def extract_flatten_output():
conv = model.get_layer(index=0)
maxPool = model.get_layer(index=1)
flatten = model.get_layer(index=2)
input_img = x_test[:5] #7 2 1 0 4
conv_output = conv(input_img)
maxPool_output = maxPool(conv_output)
flatten_output = flatten(maxPool_output)
flatten_output_7 = flatten_output[0]
tool.write_to_file(flatten_output_7, path + "flatten_output_7.csv", 1)
def extract_conv_weights():
# extract conv bias
conv_bias = conv.weights[1]
tool.write_to_file(conv_bias.numpy(), path + "conv_bias.csv", 1)
# extract conv weights
conv_weights = conv.weights[0]
conv0_weights = np.empty((5, 5), dtype=float)
conv1_weights = np.empty((5, 5), dtype=float)
conv2_weights = np.empty((5, 5), dtype=float)
for i in range(5):
for j in range(5):
conv0_weights[i][j] = conv_weights[i][j][0][0]
conv1_weights[i][j] = conv_weights[i][j][0][1]
conv2_weights[i][j] = conv_weights[i][j][0][2]
tool.write_to_file(conv0_weights, path + "conv0_weights.csv", 2)
tool.write_to_file(conv1_weights, path + "conv1_weights.csv", 2)
tool.write_to_file(conv2_weights, path + "conv2_weights.csv", 2)
def extract_conv_output():
#extract conv output
# get 5 images from test and print label
input_img = x_test[:5] #7 2 1 0 4
conv = model.get_layer(index=0)
conv_output = conv(input_img)
conv_output_7 = conv_output[0]
conv0_output_7 = np.empty((24, 24), dtype=float)
conv1_output_7 = np.empty((24, 24), dtype=float)
conv2_output_7 = np.empty((24, 24), dtype=float)
for i in range(24):
for j in range(24):
conv0_output_7[i][j] = conv_output_7[i][j][0]
conv1_output_7[i][j] = conv_output_7[i][j][1]
conv2_output_7[i][j] = conv_output_7[i][j][2]
tool.write_to_file(conv0_output_7, path + "conv0_output_7.csv", 2)
tool.write_to_file(conv1_output_7, path + "conv1_output_7.csv", 2)
tool.write_to_file(conv2_output_7, path + "conv2_output_7.csv", 2)
def extract_maxPool_output():
#extract maxPool output
import numpy as np
input_img = x_test[:5] #7 2 1 0 4
conv = model.get_layer(index=0)
conv_output = conv(input_img)
maxPool = model.get_layer(index=1)
maxPool_output = maxPool(conv_output)
maxPool_output_7 = maxPool_output[0]
maxPool0_output_7 = np.empty((12, 12), dtype=float)
maxPool1_output_7 = np.empty((12, 12), dtype=float)
maxPool2_output_7 = np.empty((12, 12), dtype=float)
for i in range(12):
for j in range(12):
maxPool0_output_7[i][j] = maxPool_output_7[i][j][0]
maxPool1_output_7[i][j] = maxPool_output_7[i][j][1]
maxPool2_output_7[i][j] = maxPool_output_7[i][j][2]
tool.write_to_file(maxPool0_output_7, path + "maxPool0_output_7.csv", 2)
tool.write_to_file(maxPool1_output_7, path + "maxPool1_output_7.csv", 2)
tool.write_to_file(maxPool2_output_7, path + "maxPool2_output_7.csv", 2)
if neuron_result > 0: return neuron_result
else: return 0.0
def conv(matrix, bias, filter):
print("now is filter " + str(filter) + " with bias: " + str(bias))
conv_result = np.empty((24, 24), dtype=float)
conv_weights = np.empty((24, 24), dtype=float)
if filter == 0:
print("conv0")
conv_weights = conv0_weights
elif filter == 1:
print("conv1")
conv_weights = conv1_weights
else:
print("conv2")
conv_weights = conv2_weights
for i in range(24):
for j in range(24):
conv_matrix = getFilter(matrix, i, j, 5, 5)
conv_result[i][j] = neuron(conv_matrix, conv_weights, 5, 5, bias)
return conv_result
tool.write_to_file(conv(x_test[0], bias[0], 0),
path + "conv0_manual_output_7.csv", 2)
tool.write_to_file(conv(x_test[0], bias[1], 1),
path + "conv1_manual_output_7.csv", 2)
tool.write_to_file(conv(x_test[0], bias[2], 2),
path + "conv2_manual_output_7.csv", 2)
# this file is to extract model parameter and write it into csv file
import tensorflow as tf
import extract_tools as tool
path = tool.ann_path
# load model from h5 file
model = tf.keras.models.load_model('./tensorflow/resources/chisel_ann.h5')
# get layers
flatten = model.get_layer(index=0)
dense = model.get_layer(index=1)
dense1 = model.get_layer(index=2)
# write layers bias and weights to csv file(flatten doesn't need)
tool.write_to_file(dense.weights[1].numpy(), path + 'dense_weights_bias.csv',
1)
tool.write_to_file(dense.weights[0].numpy(), path + 'dense_weights.csv', 2)
tool.write_to_file(dense1.weights[1].numpy(), path + 'dense1_weights_bias.csv',
1)
tool.write_to_file(dense1.weights[0].numpy(), path + 'dense1_weights.csv', 2)
def extract_dense_weights():
dense_weights = dense.weights[0]
dense_bias = dense.weights[1]
tool.write_to_file(dense_bias.numpy(), path + "dense_bias.csv", 1)
tool.write_to_file(dense_weights.numpy(), path + "dense_weights.csv", 2)
model = tf.keras.models.load_model(path + 'chisel_ann.h5')
print(model.summary())
# get layers
flatten = model.get_layer(index=0)
dense = model.get_layer(index=1)
dense1 = model.get_layer(index=2)
# get 5 images from test and print label
input_img = tf.convert_to_tensor(x_test[:5])
for i in range(0, 5):
print(y_test[i])
# get flatten layer output
flatten_output = flatten(input_img)
tool.write_to_file(flatten_output[0].numpy(), path + 'flatten_output_7.csv', 1)
tool.write_to_file(flatten_output[1].numpy(), path + 'flatten_output_2.csv', 1)
tool.write_to_file(flatten_output[2].numpy(), path + 'flatten_output_1.csv', 1)
tool.write_to_file(flatten_output[3].numpy(), path + 'flatten_output_0.csv', 1)
tool.write_to_file(flatten_output[4].numpy(), path + 'flatten_output_4.csv', 1)
# get dense layer output
dense_output = dense(flatten_output)
tool.write_to_file(dense_output[0].numpy(), path + 'dense_output_7.csv', 1)
tool.write_to_file(dense_output[1].numpy(), path + 'dense_output_2.csv', 1)
tool.write_to_file(dense_output[2].numpy(), path + 'dense_output_1.csv', 1)
tool.write_to_file(dense_output[3].numpy(), path + 'dense_output_0.csv', 1)
tool.write_to_file(dense_output[4].numpy(), path + 'dense_output_4.csv', 1)
# get dense1 layer output
dense1_output = dense1(dense_output)