DL4J Integration

[jalilsh]

Hi,

I am using DL4J MLP to train and predict my data. Is there any way to see the network weights and biases after training? the MLP model can be saved with “DL4J Model Writer”, but the saved file can not be read and interpreted.

thanks

[DaveK]

Hi @jalilsh,

the file written by the DL4J Model Writer is just a .zip file. You can just extract it. Within, there are a bunch of other files giving configuration and there is another file ending with “model”, which contain the weights. It is in turn another .zip file, which can be extracted. Within, there is a coefficients.bin binary file containing the actual weights. Its saved using the code provided by DL4J, hence you need to check their documentation for instructions how to read the file.

Another option would be to use our Keras integration instead of DL4J (which is generally recommended), which save the weights as a .h5 file. This can be easily read using, e.g., Python.

Cheers,
David

[jalilsh]

thanks for your help.

I will elaborate and will come back if needed