keras - get different results on different data size -
i'm trying build model generate chiense text. problem when use several atrticles training datas , trained them each model.fit(), got bad results loss not decrease @ all. however, when combine several articles one, performs well. loss shrinks down rapidly. causes problem , how can fix it? need memory if combine datas one.....
here's code:
for filename in os.listdir(dir_path): filename = dir_path + filename tmp = " ".join(text_to_word_sequence(open(filename).read())) raw_text.append(tmp) # raw_text list articles seq_length = 25 model = sequential() model.add(lstm(n_vocab, input_shape=(seq_length, 1), return_sequences=true)) model.add(dropout(0.2)) model.add(lstm(n_vocab)) model.add(dropout(0.2)) model.add(dense(n_vocab, activation='softmax')) model.compile(loss='categorical_crossentropy', optimizer='adam') dataxs, datays = process(raw_text) in range(1000): j in range(len(dataxs)): model.fit(dataxs[j], datays[j], epochs=1, batch_size=128)
here's modified makes performance better:
all_text = "" filename in os.listdir(dir_path): filename = dir_path + filename tmp = " ".join(text_to_word_sequence(open(filename).read())) all_text += tmp raw_text.append(all_text)
thanks.
Comments
Post a Comment