kedartatwawadi / nn_compression Goto Github PK

The code is hard to understand its function. So, would you please give us a usage to learn how to use that tool into the real application?

Dear sir,
hello,I try to run your code in Linux server, but there is a bug : ValueError: Dimensions must be equal, but are 256 and 130 for 'rnn/while/rnn/multi_rnn_cell/cell_0/cell_0/basic_rnn_cell/MatMul_1' (op: 'MatMul') with input shapes: [?,256], [130,128].
Is it my problem? Can you answer my problem? thank you

class generate_sequence_data exist：
Mismatch between array dtype ('|S1') and format specifier ('%c')

resolved method 👍
replace np.savetxt(FLAGS.file_name,data,delimiter='', fmt='%c',newline='');
use np.savetxt(FLAGS.file_name,data,delimiter='', fmt='%s',newline='');

I didn't find the source code for the arithmetic encoder&decoder, did I miss them?

The link to code on README is broken: https://github.com/kedartatwawadi/NN_compression/tree/master/tf_char_rnn

Why there is an addition with 2/N ?

In section 3.4

"We thus consider the DeepZip-Feat model which has as inputs past 50 symbols (instead of only 1),
and 5 4-gram context counts."

what's the input?
and where is the code?
thanks

I try to run the sample, and find out it seems currently the tool could support limited input characters, like string containing only 'a' and 'b', will the tool support all 256 characters given the minimum input unit is 1 byte, thanks.

Should I try higher markovity dataset to compress using Gzip? The datasets are all generate by your generating code.

where is the "data/sequence_data/input.txt"
i did not find this file

Sorry， I‘m new to train lstm. Will you kindly provide you training parameters? Since I can't get results as you provided. Or maybe it's because you didn't provide arithmetic coding block?