def parse_images(self):
if self.args.training or self.args.validation:
self.fetch_mnist_data()
if self.args.training:
self.sets['training'] = {'images': [], 'correct_answers': []}
x_training = self.mnist_ds['training']['default'][:]
y_training = self.mnist_ds['training']['targets'][:]
for x in x_training[0]:
self.sets['training']['images'].append(x)
for y in y_training[0]:
self.sets['training']['correct_answers'].append(y[0])
if self.args.validation:
self.sets['validation'] = {'images': [], 'correct_answers': []}
x_validation = self.mnist_ds['validation']['default'][:]
y_validation = self.mnist_ds['validation']['targets'][:]
for x in x_validation[0]:
self.sets['validation']['images'].append(x)
for y in y_validation[0]:
self.sets['validation']['correct_answers'].append(y[0])
if self.args.demo:
"""
import pickle
demo_set_filename = 'demo_python2'
demo_set = pickle.load(open(demo_set_filename, "rb"))
"""
from demo import load_flat_text_cases
demo_inputs, demo_targets = load_flat_text_cases(
'demo100_text.txt')
for i in range(len(demo_inputs)):
demo_inputs[i] = map(lambda value: value / 255.0,
demo_inputs[i])
adapted_image_array = []
for row_idx in range(28):
adapted_image_array.append([])
for col_idx in range(28):
adapted_image_array[row_idx].append(
demo_inputs[i][row_idx * 28 + col_idx])
demo_inputs[i] = adapted_image_array
self.sets['demo'] = {
'images': demo_inputs,
'correct_answers': demo_targets
}
if self.args.input:
self.sets['input'] = {'images': []}
if self.args.input.endswith('.png'):
image_array = img_helper.read_image(self.args.input)
self.sets['input']['images'].append(image_array)
elif self.args.input:
f = open(self.args.input)
lines = []
for line in f:
lines.append(line.strip())
f.close()
if 'png' in lines[0]:
for file_name in lines:
if file_name.endswith('.png'):
image_array = img_helper.read_image(file_name)
self.sets['input']['images'].append(image_array)
else:
print('Failed to parse custom input')
import os
from demo import load_flat_text_cases
import numpy as np
"""
This python file reads the flat mnist txt files, preprocesses it and finally saves it in a format
that the training module can read
"""
data_dir = '.'
mnist_training_file_path = os.path.join(data_dir, 'all_flat_mnist_training_cases_text.txt')
mnist_validation_file_path = os.path.join(data_dir, 'all_flat_mnist_testing_cases_text.txt')
hdf_file = os.path.join(data_dir, 'MNIST.hdf5')
print("Extracting and preprocessing MNIST data...")
training_inputs, training_targets = load_flat_text_cases(mnist_training_file_path)
for i in range(len(training_inputs)):
training_inputs[i] = map(lambda value: value / 255.0, training_inputs[i]) # normalize
training_inputs = np.array(training_inputs).reshape((1, 60000, 28, 28, 1))
training_targets = np.array(training_targets).reshape((1, 60000, 1))
validation_inputs, validation_targets = load_flat_text_cases(mnist_validation_file_path)
for i in range(len(validation_inputs)):
validation_inputs[i] = map(lambda value: value / 255.0, validation_inputs[i]) # normalize
validation_inputs = np.array(validation_inputs).reshape((1, 10000, 28, 28, 1))
validation_targets = np.array(validation_targets).reshape((1, 10000, 1))
print("Creating HDF5 dataset...")
f = h5py.File(hdf_file, 'w')
def parse_images(self):
if self.args.training or self.args.validation:
self.fetch_mnist_data()
if self.args.training:
self.sets["training"] = {"images": [], "correct_answers": []}
x_training = self.mnist_ds["training"]["default"][:]
y_training = self.mnist_ds["training"]["targets"][:]
for x in x_training[0]:
self.sets["training"]["images"].append(x)
for y in y_training[0]:
self.sets["training"]["correct_answers"].append(y[0])
if self.args.validation:
self.sets["validation"] = {"images": [], "correct_answers": []}
x_validation = self.mnist_ds["validation"]["default"][:]
y_validation = self.mnist_ds["validation"]["targets"][:]
for x in x_validation[0]:
self.sets["validation"]["images"].append(x)
for y in y_validation[0]:
self.sets["validation"]["correct_answers"].append(y[0])
if self.args.demo:
"""
import pickle
demo_set_filename = 'demo_python2'
demo_set = pickle.load(open(demo_set_filename, "rb"))
"""
from demo import load_flat_text_cases
demo_inputs, demo_targets = load_flat_text_cases("demo100_text.txt")
for i in range(len(demo_inputs)):
demo_inputs[i] = map(lambda value: value / 255.0, demo_inputs[i])
adapted_image_array = []
for row_idx in range(28):
adapted_image_array.append([])
for col_idx in range(28):
adapted_image_array[row_idx].append(demo_inputs[i][row_idx * 28 + col_idx])
demo_inputs[i] = adapted_image_array
self.sets["demo"] = {"images": demo_inputs, "correct_answers": demo_targets}
if self.args.input:
self.sets["input"] = {"images": []}
if self.args.input.endswith(".png"):
image_array = img_helper.read_image(self.args.input)
self.sets["input"]["images"].append(image_array)
elif self.args.input:
f = open(self.args.input)
lines = []
for line in f:
lines.append(line.strip())
f.close()
if "png" in lines[0]:
for file_name in lines:
if file_name.endswith(".png"):
image_array = img_helper.read_image(file_name)
self.sets["input"]["images"].append(image_array)
else:
print("Failed to parse custom input")