In [ ]:
%matplotlib inline
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
1. 讀入深度學習套件¶
In [ ]:
from tensorflow.keras.preprocessing import sequence
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Embedding
from tensorflow.keras.layers import LSTM
from tensorflow.keras.datasets import imdb
2. 讀入數據¶
一般自然語言處理, 我們會限制最大要使用的字數。
In [ ]:
(x_train, y_train), (x_test, y_test) = imdb.load_data(num_words=10000)
Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/imdb.npz 17465344/17464789 [==============================] - 0s 0us/step 17473536/17464789 [==============================] - 0s 0us/step
In [ ]:
print(f'訓練資料筆數：{len(x_train)}')
print(f'測試資料筆數：{len(x_test)}')
訓練資料筆數：25000 測試資料筆數：25000
注意每筆評論的長度當然是不一樣的。
In [ ]:
print(f'第一筆訓練資料的長度：{len(x_train[0])}')
print(f'第二筆測試資料的長度：{len(x_train[1])}')
第一筆訓練資料的長度：218 第二筆測試資料的長度：189
In [ ]:
print(f'第一筆資料的標籤：{y_train[0]}(正評)')
print(f'第二筆資料的標籤：{y_train[1]}(負評)')
第一筆資料的標籤：1(正評) 第二筆資料的標籤：0(負評)
3. 資料處理¶
雖然我們可以做真的 seq2seq, 可是資料長度不一樣對計算上有麻煩, 因此平常還是會固定一定長度, 其餘補 0。
In [ ]:
x_train = sequence.pad_sequences(x_train, maxlen=100)
x_test = sequence.pad_sequences(x_test, maxlen=100)
4. step 01: 打造一個函數學習機¶
In [ ]:
model = Sequential()
In [ ]:
model.add(Embedding(10000, 128))
In [ ]:
model.add(LSTM(128))
In [ ]:
model.add(Dense(1, activation='sigmoid'))
組裝¶
In [ ]:
model.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['accuracy'])
欣賞我們的 model¶
In [ ]:
model.summary()
Model: "sequential"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
embedding (Embedding) (None, None, 128) 1280000
lstm (LSTM) (None, 128) 131584
dense (Dense) (None, 1) 129
=================================================================
Total params: 1,411,713
Trainable params: 1,411,713
Non-trainable params: 0
_________________________________________________________________
5. step 02: 訓練¶
In [ ]:
model.fit(x_train, y_train, batch_size=32, epochs=10,
validation_data=(x_test, y_test))
Epoch 1/10 782/782 [==============================] - 22s 24ms/step - loss: 0.4183 - accuracy: 0.8095 - val_loss: 0.3533 - val_accuracy: 0.8436 Epoch 2/10 782/782 [==============================] - 18s 24ms/step - loss: 0.2572 - accuracy: 0.8975 - val_loss: 0.3465 - val_accuracy: 0.8459 Epoch 3/10 782/782 [==============================] - 19s 24ms/step - loss: 0.1825 - accuracy: 0.9292 - val_loss: 0.4315 - val_accuracy: 0.8412 Epoch 4/10 782/782 [==============================] - 18s 23ms/step - loss: 0.1317 - accuracy: 0.9512 - val_loss: 0.4378 - val_accuracy: 0.8267 Epoch 5/10 782/782 [==============================] - 19s 24ms/step - loss: 0.0930 - accuracy: 0.9672 - val_loss: 0.6900 - val_accuracy: 0.8292 Epoch 6/10 782/782 [==============================] - 18s 24ms/step - loss: 0.0794 - accuracy: 0.9730 - val_loss: 0.5700 - val_accuracy: 0.8333 Epoch 7/10 782/782 [==============================] - 19s 24ms/step - loss: 0.0505 - accuracy: 0.9833 - val_loss: 0.6798 - val_accuracy: 0.8370 Epoch 8/10 782/782 [==============================] - 18s 24ms/step - loss: 0.0460 - accuracy: 0.9848 - val_loss: 0.6846 - val_accuracy: 0.8300 Epoch 9/10 782/782 [==============================] - 18s 23ms/step - loss: 0.0302 - accuracy: 0.9912 - val_loss: 0.7452 - val_accuracy: 0.8279 Epoch 10/10 782/782 [==============================] - 18s 23ms/step - loss: 0.0195 - accuracy: 0.9941 - val_loss: 0.8609 - val_accuracy: 0.8259
Out[ ]:
<tensorflow.python.keras.callbacks.History at 0x7f8b51a9e490>
6. 換個存檔方式¶
這次是把 model 和訓練權重分開存, 使用上更有彈性。
In [ ]:
from google.colab import drive
drive.mount('/content/drive')
Mounted at /content/drive
In [ ]:
%cd '/content/drive/My Drive/Colab Notebooks'
/content/drive/My Drive/Colab Notebooks
In [ ]:
model_json = model.to_json()
open('imdb_model_architecture.json', 'w').write(model_json)
model.save_weights('imdb_model_weights.h5')