第14章 ディープラーニングの手ほどき

In [10]:

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

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

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 [6]:

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 [7]:

# 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:]

# 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:]

In [41]:

# 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 = {}

val_performance['Baseline - Last'] = baseline_last.evaluate(single_step_window.val)
test_performance['Baseline - Last'] = baseline_last.evaluate(single_step_window.test, verbose=0)

# 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 = {}

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)

# 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 = {}

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)

# 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 = {} val_performance['Baseline - Last'] = baseline_last.evaluate(single_step_window.val) test_performance['Baseline - Last'] = baseline_last.evaluate(single_step_window.test, verbose=0) # 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 = {} 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) # 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 = {} 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)

110/110 [==============================] - 1s 3ms/step - loss: 0.0133 - mean_absolute_error: 0.0831
109/109 [==============================] - 0s 3ms/step - loss: 0.1875 - mean_absolute_error: 0.3522
54/54 [==============================] - 0s 3ms/step - loss: 0.1814 - mean_absolute_error: 0.3473
109/109 [==============================] - 0s 3ms/step - loss: 0.2065 - mean_absolute_error: 0.3473
54/54 [==============================] - 0s 3ms/step - loss: 0.2018 - mean_absolute_error: 0.3413
110/110 [==============================] - 0s 3ms/step - loss: 0.0069 - mean_absolute_error: 0.0482
110/110 [==============================] - 0s 3ms/step - loss: 0.0069 - mean_absolute_error: 0.0482

線形モデルを実装する¶

In [42]:

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

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

シングルステップの線形モデルを実装する¶

In [43]:

linear = Sequential([Dense(units=1)])
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)

linear = Sequential([Dense(units=1)]) 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)

Epoch 1/50
384/384 [==============================] - 2s 4ms/step - loss: 0.3284 - mean_absolute_error: 0.4560 - val_loss: 0.0977 - val_mean_absolute_error: 0.2534
Epoch 2/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0829 - mean_absolute_error: 0.2325 - val_loss: 0.0570 - val_mean_absolute_error: 0.1893
Epoch 3/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0478 - mean_absolute_error: 0.1743 - val_loss: 0.0325 - val_mean_absolute_error: 0.1410
Epoch 4/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0279 - mean_absolute_error: 0.1319 - val_loss: 0.0200 - val_mean_absolute_error: 0.1100
Epoch 5/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0184 - mean_absolute_error: 0.1060 - val_loss: 0.0147 - val_mean_absolute_error: 0.0930
Epoch 6/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0146 - mean_absolute_error: 0.0922 - val_loss: 0.0128 - val_mean_absolute_error: 0.0844
Epoch 7/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0131 - mean_absolute_error: 0.0851 - val_loss: 0.0119 - val_mean_absolute_error: 0.0798
Epoch 8/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0124 - mean_absolute_error: 0.0813 - val_loss: 0.0114 - val_mean_absolute_error: 0.0772
Epoch 9/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0118 - mean_absolute_error: 0.0786 - val_loss: 0.0109 - val_mean_absolute_error: 0.0749
Epoch 10/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0113 - mean_absolute_error: 0.0762 - val_loss: 0.0105 - val_mean_absolute_error: 0.0728
Epoch 11/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0108 - mean_absolute_error: 0.0740 - val_loss: 0.0101 - val_mean_absolute_error: 0.0708
Epoch 12/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0104 - mean_absolute_error: 0.0721 - val_loss: 0.0098 - val_mean_absolute_error: 0.0692
Epoch 13/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0101 - mean_absolute_error: 0.0706 - val_loss: 0.0095 - val_mean_absolute_error: 0.0675
Epoch 14/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0098 - mean_absolute_error: 0.0693 - val_loss: 0.0093 - val_mean_absolute_error: 0.0670
Epoch 15/50
384/384 [==============================] - 3s 6ms/step - loss: 0.0096 - mean_absolute_error: 0.0685 - val_loss: 0.0092 - val_mean_absolute_error: 0.0656
Epoch 16/50
384/384 [==============================] - 2s 6ms/step - loss: 0.0095 - mean_absolute_error: 0.0680 - val_loss: 0.0091 - val_mean_absolute_error: 0.0657
Epoch 17/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0094 - mean_absolute_error: 0.0676 - val_loss: 0.0091 - val_mean_absolute_error: 0.0656
Epoch 18/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0093 - mean_absolute_error: 0.0676 - val_loss: 0.0091 - val_mean_absolute_error: 0.0649
Epoch 19/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0093 - mean_absolute_error: 0.0675 - val_loss: 0.0091 - val_mean_absolute_error: 0.0645
Epoch 20/50
384/384 [==============================] - 3s 7ms/step - loss: 0.0093 - mean_absolute_error: 0.0675 - val_loss: 0.0091 - val_mean_absolute_error: 0.0652
Epoch 21/50
384/384 [==============================] - 3s 6ms/step - loss: 0.0093 - mean_absolute_error: 0.0676 - val_loss: 0.0091 - val_mean_absolute_error: 0.0652
Epoch 22/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0093 - mean_absolute_error: 0.0675 - val_loss: 0.0091 - val_mean_absolute_error: 0.0648
Epoch 23/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0093 - mean_absolute_error: 0.0676 - val_loss: 0.0091 - val_mean_absolute_error: 0.0645
110/110 [==============================] - 0s 4ms/step - loss: 0.0091 - mean_absolute_error: 0.0645

In [44]:

wide_window.plot('traffic_volume', linear)

wide_window.plot('traffic_volume', linear)

マルチステップの線形モデルを実装する¶

In [45]:

ms_linear = Sequential([Dense(units=1, kernel_initializer=tf.initializers.zeros)])
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)

ms_linear = Sequential([Dense(units=1, kernel_initializer=tf.initializers.zeros)]) 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)

Epoch 1/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0897 - mean_absolute_error: 0.2463 - val_loss: 0.0456 - val_mean_absolute_error: 0.1819
Epoch 2/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0297 - mean_absolute_error: 0.1366 - val_loss: 0.0261 - val_mean_absolute_error: 0.1251
Epoch 3/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0200 - mean_absolute_error: 0.1044 - val_loss: 0.0213 - val_mean_absolute_error: 0.1074
Epoch 4/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0177 - mean_absolute_error: 0.0955 - val_loss: 0.0196 - val_mean_absolute_error: 0.1000
Epoch 5/50
383/383 [==============================] - 2s 6ms/step - loss: 0.0168 - mean_absolute_error: 0.0914 - val_loss: 0.0189 - val_mean_absolute_error: 0.0956
Epoch 6/50
383/383 [==============================] - 2s 6ms/step - loss: 0.0164 - mean_absolute_error: 0.0887 - val_loss: 0.0185 - val_mean_absolute_error: 0.0924
Epoch 7/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0162 - mean_absolute_error: 0.0869 - val_loss: 0.0183 - val_mean_absolute_error: 0.0905
Epoch 8/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0162 - mean_absolute_error: 0.0857 - val_loss: 0.0182 - val_mean_absolute_error: 0.0894
Epoch 9/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0161 - mean_absolute_error: 0.0850 - val_loss: 0.0182 - val_mean_absolute_error: 0.0887
Epoch 10/50
383/383 [==============================] - 2s 6ms/step - loss: 0.0161 - mean_absolute_error: 0.0846 - val_loss: 0.0182 - val_mean_absolute_error: 0.0883
Epoch 11/50
383/383 [==============================] - 2s 4ms/step - loss: 0.0161 - mean_absolute_error: 0.0843 - val_loss: 0.0182 - val_mean_absolute_error: 0.0887
Epoch 12/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0161 - mean_absolute_error: 0.0841 - val_loss: 0.0182 - val_mean_absolute_error: 0.0883
Epoch 13/50
383/383 [==============================] - 2s 5ms/step - loss: 0.0161 - mean_absolute_error: 0.0840 - val_loss: 0.0182 - val_mean_absolute_error: 0.0881
Epoch 14/50
383/383 [==============================] - 2s 4ms/step - loss: 0.0161 - mean_absolute_error: 0.0839 - val_loss: 0.0182 - val_mean_absolute_error: 0.0880
Epoch 15/50
383/383 [==============================] - 2s 4ms/step - loss: 0.0161 - mean_absolute_error: 0.0839 - val_loss: 0.0182 - val_mean_absolute_error: 0.0881
109/109 [==============================] - 0s 3ms/step - loss: 0.0182 - mean_absolute_error: 0.0881
54/54 [==============================] - 0s 2ms/step - loss: 0.0142 - mean_absolute_error: 0.0759

In [46]:

multi_window.plot('traffic_volume', ms_linear)

multi_window.plot('traffic_volume', ms_linear)

多出力の線形モデルを実装する¶

In [47]:

mo_linear = Sequential([Dense(units=2)])
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)

mo_linear = Sequential([Dense(units=2)]) 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)

Epoch 1/50
384/384 [==============================] - 2s 5ms/step - loss: 0.2844 - mean_absolute_error: 0.4397 - val_loss: 0.1930 - val_mean_absolute_error: 0.3678
Epoch 2/50
384/384 [==============================] - 2s 4ms/step - loss: 0.1347 - mean_absolute_error: 0.3038 - val_loss: 0.1113 - val_mean_absolute_error: 0.2788
Epoch 3/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0795 - mean_absolute_error: 0.2296 - val_loss: 0.0696 - val_mean_absolute_error: 0.2200
Epoch 4/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0513 - mean_absolute_error: 0.1817 - val_loss: 0.0473 - val_mean_absolute_error: 0.1811
Epoch 5/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0358 - mean_absolute_error: 0.1510 - val_loss: 0.0335 - val_mean_absolute_error: 0.1521
Epoch 6/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0259 - mean_absolute_error: 0.1284 - val_loss: 0.0242 - val_mean_absolute_error: 0.1290
Epoch 7/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0188 - mean_absolute_error: 0.1092 - val_loss: 0.0173 - val_mean_absolute_error: 0.1085
Epoch 8/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0138 - mean_absolute_error: 0.0924 - val_loss: 0.0126 - val_mean_absolute_error: 0.0910
Epoch 9/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0103 - mean_absolute_error: 0.0778 - val_loss: 0.0093 - val_mean_absolute_error: 0.0756
Epoch 10/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0080 - mean_absolute_error: 0.0660 - val_loss: 0.0074 - val_mean_absolute_error: 0.0634
Epoch 11/50
384/384 [==============================] - 1s 4ms/step - loss: 0.0067 - mean_absolute_error: 0.0568 - val_loss: 0.0061 - val_mean_absolute_error: 0.0532
Epoch 12/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0059 - mean_absolute_error: 0.0501 - val_loss: 0.0055 - val_mean_absolute_error: 0.0464
Epoch 13/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0055 - mean_absolute_error: 0.0457 - val_loss: 0.0052 - val_mean_absolute_error: 0.0429
Epoch 14/50
384/384 [==============================] - 1s 4ms/step - loss: 0.0053 - mean_absolute_error: 0.0431 - val_loss: 0.0050 - val_mean_absolute_error: 0.0403
Epoch 15/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0052 - mean_absolute_error: 0.0416 - val_loss: 0.0048 - val_mean_absolute_error: 0.0395
Epoch 16/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0051 - mean_absolute_error: 0.0407 - val_loss: 0.0048 - val_mean_absolute_error: 0.0383
Epoch 17/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0050 - mean_absolute_error: 0.0402 - val_loss: 0.0047 - val_mean_absolute_error: 0.0384
Epoch 18/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0050 - mean_absolute_error: 0.0399 - val_loss: 0.0047 - val_mean_absolute_error: 0.0382
Epoch 19/50
384/384 [==============================] - 1s 4ms/step - loss: 0.0050 - mean_absolute_error: 0.0397 - val_loss: 0.0047 - val_mean_absolute_error: 0.0383
Epoch 20/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0049 - mean_absolute_error: 0.0396 - val_loss: 0.0047 - val_mean_absolute_error: 0.0382
Epoch 21/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0049 - mean_absolute_error: 0.0396 - val_loss: 0.0047 - val_mean_absolute_error: 0.0391
Epoch 22/50
384/384 [==============================] - 1s 4ms/step - loss: 0.0049 - mean_absolute_error: 0.0396 - val_loss: 0.0046 - val_mean_absolute_error: 0.0380
Epoch 23/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0049 - mean_absolute_error: 0.0395 - val_loss: 0.0046 - val_mean_absolute_error: 0.0381
Epoch 24/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0049 - mean_absolute_error: 0.0395 - val_loss: 0.0046 - val_mean_absolute_error: 0.0386
Epoch 25/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0049 - mean_absolute_error: 0.0396 - val_loss: 0.0046 - val_mean_absolute_error: 0.0381
Epoch 26/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0049 - mean_absolute_error: 0.0395 - val_loss: 0.0046 - val_mean_absolute_error: 0.0380
Epoch 27/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0049 - mean_absolute_error: 0.0395 - val_loss: 0.0046 - val_mean_absolute_error: 0.0377
Epoch 28/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0049 - mean_absolute_error: 0.0395 - val_loss: 0.0046 - val_mean_absolute_error: 0.0378
Epoch 29/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0049 - mean_absolute_error: 0.0395 - val_loss: 0.0046 - val_mean_absolute_error: 0.0381
110/110 [==============================] - 0s 3ms/step - loss: 0.0046 - mean_absolute_error: 0.0381
55/55 [==============================] - 0s 3ms/step - loss: 0.0043 - mean_absolute_error: 0.0361

In [48]:

mo_wide_window.plot('traffic_volume', mo_linear)

mo_wide_window.plot('traffic_volume', mo_linear)

In [49]:

pd.DataFrame(history.history)[['loss', 'val_loss']].plot()

pd.DataFrame(history.history)[['loss', 'val_loss']].plot()

Out[49]:

<Axes: >

ディープニューラルネットワーク(DNN)を実装する¶

In [50]:

dense = Sequential([
    Dense(units=64, activation='relu'),
    Dense(units=64, activation='relu'),
    Dense(units=1)
])
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)

dense = Sequential([ Dense(units=64, activation='relu'), Dense(units=64, activation='relu'), Dense(units=1) ]) 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)

Epoch 1/50
384/384 [==============================] - 3s 6ms/step - loss: 0.0116 - mean_absolute_error: 0.0700 - val_loss: 0.0044 - val_mean_absolute_error: 0.0521
Epoch 2/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0040 - mean_absolute_error: 0.0467 - val_loss: 0.0033 - val_mean_absolute_error: 0.0435
Epoch 3/50
384/384 [==============================] - 3s 6ms/step - loss: 0.0035 - mean_absolute_error: 0.0431 - val_loss: 0.0029 - val_mean_absolute_error: 0.0406
Epoch 4/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0033 - mean_absolute_error: 0.0415 - val_loss: 0.0027 - val_mean_absolute_error: 0.0389
Epoch 5/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0031 - mean_absolute_error: 0.0400 - val_loss: 0.0027 - val_mean_absolute_error: 0.0382
Epoch 6/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0030 - mean_absolute_error: 0.0396 - val_loss: 0.0027 - val_mean_absolute_error: 0.0385
Epoch 7/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0028 - mean_absolute_error: 0.0381 - val_loss: 0.0023 - val_mean_absolute_error: 0.0348
Epoch 8/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0028 - mean_absolute_error: 0.0381 - val_loss: 0.0022 - val_mean_absolute_error: 0.0341
Epoch 9/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0027 - mean_absolute_error: 0.0368 - val_loss: 0.0023 - val_mean_absolute_error: 0.0360
Epoch 10/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0027 - mean_absolute_error: 0.0373 - val_loss: 0.0026 - val_mean_absolute_error: 0.0384
Epoch 11/50
384/384 [==============================] - 2s 4ms/step - loss: 0.0026 - mean_absolute_error: 0.0365 - val_loss: 0.0021 - val_mean_absolute_error: 0.0332
Epoch 12/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0025 - mean_absolute_error: 0.0355 - val_loss: 0.0023 - val_mean_absolute_error: 0.0362
Epoch 13/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0025 - mean_absolute_error: 0.0356 - val_loss: 0.0022 - val_mean_absolute_error: 0.0335
Epoch 14/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0025 - mean_absolute_error: 0.0350 - val_loss: 0.0020 - val_mean_absolute_error: 0.0319
Epoch 15/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0025 - mean_absolute_error: 0.0349 - val_loss: 0.0022 - val_mean_absolute_error: 0.0339
Epoch 16/50
384/384 [==============================] - 2s 6ms/step - loss: 0.0025 - mean_absolute_error: 0.0352 - val_loss: 0.0025 - val_mean_absolute_error: 0.0383
Epoch 17/50
384/384 [==============================] - 2s 6ms/step - loss: 0.0024 - mean_absolute_error: 0.0349 - val_loss: 0.0021 - val_mean_absolute_error: 0.0337
110/110 [==============================] - 1s 4ms/step - loss: 0.0021 - mean_absolute_error: 0.0337

In [51]:

wide_window.plot('traffic_volume', dense)

wide_window.plot('traffic_volume', dense)

In [52]:

pd.DataFrame(history.history)[['loss', 'val_loss']].plot()

pd.DataFrame(history.history)[['loss', 'val_loss']].plot()

Out[52]:

<Axes: >

In [53]:

ms_dense = Sequential([
    Dense(units=64, activation='relu'),
    Dense(units=64, activation='relu'),
    Dense(units=1, kernel_initializer=tf.initializers.zeros)
])
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)

ms_dense = Sequential([ Dense(units=64, activation='relu'), Dense(units=64, activation='relu'), Dense(units=1, kernel_initializer=tf.initializers.zeros) ]) 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)

Epoch 1/50

383/383 [==============================] - 3s 7ms/step - loss: 0.0307 - mean_absolute_error: 0.1132 - val_loss: 0.0169 - val_mean_absolute_error: 0.0860
Epoch 2/50
383/383 [==============================] - 3s 6ms/step - loss: 0.0141 - mean_absolute_error: 0.0790 - val_loss: 0.0161 - val_mean_absolute_error: 0.0853
Epoch 3/50
383/383 [==============================] - 2s 6ms/step - loss: 0.0136 - mean_absolute_error: 0.0779 - val_loss: 0.0153 - val_mean_absolute_error: 0.0842
Epoch 4/50
383/383 [==============================] - 2s 6ms/step - loss: 0.0133 - mean_absolute_error: 0.0772 - val_loss: 0.0152 - val_mean_absolute_error: 0.0851
Epoch 5/50
383/383 [==============================] - 2s 6ms/step - loss: 0.0132 - mean_absolute_error: 0.0766 - val_loss: 0.0156 - val_mean_absolute_error: 0.0813
Epoch 6/50
383/383 [==============================] - 3s 7ms/step - loss: 0.0130 - mean_absolute_error: 0.0758 - val_loss: 0.0147 - val_mean_absolute_error: 0.0806
Epoch 7/50
383/383 [==============================] - 3s 7ms/step - loss: 0.0128 - mean_absolute_error: 0.0752 - val_loss: 0.0145 - val_mean_absolute_error: 0.0804
Epoch 8/50
383/383 [==============================] - 2s 6ms/step - loss: 0.0127 - mean_absolute_error: 0.0749 - val_loss: 0.0145 - val_mean_absolute_error: 0.0799
Epoch 9/50
383/383 [==============================] - 2s 6ms/step - loss: 0.0128 - mean_absolute_error: 0.0751 - val_loss: 0.0145 - val_mean_absolute_error: 0.0788
Epoch 10/50
383/383 [==============================] - 2s 6ms/step - loss: 0.0127 - mean_absolute_error: 0.0747 - val_loss: 0.0143 - val_mean_absolute_error: 0.0793
Epoch 11/50
383/383 [==============================] - 2s 6ms/step - loss: 0.0126 - mean_absolute_error: 0.0743 - val_loss: 0.0143 - val_mean_absolute_error: 0.0794
Epoch 12/50
383/383 [==============================] - 3s 7ms/step - loss: 0.0126 - mean_absolute_error: 0.0744 - val_loss: 0.0143 - val_mean_absolute_error: 0.0785
Epoch 13/50
383/383 [==============================] - 3s 7ms/step - loss: 0.0125 - mean_absolute_error: 0.0741 - val_loss: 0.0144 - val_mean_absolute_error: 0.0775
Epoch 14/50
383/383 [==============================] - 3s 7ms/step - loss: 0.0125 - mean_absolute_error: 0.0740 - val_loss: 0.0142 - val_mean_absolute_error: 0.0782
Epoch 15/50
383/383 [==============================] - 3s 6ms/step - loss: 0.0125 - mean_absolute_error: 0.0739 - val_loss: 0.0142 - val_mean_absolute_error: 0.0781
Epoch 16/50
383/383 [==============================] - 2s 6ms/step - loss: 0.0125 - mean_absolute_error: 0.0742 - val_loss: 0.0142 - val_mean_absolute_error: 0.0775
Epoch 17/50
383/383 [==============================] - 3s 6ms/step - loss: 0.0124 - mean_absolute_error: 0.0738 - val_loss: 0.0142 - val_mean_absolute_error: 0.0801
Epoch 18/50
383/383 [==============================] - 3s 6ms/step - loss: 0.0124 - mean_absolute_error: 0.0736 - val_loss: 0.0142 - val_mean_absolute_error: 0.0793
Epoch 19/50
383/383 [==============================] - 3s 7ms/step - loss: 0.0123 - mean_absolute_error: 0.0734 - val_loss: 0.0143 - val_mean_absolute_error: 0.0815
Epoch 20/50
383/383 [==============================] - 2s 6ms/step - loss: 0.0123 - mean_absolute_error: 0.0732 - val_loss: 0.0145 - val_mean_absolute_error: 0.0823
Epoch 21/50
383/383 [==============================] - 3s 7ms/step - loss: 0.0122 - mean_absolute_error: 0.0731 - val_loss: 0.0140 - val_mean_absolute_error: 0.0766
Epoch 22/50
383/383 [==============================] - 2s 6ms/step - loss: 0.0122 - mean_absolute_error: 0.0732 - val_loss: 0.0139 - val_mean_absolute_error: 0.0777
Epoch 23/50
383/383 [==============================] - 3s 7ms/step - loss: 0.0121 - mean_absolute_error: 0.0727 - val_loss: 0.0141 - val_mean_absolute_error: 0.0767
Epoch 24/50
383/383 [==============================] - 2s 6ms/step - loss: 0.0121 - mean_absolute_error: 0.0726 - val_loss: 0.0140 - val_mean_absolute_error: 0.0756
Epoch 25/50
383/383 [==============================] - 3s 6ms/step - loss: 0.0121 - mean_absolute_error: 0.0725 - val_loss: 0.0142 - val_mean_absolute_error: 0.0751
109/109 [==============================] - 1s 4ms/step - loss: 0.0142 - mean_absolute_error: 0.0751
54/54 [==============================] - 0s 4ms/step - loss: 0.0096 - mean_absolute_error: 0.0618

In [54]:

multi_window.plot('traffic_volume', ms_dense)

multi_window.plot('traffic_volume', ms_dense)

In [55]:

pd.DataFrame(history.history)[['loss', 'val_loss']].plot()

pd.DataFrame(history.history)[['loss', 'val_loss']].plot()

Out[55]:

<Axes: >

In [56]:

mo_dense = Sequential([
    Dense(units=64, activation='relu'),
    Dense(units=64, activation='relu'),
    Dense(units=2)
])
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)

mo_dense = Sequential([ Dense(units=64, activation='relu'), Dense(units=64, activation='relu'), Dense(units=2) ]) 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)

Epoch 1/50
384/384 [==============================] - 3s 5ms/step - loss: 0.0127 - mean_absolute_error: 0.0642 - val_loss: 0.0031 - val_mean_absolute_error: 0.0349
Epoch 2/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0028 - mean_absolute_error: 0.0324 - val_loss: 0.0023 - val_mean_absolute_error: 0.0302
Epoch 3/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0024 - mean_absolute_error: 0.0299 - val_loss: 0.0022 - val_mean_absolute_error: 0.0304
Epoch 4/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0022 - mean_absolute_error: 0.0285 - val_loss: 0.0017 - val_mean_absolute_error: 0.0262
Epoch 5/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0020 - mean_absolute_error: 0.0277 - val_loss: 0.0016 - val_mean_absolute_error: 0.0255
Epoch 6/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0020 - mean_absolute_error: 0.0277 - val_loss: 0.0016 - val_mean_absolute_error: 0.0258
Epoch 7/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0019 - mean_absolute_error: 0.0270 - val_loss: 0.0015 - val_mean_absolute_error: 0.0256
Epoch 8/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0019 - mean_absolute_error: 0.0264 - val_loss: 0.0015 - val_mean_absolute_error: 0.0242
Epoch 9/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0018 - mean_absolute_error: 0.0260 - val_loss: 0.0015 - val_mean_absolute_error: 0.0243
Epoch 10/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0018 - mean_absolute_error: 0.0257 - val_loss: 0.0018 - val_mean_absolute_error: 0.0288
Epoch 11/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0017 - mean_absolute_error: 0.0255 - val_loss: 0.0014 - val_mean_absolute_error: 0.0239
Epoch 12/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0017 - mean_absolute_error: 0.0249 - val_loss: 0.0014 - val_mean_absolute_error: 0.0240
Epoch 13/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0017 - mean_absolute_error: 0.0250 - val_loss: 0.0013 - val_mean_absolute_error: 0.0233
Epoch 14/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0016 - mean_absolute_error: 0.0244 - val_loss: 0.0013 - val_mean_absolute_error: 0.0230
Epoch 15/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0016 - mean_absolute_error: 0.0241 - val_loss: 0.0014 - val_mean_absolute_error: 0.0274
Epoch 16/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0016 - mean_absolute_error: 0.0242 - val_loss: 0.0014 - val_mean_absolute_error: 0.0238
Epoch 17/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0016 - mean_absolute_error: 0.0243 - val_loss: 0.0012 - val_mean_absolute_error: 0.0227
Epoch 18/50
384/384 [==============================] - 2s 6ms/step - loss: 0.0016 - mean_absolute_error: 0.0240 - val_loss: 0.0012 - val_mean_absolute_error: 0.0224
Epoch 19/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0016 - mean_absolute_error: 0.0241 - val_loss: 0.0013 - val_mean_absolute_error: 0.0236
Epoch 20/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0015 - mean_absolute_error: 0.0235 - val_loss: 0.0015 - val_mean_absolute_error: 0.0271
Epoch 21/50
384/384 [==============================] - 2s 5ms/step - loss: 0.0015 - mean_absolute_error: 0.0237 - val_loss: 0.0013 - val_mean_absolute_error: 0.0239
110/110 [==============================] - 0s 3ms/step - loss: 0.0013 - mean_absolute_error: 0.0239
55/55 [==============================] - 0s 3ms/step - loss: 9.0741e-04 - mean_absolute_error: 0.0203

In [57]:

mo_wide_window.plot('traffic_volume', mo_dense)

mo_wide_window.plot('traffic_volume', mo_dense)

In [58]:

pd.DataFrame(history.history)[['loss', 'val_loss']].plot()

pd.DataFrame(history.history)[['loss', 'val_loss']].plot()

Out[58]:

<Axes: >

In [72]:

# 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.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.legend() plt.tight_layout()

In [73]:

# 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.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.tight_layout()

In [74]:

# 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.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.legend() plt.tight_layout()

In [ ]: