keras中变长序列的softmax

我有二维形状的样品 (say [a, 10]) . 其中a从一个样本变为另一个样本。我正在和 batch size = 1 避免批量大小可变的问题。我创造了如下 LSTM 网络。现在问题是我的目标是形状的概率向量 [1,a,1] . 每个样本的概率向量之和为1。

我想申请 softmax 激活最后一层,这样我就可以与目标进行比较。我该怎么办?

    Layer (type)                            Output Shape                        Param #       
==========================================================================================
lstm_21 (LSTM)                          (1, None, 32)                       7808          
__________________________________________________________________________________________
lstm_22 (LSTM)                          (1, None, 8)                        1312          
__________________________________________________________________________________________
time_distributed_6 (TimeDistributed)    (1, None, 1)                        9             
==========================================================================================

这是我的密码

model = tf.keras.models.Sequential()
model.add(tf.keras.layers.LSTM(32, return_sequences=True, batch_input_shape=(1, None, len(features))))
model.add(tf.keras.layers.LSTM(8, return_sequences=True))
model.add(tf.keras.layers.Dense(1, activation='softmax'))
# model.add(tf.keras.layers.TimeDistributed(tf.keras.layers.Dense(1, activation='softmax')))

print(model.summary(90))

model.compile(loss = 'mean_squared_error',
              optimizer = 'adam')


def generate_arrays_from_pd(df, arr_df):
    while True:
        for i in range(arr_df.shape[0]):
            a1 = arr_df[i, 0]
            a2 = arr_df[i, 1]
            batch_x = df.loc[a1:a2, features].as_matrix().reshape((1, -1, len(features)))
            batch_y = df.loc[a1:a2, "mkt_shr"].as_matrix().reshape((1, -1, 1))
            yield(batch_x, batch_y)

model.fit_generator(generate_arrays_from_pd(dat_train, arr_train), steps_per_epoch=arr_train.shape[0], epochs = 10, verbose=1, shuffle=False)