Question

我使用Recurrent NN生成分类模型。以下是使用keras的简单RNN模型（由blog提供）：

tr_C, ts_C, tr_r, ts_r = train_test_split(C, r, train_size=.8)
tr_C = tr_C.reshape((tr_C.shape[0], tr_C.shape[1], 1))
ts_C = ts_C.reshape((ts_C.shape[0], ts_C.shape[1], 1))
tr_r = tr_r.reshape(tr_r.shape[0], 1)
ts_r = ts_r.reshape(ts_r.shape[0], 1)

print('Build model...')
model = Sequential()
model.add(LSTM(16,
               batch_input_shape=(1, 1, 1),
               activation='tanh',
               recurrent_activation='tanh',
               kernel_initializer='he_uniform',
               bias_initializer='he_uniform',
               # dropout=0.1,
               # recurrent_dropout=0.1,
               stateful=True,
               return_sequences=False,
               )
          )
model.add(Dense(1, activation='tanh'))
model.compile(loss='mean_absolute_error', optimizer='adam', metrics=['accuracy'])


for epoch in range(10):
    print ('epoch(%s): ' % epoch)
    mean_tr_acc = []
    mean_tr_loss = []
    for i in range(len(tr_C)):
        for j in range(max_len):
            tr_loss, tr_acc = model.train_on_batch(np.expand_dims(np.expand_dims(tr_C[i][j], axis=1), axis=1), tr_r[i])
            mean_tr_acc.append(tr_acc)
            mean_tr_loss.append(tr_loss)
        model.reset_states()

    print('training_acc = {}'.format(np.mean(mean_tr_acc)))
    print('loss training = {}'.format(np.mean(mean_tr_loss)))
    print('___________________________________')

    mean_te_acc = []
    mean_te_loss = []
    for i in range(len(ts_C)):
        for j in range(max_len):
            te_loss, te_acc = model.test_on_batch(np.expand_dims(np.expand_dims(ts_C[i][j], axis=1), axis=1), ts_r[i])
            mean_te_acc.append(te_acc)
            mean_te_loss.append(te_loss)
        model.reset_states()

        for j in range(max_len):
            y_pred = model.predict_on_batch(np.expand_dims(np.expand_dims(ts_C[i][j], axis=1), axis=1))
        model.reset_states()

    print('testing_acc = {}'.format(np.mean(mean_te_acc)))
    print('loss testing = {}'.format(np.mean(mean_te_loss)))
    print('___________________________________')

这是模型的摘要：

Layer (type)                 Output Shape              Param #   
=================================================================
lstm_1 (LSTM)                (1, 16)                   1152      
_________________________________________________________________
dense_1 (Dense)              (1, 1)                    17        
=================================================================
Total params: 1,169
Trainable params: 1,169
Non-trainable params: 0
_________________________________________________________________

我可以修改此代码以训练一批样本，而不是一次训练一个样本吗？如果是这样，怎么样？任何建议表示赞赏谢谢。

Answer 1

你在这里声明：

model.add(LSTM(16,
               batch_input_shape=(1, 1, 1),
               activation='tanh',
               recurrent_activation='tanh',
               kernel_initializer='he_uniform',
               bias_initializer='he_uniform',
               # dropout=0.1,
               # recurrent_dropout=0.1,
               stateful=True,
               return_sequences=False,
               )
          )

批量大小应为：（1,1,1）仅表示1个示例，1个时间戳，每个时间戳为标量。
如果您想要超过1个培训示例，您应该这样做：

batch_input_shape=(None,1,1)

或，将None更改为批次中的示例数（无意味着您可以根据需要插入多少个示例）

将一个简单的火车/测试转换为一批火车/测试

1 个答案: