How does keras handle complex dataset?

[Noob]

I am currently working with a dataset which has complex values, and seem to get a warning that I am bit concerned about. 

Train start
Train on 221286 samples, validate on 24588 samples
Epoch 1/150
/usr/local/lib/python2.7/dist-packages/numpy/core/numeric.py:482: ComplexWarning: Casting complex values to real discards the imaginary part
  return array(a, dtype, copy=False, order=order)
221286/221286 [==============================] - 22s - loss: 0.0154 - val_loss: 0.0166
Epoch 2/150

How do I make it not discard the complex part.. ?

Any way around it?

Here is the code:

print "Model definition!"
    model = Sequential()
model.add(Dense(output_dim=13, input_dim=1025, init="normal",activation=K.tanh))

reduce_lr=ReduceLROnPlateau(monitor='val_loss', factor=0.01, patience=3, verbose=1, mode='auto', epsilon=0.001, cooldown=0, min_lr=0.00000001)
    stop  = EarlyStopping(monitor='val_loss', min_delta=0, patience=5, verbose=1, mode='auto')


    log=csv_logger = CSVLogger('training_'+str(i)+'.csv')




    hist_current = model.fit(train_set_data_vstacked_normalized[train],
                                train_set_output_vstacked_normlized[train],
                                shuffle=False,
                                validation_data=(train_set_data_vstacked_normalized[test],train_set_output_vstacked_normlized[test]),
                                validation_split=0.1,
                                nb_epoch=150,
                                verbose=1,
                                callbacks=[reduce_lr,log,stop])

[Matias Valdenegro]

You could just treat complex numbers as 2D vectors (real and imaginary parts) and feed that to the neural network. You just need to preprocess the data correctly to just keep the real coefficients and not have complex numbers.

[Noob]

I am working with complex matrices.. so i've kind of done what you mention. Splitted into real and img, concatenated them, and doubled the input_dim. 
Lets see if it works. 

[cr88...@gmail.com]

I would like to know if finally your Idea worked because I'm facing the same situation with a matrix of complex values to train in Keras

[manue...@gmail.com]

Does anyone knows how to deal with complex numbers in keras? I have the same issue.

Regards

[pachpan...@gmail.com]

Did anyone solve this issue?

[pachpan...@gmail.com]

Did you solve this issue?