def do_sequence_detector():
with open('hypes/sequence_detector.json', 'r') as f:
H = json.load(f)
p = Preprocess()
images = np.array([]).reshape(0, side, side)
labels = np.array([]).reshape(0, 5)
for i in range(H['dataset_size']):
data = p.load_file('data/train_sequences{:02d}.pickle'.format(i))
imagesi, labelsi = preprocess_for_length_learner(data)
images = np.vstack((images, imagesi))
labels = np.vstack((labels, labelsi))
data = None
labels = labels.astype(int)
validation_images, validation_labels = preprocess_for_length_learner(
p.load_file('data/valid_sequences.pickle'))
validation_labels = validation_labels.astype(int)
NeuralNet.execute(H, images, labels, validation_images,
validation_labels)
import json
from neural_net import NeuralNet
from preprocess import Preprocess
import numpy as np
def preprocess_for_length_learner(data, sequence_length=5, side=280):
images = np.array(data['sequences'])
labels = np.array(data['labels'])[:, [0]].astype(int) - 1
return images, labels
p = Preprocess()
d = p.load_file('data/preprocessed_images.pickle')
images = d['trainset']
labels = d['train_labels'].reshape((len(d['train_labels']), 1))
validation_images = d['validset']
validation_labels = d['valid_labels'].reshape((len(d['valid_labels']), 1))
with open('hypes/single_digit.json', 'r') as f:
H = json.load(f)
NeuralNet.execute(H, images, labels, validation_images, validation_labels)
