def pre_train(mini_batch_size=1000):
lsq = LSQ()
'''bug bug bug'''
sample_dict = tcdp.random_select_samples()
mode = 'train'
for _, (index, features, labels) in enumerate(tcdl.iter_mini_batches(mode, sample_dict[mode], mini_batch_size=mini_batch_size)):
y, rmse = lsq.train(features, labels)
# np.savetxt('ys_train_lsq.txt', np.hstack((index, labels, y)))
print(y.tolist()[-6:], '\n', labels.tolist()[-6:], '\n', rmse)
break
mode = 'valid'
for _, (index, features, labels) in enumerate(tcdl.iter_mini_batches('train', sample_dict[mode], mini_batch_size=2000)):
y, rmse = lsq.predict(features, labels)
# np.savetxt('ys_valid_lsq.txt', np.hstack((index, labels, y)))
print(y.tolist()[-6:], '\n', labels.tolist()[-6:], '\n', rmse)
break
# mode = 'testA'
# for _, (index, features, labels) in enumerate(tcdl.iter_mini_batches(mode, sample_dict[mode], mini_batch_size=2000)):
# y, rmse = lsq.predict(features, labels)
# np.savetxt('ys_testA_lsq.txt', np.hstack((index, labels, y)))
# print(y.tolist()[-6:], '\n', labels.tolist()[-6:], '\n', rmse)
# break
#
# print('\n')
return (lsq.weights).astype(np.float32)
return ori_labels, pre_labels
if __name__ == '__main__':
n_input = 4 * 101 * 101
n_hidden = 1024
n_output = 1
n_steps = 15
dropout = 0.75
cnn = cnn2d(n_input, n_hidden, n_output, n_steps, dropout)
print('a cnn model is created! ')
for i in range(128):
sample_dict = tcdp.random_select_samples()
print('training ... ')
cnn2d_train('train', cnn, sample_dict, i)
print('validating ... ')
ys_train = cnn2d_valid('train', cnn, sample_dict, i)
ys_valid = cnn2d_valid('valid', cnn, sample_dict, i)
ys_testA = cnn2d_valid('testA', cnn, sample_dict, i)
obj = {'train': ys_train, 'valid': ys_valid, 'testA': ys_testA}
fn2 = 'ys_cnn_model_%d.cpk' % (i + 1)
loader.saveObject(obj, fn2)
''''''
# #break