可变序列长度数据的分类

Question

I have this data with varying number of elements in a row

sample feat1  feat2 feat3 feat4 feat5 feat6 feat7
 1       1      200  250    312   474  
 1       2      170  280    370
 ...
 1       12     220  400    470   520  620   720
 2       1      130  320    430   580  612   
 ...
 N       12     70   180    270   410

I found this sequence classification

from keras.datasets import imdb
from keras.models import Sequential
from keras.layers import Dense
from keras.layers import LSTM
from keras.layers.convolutional import Convolution1D
from keras.layers.convolutional import MaxPooling1D
from keras.layers.embeddings import Embedding
from keras.preprocessing import sequence
numpy.random.seed(7)
top_words = 5000
(X_train, y_train), (X_test, y_test) = imdb.load_data(nb_words=top_words)
# truncate and pad input sequences
max_review_length = 500
X_train = sequence.pad_sequences(X_train, maxlen=max_review_length)
X_test = sequence.pad_sequences(X_test, maxlen=max_review_length)
# create the model
embedding_vecor_length = 32
model = Sequential()
model.add(Embedding(top_words, embedding_vecor_length, input_length=max_review_length))
model.add(Convolution1D(nb_filter=32, filter_length=3, border_mode='same', activation='relu'))
model.add(MaxPooling1D(pool_length=2))
model.add(LSTM(100))
model.add(Dense(1, activation='sigmoid'))
model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])
print(model.summary())
model.fit(X_train, y_train, nb_epoch=3, batch_size=64)

Can I use this or modify to use it? Some direction would be nice.

Also, if you have better suggestion which algorithm to use or how to do it please suggest.

Answer 1

A general approach would be to specify a specific value that means "unknown". For example, if all your values are positive you can pick it to be -1.

sample feat1  feat2 feat3 feat4 feat5 feat6 feat7
 1       1      200  250    312   474    -1    -1
 1       2      170  280    370    -1    -1    -1
 ...
 1       12     220  400    470   520   620   720
 2       1      130  320    430   580   612    -1  
 ...
 N       12     70   180    270   410    -1    -1

The network then learns to ignore this value.

There is even a built in function called pad_sequences that does this for you.