前提・実現したいこと
①時系列データから②いくつかのトレーニング期間とラベル期間を指定して学習③アンサンブルする
このようなモデルを作成したく、コードを書いています。
参考はこちらを参照させていただいています。
https://www.tensorflow.org/tutorials/structured_data/time_series
loss計算や評価計算がnanになってしまうため、学習させるコードを確認したく
質問させていただきました。
tensorflow version : 2.4.1
keras version : 2.4.3
発生している問題・エラーメッセージ
参考先で与えられているデータ
zip_path = tf.keras.utils.get_file(
origin='https://storage.googleapis.com/tensorflow/tf-keras-
# 文字数の関係で一部省略
では発生しなかったのですが
私が取得したデータ
import pandas_datareader.data as pdr
df = pdr.DataReader('aapl', 'yahoo', '2018/01/01')
# 一部省略
では
モデル内で採用する損失関数や評価指標
tensorflow.keras.Sequential.Dense(losses=MeanSquaredError(),metrics=MeanAbsoluteError())
が欠損値nanになってしまいます。
該当のソースコード
Python
# -*- coding: utf-8 -*-
import os
import datetime
import IPython
import IPython.display
import matplotlib as mpl
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import seaborn as sns
import tensorflow as tf
mpl.rcParams['figure.figsize'] = (12,9)
mpl.rcParams['axes.grid'] = False
import requests
from PIL import Image
from io import BytesIO
def image(url):
req = requests.get(url)
return Image.open(BytesIO(req.content))
import pandas_datareader.data as pdr
df = pdr.DataReader('aapl', 'yahoo', '2018/01/01')
df['Date Time'] = df.index
df.reset_index(inplace=True)
date_time = pd.to_datetime(df.pop('Date Time'), format='%d.%m.%Y %H:%M:%S')
df.drop(columns='Date', inplace=True)
# converting it to seconds:
timestamp_s = date_time.map(datetime.datetime.timestamp)
# del print
day = 24*60*60
year = (365.2425)*day
df['Day sin'] = np.sin(timestamp_s * (2*np.pi/day))
df['Day cos'] = np.cos(timestamp_s * (2*np.pi/day))
df['Year sin'] = np.sin(timestamp_s * (2*np.pi/year))
df['Year cos'] = np.cos(timestamp_s * (2*np.pi/year))
plt.plot(np.array(df['Day sin']))
# plt.plot(np.array(df['Day cos']))
plt.xlabel('Time [h]')
plt.title('Time of day signal')
"""## Split data"""
column_indices = {name:i for i, name in enumerate(df.columns)}
n = len(df)
train_df = df[0:int(n*0.7)]
val_df = df[int(n*0.7):int(n*0.9)]
test_df = df[int(n*0.9):]
num_features = df.shape[1]
train_mean = train_df.mean()
train_std = train_df.std()
train_df = (train_df - train_mean) / train_std
val_df = (val_df - train_mean) / train_std
test_df = (test_df - train_mean) / train_std
df_std = (df - train_mean) / train_std
df_std = df_std.melt(var_name='Column', value_name='Normalized')
"""# Data windowing"""
class WindowGenerator():
def __init__(self, input_width, label_width, shift,
train_df=train_df, val_df=val_df, test_df=test_df,
label_columns=None):
self.train_df = train_df
self.val_df = val_df
self.test_df =test_df
# for column indices of 'Label'
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(label_columns)}
# for window params
self.input_width = input_width
self.label_width = label_width
self.shift = shift
self.total_window_size = input_width + shift
self.input_slice = slice(0, input_width)
self.input_indices = np.arange(self.total_window_size)[self.input_slice]
self.label_start = self.total_window_size - self.label_width
self.label_slice = slice(self.label_start, None)
self.label_indices = np.arange(self.total_window_size)[self.label_slice]
def __repr__(self):
return '\n'.join([
f'Total window size: {self.total_window_size}',
f'Input indices: {self.input_indices}',
f'Label indices: {self.label_indices}',
f'Label column name(s): {self.label_columns}'
])
# 2 samples
w1 = WindowGenerator(input_width=24, label_width=1, shift=1, label_columns=['Adj Close'])
w1
w2 = WindowGenerator(input_width=6, label_width=1, shift=1, label_columns=['Adj Close'])
w2
def split_window(self, features):
inputs = features[:, self.input_slice, :]
labels = features[:, self.label_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
WindowGenerator.split_window = split_window
# stack 3 slices,,
exam_window = tf.stack([
np.array(train_df[:w2.total_window_size]),
np.array(train_df[100:100+w2.total_window_size]),
np.array(train_df[200:200+w2.total_window_size])
])
exam_inputs, exam_labels = w2.split_window(exam_window)
print('All shapes: (batch, time, features)')
print(f'Window shape: {exam_window.shape}')
print(f'Inputs shape: {exam_inputs.shape}')
print(f'Labels shape: {exam_labels.shape}')
w2.example = exam_inputs, exam_labels
def plot(self, model=None, plot_col='Adj Close', max_subplot=3):
inputs, labels = self.example
plt.figure(figsize=(12,8))
plot_col_index = self.column_indices[plot_col]
max_n = min(max_subplot, len(inputs))
for n in range(max_n):
plt.subplot(3, 1, n+1)
plt.ylabel(f'{plot_col} [normed]')
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='#2ca02c', 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='#ff7f0e', s=64)
if n==0:
plt.legend()
plt.xlabel('Time [h]')
WindowGenerator.plot = plot
def make_dataset(self, data):
data = np.array(data, dtype=np.float32)
ds = tf.keras.preprocessing.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_window)
return ds
WindowGenerator.make_dataset = make_dataset
@property
def train(self):
return self.make_dataset(self.train_df)
@property
def val(self):
return self.make_dataset(self.val_df)
@property
def test(self):
return self.make_dataset(self.test_df)
WindowGenerator.train = train
WindowGenerator.val = val
WindowGenerator.test = test
for exam_inputs, exam_labels in w2.train.take(1):
print(f'Inputs shape (batch, time, features): {exam_inputs.shape}')
print(f'Labels shape (batch, time, features): {exam_labels.shape}')
# by Dense
MAX_EPOCHS = 20
def compile_and_fit(model, window, patience=2):
early_stopping = tf.keras.callbacks.EarlyStopping(monitor='val_loss',
patience=patience,
mode='min')
model.compile(loss=tf.losses.MeanSquaredError(),
optimizer=tf.optimizers.Adam(),
metrics=[tf.metrics.MeanAbsoluteError()])
history = model.fit(window.train, epochs=MAX_EPOCHS,
validation_data=window.val,
callbacks=[early_stopping])
return history
dense = tf.keras.Sequential([
tf.keras.layers.Dense(units=64, activation='relu'),
tf.keras.layers.Dense(units=64, activation='relu'),
tf.keras.layers.Dense(units=1)
])
CONV_WIDTH = 20
conv_window = WindowGenerator(
input_width=CONV_WIDTH,
label_width=1,
shift=1,
label_columns=['Adj Close']
)
multi_step_dense = tf.keras.Sequential([
# Shape: (time, features) => (time*features)
tf.keras.layers.Flatten(),
tf.keras.layers.Dense(units=32, activation='relu'),
tf.keras.layers.Dense(units=32, activation='relu'),
tf.keras.layers.Dense(units=1),
# Add back the time dimension.
# Shape: (outputs) => (1, outputs)
tf.keras.layers.Reshape([1, -1]),
])
history = compile_and_fit(multi_step_dense, conv_window)
# nan event
'''======
Epoch 1/20
17/17 [==============================] - 1s 34ms/step - loss: nan - mean_absolute_error: nan - val_loss: nan - val_mean_absolute_error: nan
Epoch 2/20
17/17 [==============================] - 0s 12ms/step - loss: nan - mean_absolute_error: nan - val_loss: nan - val_mean_absolute_error: nan
[ ]
======='''
試したこと
これらを修正するために
こちら https://teratail.com/questions/299561 や
こちら https://teratail.com/questions/295771 を参考にさせていただき
前処理方法から着手してみましたが
変化がなく困っております。
私が取得したデータについて、なぜ欠損値になってしまうのでしょうか。
学習できるようにするにはどの視点からコードを修正したらよいでしょうか。
教えていただけると幸いです。
補足情報（FW/ツールのバージョンなど）
Google Colab
chrome webbrowser
＜参考＞
１．[TensorFlow時系列分析]
https://www.tensorflow.org/tutorials/structured_data/time_series
２．[tensorflow.keras _ 損失関数や評価指標メトリックについて]
https://keras.io/api/metrics/
＜ゴール＞
最終的にはこちら https://teratail.com/questions/299561 のようなアンサンブルにできる
モデル生成器をを完成させたいです。
