具有不同模型参数的相同结果

Question

我正在建立一个神经网络模型来将情感分配给短语。我的问题是，无论我对模型进行什么更改，准确性得分始终相同。

我尝试添加或删除图层，更改批处理大小，更改时期数，更改激活器等。

tokenizer = Tokenizer()
total_phrases = phrases_train.append([phrases_valid,  phrases_test], 
                                     ignore_index = True) 

tokenizer.fit_on_texts(total_phrases)

maxlen = max([len(s.split()) for s in total_phrases])

vocab_size = len(tokenizer.word_index) + 1

phrases_train_tokens = tokenizer.texts_to_sequences(phrases_train)
phrases_valid_tokens = tokenizer.texts_to_sequences(phrases_valid)
phrases_test_tokens = tokenizer.texts_to_sequences(phrases_test)

pad_phrases_train = pad_sequences(phrases_train_tokens, maxlen=maxlen, 
                                  padding='post') 
pad_phrases_valid = pad_sequences(phrases_valid_tokens, maxlen=maxlen, 
                                  padding='post') 
pad_phrases_test = pad_sequences(phrases_valid_tokens, maxlen=maxlen, 
                                 padding='post')

model = Sequential()
model.add(Embedding(vocab_size, 100, input_length=maxlen))
model.add(Dense(50, kernel_initializer='truncated_normal', 
                activation='relu'))
model.add(GRU(units=32, dropout=0.2, recurrent_dropout=0.2))
model.add(Dense(1, activation='softmax'))

model.compile(loss='binary_crossentropy', optimizer='adam', metrics= 
              ['accuracy'])

model.fit(pad_phrases_train, sents_train, batch_size=250, epochs=50, 
          validation_data=(pad_phrases_valid, sents_valid), verbose=2)

model.evaluate(pad_phrases_test, sents_test)

评估始终准确显示0.2057分。我想达到50％左右的精度