test_performance['CNN + LSTM'] = mo_cnn_lstm_model.evaluate(mo_conv_window.test)

Epoch 1/50
384/384 [==============================] - 6s 13ms/step - loss: 0.0230 - mean_absolute_error: 0.0913 - val_loss: 0.0041 - val_mean_absolute_error: 0.0415
Epoch 2/50
384/384 [==============================] - 5s 13ms/step - loss: 0.0032 - mean_absolute_error: 0.0363 - val_loss: 0.0024 - val_mean_absolute_error: 0.0324
Epoch 3/50
384/384 [==============================] - 5s 14ms/step - loss: 0.0026 - mean_absolute_error: 0.0322 - val_loss: 0.0020 - val_mean_absolute_error: 0.0292
Epoch 4/50
384/384 [==============================] - 5s 14ms/step - loss: 0.0022 - mean_absolute_error: 0.0298 - val_loss: 0.0017 - val_mean_absolute_error: 0.0279
Epoch 5/50
384/384 [==============================] - 5s 14ms/step - loss: 0.0019 - mean_absolute_error: 0.0278 - val_loss: 0.0017 - val_mean_absolute_error: 0.0279
Epoch 6/50
384/384 [==============================] - 5s 14ms/step - loss: 0.0017 - mean_absolute_error: 0.0263 - val_loss: 0.0015 - val_mean_absolute_error: 0.0274
Epoch 7/50
384/384 [==============================] - 6s 15ms/step - loss: 0.0016 - mean_absolute_error: 0.0255 - val_loss: 0.0013 - val_mean_absolute_error: 0.0238
Epoch 8/50
384/384 [==============================] - 5s 14ms/step - loss: 0.0016 - mean_absolute_error: 0.0246 - val_loss: 0.0013 - val_mean_absolute_error: 0.0237
Epoch 9/50
384/384 [==============================] - 6s 15ms/step - loss: 0.0015 - mean_absolute_error: 0.0240 - val_loss: 0.0013 - val_mean_absolute_error: 0.0238
Epoch 10/50
384/384 [==============================] - 5s 14ms/step - loss: 0.0015 - mean_absolute_error: 0.0236 - val_loss: 0.0012 - val_mean_absolute_error: 0.0238
Epoch 11/50
384/384 [==============================] - 6s 14ms/step - loss: 0.0015 - mean_absolute_error: 0.0232 - val_loss: 0.0013 - val_mean_absolute_error: 0.0246
Epoch 12/50
384/384 [==============================] - 6s 15ms/step - loss: 0.0014 - mean_absolute_error: 0.0229 - val_loss: 0.0012 - val_mean_absolute_error: 0.0224
Epoch 13/50
384/384 [==============================] - 6s 15ms/step - loss: 0.0014 - mean_absolute_error: 0.0227 - val_loss: 0.0011 - val_mean_absolute_error: 0.0221
Epoch 14/50
384/384 [==============================] - 6s 15ms/step - loss: 0.0014 - mean_absolute_error: 0.0225 - val_loss: 0.0011 - val_mean_absolute_error: 0.0211
Epoch 15/50
384/384 [==============================] - 6s 15ms/step - loss: 0.0014 - mean_absolute_error: 0.0224 - val_loss: 0.0011 - val_mean_absolute_error: 0.0222
Epoch 16/50
384/384 [==============================] - 6s 15ms/step - loss: 0.0013 - mean_absolute_error: 0.0221 - val_loss: 0.0011 - val_mean_absolute_error: 0.0224
Epoch 17/50
384/384 [==============================] - 6s 16ms/step - loss: 0.0013 - mean_absolute_error: 0.0219 - val_loss: 0.0011 - val_mean_absolute_error: 0.0217
109/109 [==============================] - 1s 6ms/step - loss: 0.0011 - mean_absolute_error: 0.0217
55/55 [==============================] - 0s 6ms/step - loss: 8.2616e-04 - mean_absolute_error: 0.0185

In [23]:

# single stepの結果比較
plt.title('Single Step')
plt.xlabel('Models')
plt.ylabel('MAE')
plt.bar(val_performance.keys(), [v[1] for v in val_performance.values()], width=-0.25, align='edge', label='Validation')
plt.bar(test_performance.keys(), [v[1] for v in test_performance.values()], width=0.25, align='edge', label='Test', hatch='/')
plt.xticks(rotation=45, ha='right')
plt.legend()
plt.tight_layout()

# single stepの結果比較 plt.title('Single Step') plt.xlabel('Models') plt.ylabel('MAE') plt.bar(val_performance.keys(), [v[1] for v in val_performance.values()], width=-0.25, align='edge', label='Validation') plt.bar(test_performance.keys(), [v[1] for v in test_performance.values()], width=0.25, align='edge', label='Test', hatch='/') plt.xticks(rotation=45, ha='right') plt.legend() plt.tight_layout()

In [22]:

# multi stepの結果比較
plt.title('Multi Step')
plt.xlabel('Models')
plt.ylabel('MAE')
plt.bar(ms_val_performance.keys(), [v[1] for v in ms_val_performance.values()], width=-0.25, align='edge', label='Validation')
plt.bar(ms_test_performance.keys(), [v[1] for v in ms_test_performance.values()], width=0.25, align='edge', label='Test', hatch='/')
plt.legend()
plt.xticks(rotation=45, ha='right')
plt.tight_layout()

# multi stepの結果比較 plt.title('Multi Step') plt.xlabel('Models') plt.ylabel('MAE') plt.bar(ms_val_performance.keys(), [v[1] for v in ms_val_performance.values()], width=-0.25, align='edge', label='Validation') plt.bar(ms_test_performance.keys(), [v[1] for v in ms_test_performance.values()], width=0.25, align='edge', label='Test', hatch='/') plt.legend() plt.xticks(rotation=45, ha='right') plt.tight_layout()

In [24]:

# multi outputの結果比較
plt.title('Multi Output')
plt.xlabel('Models')
plt.ylabel('MAE')
plt.bar(mo_val_performance.keys(), [v[1] for v in mo_val_performance.values()], width=-0.25, align='edge', label='Validation')
plt.bar(mo_test_performance.keys(), [v[1] for v in mo_test_performance.values()], width=0.25, align='edge', label='Test', hatch='/')
plt.xticks(rotation=45, ha='right')
plt.legend()
plt.tight_layout()

# multi outputの結果比較 plt.title('Multi Output') plt.xlabel('Models') plt.ylabel('MAE') plt.bar(mo_val_performance.keys(), [v[1] for v in mo_val_performance.values()], width=-0.25, align='edge', label='Validation') plt.bar(mo_test_performance.keys(), [v[1] for v in mo_test_performance.values()], width=0.25, align='edge', label='Test', hatch='/') plt.xticks(rotation=45, ha='right') plt.legend() plt.tight_layout()

In [ ]: