def test_net():
'''
Tests stuff
'''
import data_loader
d = data_loader.Loader()
d.load()
sizes = [d.x_dim, 30, d.y_dim]
net = Network(sizes)
k = d.size / 3
training_data, test_data = d.data[k:], d.data[:k]
net.SGD2(training_data, 500, 3.0, 2, test_data=test_data)
def test_net2():
import data_loader
d = data_loader.Loader()
d.load2()
np.random.shuffle(d.data)
k = d.size / 3
data_r = [(x, y[4:] * 10e10) for x, y in d.data]
training_data, test_data = data_r[k:], data_r[:k]
sizes = [d.x_dim, len(training_data[0][1])]
net = Network(sizes)
net.SGDR(training_data, 100, 3.0, 10, test_data=test_data)
results = [net.feedforward(x) for x, y in test_data]
training_data_c = [(np.concatenate((x, y[4:])), y[:2])
for x, y in d.data[k:]]
test_data_c = [(np.concatenate((x[0], y)), x[1][:2])
for x, y in zip(d.data[:k], results)]
x_dim = training_data_c[0][0].shape[0]
y_dim = training_data_c[0][1].shape[0]
#sizes_c = [x_dim,10,10,10,10,y_dim]
sizes_c = [x_dim, y_dim]
net2 = Network(sizes_c)
net2.SGD2(training_data_c, 200, 0.1, 5, test_data=test_data_c)
print 'sizes %s' % sizes
print 'sizes_c %s ' % sizes_c
# 默认从data/train.pkl中读取训练数据
# 保存训练结果到save文件夹
import tensorflow as tf
import compo_net
import data_loader
import os
if __name__ == '__main__':
config = compo_net.Config(training=True, learning_rate=0)
config.restore = os.path.exists(os.path.join(config.save_dir,
'checkpoint'))
if config.restore:
print(':: restore from saved model')
loader = data_loader.Loader(config)
model = compo_net.Model(config)
merged_summary = tf.summary.merge_all()
sess = tf.Session()
summary_writer = tf.summary.FileWriter(config.log_dir, sess.graph)
sess.run(tf.global_variables_initializer())
if config.restore:
model.restore(sess)
input('NOTICE: initial learning rate is 0, reset after start')
input(':: press enter to start training')
times = 1
cost = 0
while True:
#print('predict_for_list_of_models')
best_prediction_forest, best_forest = self.predict_for_list_of_models(list_of_forests)
#print('plot conf matrix')
self.plot_confusion_matrix(best_prediction_forest, 'rforest')
classes_dict_pos_svm, classes_dict_neg_svm = self.roc_curve(best_prediction_svm)
classes_dict_pos_fr, classes_dict_neg_fr =self.roc_curve(best_prediction_forest)
self.plot_roc(classes_dict_pos_svm, classes_dict_neg_svm)
self.plot_predicted(best_prediction_svm, 'svm')
self.plot_predicted(best_prediction_forest, 'forest')
data_load = data_loader.Loader("consensus_county15.txt")
data = data_load.estimation_data
validation_data = data_load.validation_data[:,1:]
header = data_load.header[1:]
print(len(header))
pca = Main(data, header, validation_data)
#categories should stay the same
pca.choose_categories(header, 'Poverty')
pca.choose_categories(header, 'Poverty', for_validation=True)
#pca.choose_categories(['County', 'Income', 'White', 'TotalPop', 'Unemployment','PublicWork'], 'Poverty')
#pca.choose_categories(['County', 'Income', 'White', 'TotalPop', 'Unemployment','PublicWork'], 'Poverty', for_validation=True)
#print('classify_poverty')
#!/usr/bin/python3
import value_net
import data_loader
import config
import tensorflow as tf
import os.path
if __name__ == '__main__':
model = value_net.Model()
data_loader = data_loader.Loader()
input(':: press enter to start training')
learning_rate = config.default_lr
times = 1
while True:
try:
inputs, targets = data_loader.get_next_batch()
cost = model.train(inputs, targets, learning_rate)
# when use cross entroy, this may happen
assert cost == cost, 'cost is nan'
print('batch: {0}, cost: {1}'.format(times, cost))
times += 1
except KeyboardInterrupt:
cmd = input('\noperation(w/q/c/l/t):')
if cmd == 'w':
model.save()