Reload weights into fc layer after compression

[pius kwao]

Hi,

    I am trying to store weights in a the fc layers in a compressed sparse row format. When I retrieve the weights and convert them to csr matrix format , it's size in memory 

reduces drastically but when I load it back into caffe my model size remains the same. Basically this is what I'm doing:

                                                             temp2 = net.params['ip1'][0].data.shape

                                                             sparse_csr1 = sparse.scr_matrix(temp2, shape)

                                                             net.params['ip1'][0].data[...] = sparse_csr1

                                                             net.save('compressed.caffemodel')

Any suggestions will be appreciated