A problem about LSTM learning

[LinRick]

Dear sir,
When I use the following LSTM architecture along with keras/elephas, as:
conf = SparkConf().setAppName('RNN_LSTM_Elephas_App').setMaster('spark://master:7077')

one driver and four workers

sc = SparkContext(conf=conf)

rdd = to_simple_rdd(sc, train_X, train_y)

model = Sequential()
model.add(LSTM(32, input_shape=(train_X.shape[1],train_X.shape[2]), activation='tanh', kernel_initializer='lecun_uniform'))
model.add(Dense(1))
model.compile(loss="mse", optimizer='adam')
sparkRNN_model = SparkModel(model, frequency='epoch', mode='asynchronous')
sparkRNN_model.fit(rdd, epochs = 150, batch_size = 32, verbose = 2, validation_split = 0.1)

The predicting results are the same values as follows:
array([[33.390114],
[33.390114],
[33.390114],
[33.390114],
..., [33.390114]]

By the way, when I use setMaster('local[*]'), the results are different values.
Hence, I have no idea about this problem.

Rick

[danielenricocahall]

Elephas currently doesn't support LSTM training/prediction - it can be added if enough users request it! Although I would definitely want some assistance/feedback.

[danielenricocahall]

Moved this issue to the new fork: danielenricocahall#10. Closing this for now but still on the radar!