Python の pkl ファイルを GitHub からダウンロードし、適当に変換して保存する

main.py

import numpy
import os
import pickle
import requests



def convert_data(targetData, fileName):

    with open(fileName, 'wb') as writingFileStream:

        # Output of npy file for comparison and verification
        numpy.save(fileName, targetData)

        # Obtain the dimensions of `targetData`
        ndimTargetData = numpy.ndim(targetData)

        # Write the dimension of `targetData`
        writingFileStream.write( ndimTargetData.to_bytes(4, 'little') )



        if (ndimTargetData == 1): # if `targetData` was 1D array

            # Write the number of elements in a 1D array
            writingFileStream.write( targetData.shape[0].to_bytes(4, 'little') )

            # Write the each element in this 1D array
            writingFileStream.write( targetData )



        elif (ndimTargetData == 2): # if `targetData` was 2D array

            # Write the number of elements in each dimension of a 2D array
            writingFileStream.write( targetData.shape[0].to_bytes(4, 'little') )
            writingFileStream.write( targetData.shape[1].to_bytes(4, 'little') )

            # Write the each element in this 2D array
            writingFileStream.write( targetData.ravel() )



if __name__ == '__main__':

    # URL of the file you want to download
    targetFileURL = \
        "https://github.com/oreilly-japan/deep-learning-from-scratch/blob/master/ch03/sample_weight.pkl"\
        .replace("github.com", "raw.githubusercontent.com")\
        .replace("/blob", "")

    # filename you want to download & save
    targetFileName = os.path.basename(targetFileURL)




    if os.path.isfile(targetFileName):

        print("The target file `" + targetFileName + "` already exists.")

    else:

        requestResponse = requests.get(url=targetFileURL)
        targetFileData  = pickle.loads(requestResponse.content)

        convert_data( targetFileData['b1'], 'sample_weight_b1.bin' )
        convert_data( targetFileData['b2'], 'sample_weight_b2.bin' )
        convert_data( targetFileData['b3'], 'sample_weight_b3.bin' )

        convert_data( targetFileData['W1'], 'sample_weight_W1.bin' )
        convert_data( targetFileData['W2'], 'sample_weight_W2.bin' )
        convert_data( targetFileData['W3'], 'sample_weight_W3.bin' )

        with open(targetFileName, 'wb') as writingFileStream:
            pickle.dump(targetFileData, writingFileStream)

# EOF