Subversion Repositories keras_to_fpga

from keras.models import Model

from tensorflow.keras.models import load_model

import numpy as np

import os

import struct

import shutil

# -------------------------------------------------------

model = load_model('xor.h5')

model.summary()

# -------------------------------------------------------

x = np.array([[0,0],[0,1],[1,0],[1,1]])

# print(model.predict(x).round())

# print(model.predict(x))

# print('-' * 60)

def float_to_hex(f):

    # return hex(struct.unpack('<I', struct.pack('<f', f))[0])

    return format(struct.unpack('<I', struct.pack('<f', f))[0], 'x')

# -------------------------------------------------------

dir = 'weights'

if os.path.exists(dir):

    shutil.rmtree(dir)

os.makedirs(dir)

for i in range(len(model.layers)):

    layer = model.layers[i]

    print(layer.name)

    w = layer.get_weights()

    # for w in layer.get_weights():

      # print(w.shape)

      # print(w)

      # print('^' * 60)

    print(w[0].shape, w[1].shape)

    print('+' * 60)

    for y in range(0, w[0].shape[1]):

      print('-' * 60)

      file_name = dir + '/' + layer.name + '_' + str(y) + '.txt'

      print(file_name)

      with open(file_name, "w") as text_file:

        for x in range(0, w[0].shape[0]):

          # print(float_to_hex(w[0][x][0]))

          print(float_to_hex(w[0][x][y]), file=text_file)

        print(float_to_hex(w[1][y]), file=text_file)

      # close(text_file)

      print('^' * 60)

    # print(layer.get_weights())

    # a = layer.get_weights()

    # bias = a[1]

    # w = np.concatenate((a[0], bias[np.newaxis,:]), axis=0)

    # # w.astype('float32').tofile(layer.name)

    # print('+' * 60)

    # for x in range(0, w.shape[0]):

      # for y in range(0, w.shape[1]):        

        # print(float_to_hex(w[x,y]))

    # print('+' * 60)