MXBoard with keras-mxnet

Discussion
1

s MXBoard supported in Keras with MXNet backend? I managed to get my network (which I previously used with the TensorFlow backend) training with the MXNet backend, but now I want to monitor the performance as I was used to with TensorBoard. This is what I tried:

from keras.losses import mean_squared_error
sw = SummaryWriter(logdir="logs/test", flush_secs=10)
sw.add_scalar(tag='training_MSE', value=mean_squared_error)

but this gives the following error:
TypeError: _make_numpy_array only accepts input types of numpy.ndarray, scalar, and MXNet NDArray, while received type <class 'function'>

How can MXBoard be used in Keras?

2

Did you try using Tensorboard instead of MXBoard? E.g.

tensorboard_callback = keras.callbacks.TensorBoard(log_dir="logs/test")
model.fit(x_train, y_train,
          batch_size=batch_size,
          epochs=epochs,
          validation_data=(x_test, y_test),
         callbacks=[tensorboard_callback])
3

Thank you, installing TensorFlow in my MXNet environment actually allowed to use the TensorBoard module!