我正在学习深度学习，
我正在使用IMDB数据集。 它正在处理[整数编码]？
一些示例表明您只是在进行深度学习，而不是转换为单热编码。
这足以获得有效的结果吗？
如果是这样的话
单热编码的优势是什么？
这是我的代码吗
import numpy
from keras.datasets import imdb
from keras.models import Sequential
from keras.layers import Dense
from keras.layers import Flatten
from keras.layers.convolutional import Conv1D
from keras.layers.convolutional import MaxPooling1D
from keras.layers.embeddings import Embedding
from keras.preprocessing import sequence
# load the dataset but only keep the top n words, zero the rest
top_words = 5000
max_words = 500
X_train = train_result
y_train = train_label
X_test = test_result
y_test = test_label# pad dataset to a maximum review length in words
X_train = sequence.pad_sequences(X_train, maxlen=max_words)
X_test = sequence.pad_sequences(X_test, maxlen=max_words)
print(X_train[:1])
# create the model
model = Sequential()
model.add(Embedding(top_words, 32, input_length=max_words))
model.add(Conv1D(filters=32, kernel_size=3, padding='same', activation='relu'))
model.add(MaxPooling1D(pool_size=2))
model.add(Flatten())
model.add(Dense(250, activation='relu'))
model.add(Dense(1, activation='sigmoid'))
model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])
model.summary()
# Fit the model
hist = model.fit(X_train, y_train, validation_data=(X_test, y_test), epochs=20, batch_size=128, verbose=1)
# Final evaluation of the model
scores = model.evaluate(X_test, y_test, verbose=0)
print("Accuracy: %.2f%%" % (scores[1]*100))
在X_train[1]。
[[ 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 3 284 2452 756 1 3075 194
54 3717 10 757 169 2216 5 1 1906 843 54 52 2732 3403
5 1819 3 34 4 54 1819 5 2532 42 668 23 54 709
52 7 9 2 80 172 3258 265 33 1 1467 4 683 4
11 21 988 1 3 110 631 2 4 321 3 3040 294 284
478 33 1 33 54 4349 33 54 213 2 86 54 516 420
754 1 84 2 8 526 473 63 20 184 20 184 20 184
1138 52 3 23 1 1468 101 3 1850 4 61 6 777 20
237 185 52 3846 5 54 149 7 34 4 1 18 54 4802
929 2 5 98 8 13 17 9 1 993 117 101 3 165
41 653 781 3 286 923 2882 7 210 3 181 5 1743 3
120 814 1630 1517 3 2317 4606 4425 9 43 686 5 744 1018
910 223 136 3782 1585 775 1391 3041 155 3 292 4 2975 2
136 135 120 864 24 869 3655 245 421 1 1803 10 1 120
2 1 261 78 1671 19 43 1288 16 1 1036 5 380 1
1744 121 10 1 84 252 55 51 670 2 24 200 51 1709
1 1256 1469 2 1 217 5 2453 423 79 929 36 9 3
1106 4 2754 4526 14 29 24 2393 74 34 4049 17 42 72
9 365 1 69 41 1804 572 41 559 76 92 2 153 112
11 15 835 1423 136 1 59 15 67 1 1320 5 441 2
733 17 1 688 890 5 26 421 55 23 208 2 31 2070
23 1 2998 136 6 413 44 33 40 7 119 9 668 4
22 3213 40 7 119 151 359 5 25 185]]
这是输出，
Epoch 1/20
10103/10103 [==============================] - 5s 523us/step - loss: 0.5812 - acc: 0.6589 - val_loss: 0.1229 - val_acc: 0.9551
Epoch 2/20
10103/10103 [==============================] - 5s 478us/step - loss: 0.1299 - acc: 0.9485 - val_loss: 0.0693 - val_acc: 0.9663
Epoch 3/20
10103/10103 [==============================] - 5s 488us/step - loss: 0.0544 - acc: 0.9824 - val_loss: 0.0589 - val_acc: 0.9775
Epoch 4/20
10103/10103 [==============================] - 5s 488us/step - loss: 0.0258 - acc: 0.9923 - val_loss: 0.0371 - val_acc: 0.9850
Epoch 5/20
10103/10103 [==============================] - 5s 483us/step - loss: 0.0120 - acc: 0.9976 - val_loss: 0.0528 - val_acc: 0.9813
Epoch 6/20
10103/10103 [==============================] - 5s 483us/step - loss: 0.0058 - acc: 0.9991 - val_loss: 0.0464 - val_acc: 0.9850
Epoch 7/20
10103/10103 [==============================] - 5s 482us/step - loss: 0.0032 - acc: 0.9994 - val_loss: 0.0707 - val_acc: 0.9738
Epoch 8/20
10103/10103 [==============================] - 5s 485us/step - loss: 0.0022 - acc: 0.9997 - val_loss: 0.0471 - val_acc: 0.9925
Epoch 9/20
10103/10103 [==============================] - 5s 482us/step - loss: 0.0011 - acc: 0.9998 - val_loss: 0.0698 - val_acc: 0.9775
Epoch 10/20
10103/10103 [==============================] - 5s 481us/step - loss: 6.8280e-04 - acc: 1.0000 - val_loss: 0.0728 - val_acc: 0.9775
Epoch 11/20
10103/10103 [==============================] - 5s 483us/step - loss: 4.8174e-04 - acc: 1.0000 - val_loss: 0.0873 - val_acc: 0.9738
Epoch 12/20
10103/10103 [==============================] - 5s 477us/step - loss: 3.4037e-04 - acc: 1.0000 - val_loss: 0.0674 - val_acc: 0.9813
Epoch 13/20
10103/10103 [==============================] - 5s 478us/step - loss: 2.6164e-04 - acc: 1.0000 - val_loss: 0.0847 - val_acc: 0.9775
Epoch 14/20
10103/10103 [==============================] - 5s 478us/step - loss: 2.0453e-04 - acc: 1.0000 - val_loss: 0.0812 - val_acc: 0.9775
Epoch 15/20
10103/10103 [==============================] - 5s 473us/step - loss: 1.6034e-04 - acc: 1.0000 - val_loss: 0.0831 - val_acc: 0.9775
Epoch 16/20
10103/10103 [==============================] - 5s 469us/step - loss: 1.3443e-04 - acc: 1.0000 - val_loss: 0.0874 - val_acc: 0.9775
Epoch 17/20
10103/10103 [==============================] - 5s 467us/step - loss: 1.1035e-04 - acc: 1.0000 - val_loss: 0.0891 - val_acc: 0.9775
Epoch 18/20
10103/10103 [==============================] - 5s 471us/step - loss: 9.3257e-05 - acc: 1.0000 - val_loss: 0.0956 - val_acc: 0.9775
Epoch 19/20
10103/10103 [==============================] - 5s 465us/step - loss: 7.9740e-05 - acc: 1.0000 - val_loss: 0.0965 - val_acc: 0.9775
Epoch 20/20
10103/10103 [==============================] - 5s 467us/step - loss: 6.7700e-05 - acc: 1.0000 - val_loss: 0.0919 - val_acc: 0.9775
Accuracy: 97.75%
整数编码意味着标签中存在一些有序关系，因此在构建分类模型时需要独热嵌入。本质上，独热嵌入是将离散数据映射到欧几里得空间。
例如，这里的数据集包括 3 个类别：苹果、橙子、香蕉。如果你使用整数编码：{apple => 0，橙色=> 1，香蕉=> 2}，你永远不能说"橙色"大于或大于"苹果"。
在您的情况下，IMDB评论数据集是一个二元分类数据集，有两种标签：负面和正面。您可以将它们作为连续特征处理：如果预测值接近 1，则审查率更积极，反之亦然。
https://www.quora.com/What-are-good-ways-to-handle-discrete-and-continuous-inputs-together
为什么一个热编码可以提高机器学习性能？