lr = 1e-2
decay = (1, 0.2)
n_epoch = 5
batch_size = 1000
# data
x_tr = np.load('data/x_tr.npy')
y_tr = np.load('data/y_tr.npy')
x_te = np.load('data/x_te.npy')
y_te = np.load('data/y_te.npy')
y_tr = y_tr.astype(int)
y_te = y_te.astype(int)
scaler_x = StandardScaler()
x_tr = scaler_x.fit_transform(x_tr) / 3
x_te = scaler_x.transform(x_te) / 3
x_tr = make_tensor(x_tr, 'x_tr')
y_tr = make_tensor(y_tr, 'y_tr', dtype=tf.int64)
x_te = make_tensor(x_te, 'x_te')
y_te = make_tensor(y_te, 'y_te', dtype=tf.int64)
# batches
x_batch, y_batch = _make_batches(x_tr, y_tr, batch_size)
x_te_batch, y_te_batch = _make_batches(x_te, y_te, batch_size, test=True)
# Loss
y_batch_oh = tf.one_hot(y_batch, 2)
pred = net.predict(x_batch)
loss = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits(labels=y_batch_oh,
logits=pred))
pred_labels = tf.argmax(pred, axis=1)
np.save('W1.npy', W1)
np.save('b1.npy', b1)
np.save('W2.npy', W2)
np.save('b2.npy', b2)
np.save('W3.npy', W3)
np.save('b3.npy', b3)
np.save('W4.npy', W4)
with tf.Graph().as_default():
data_dir = "data_reg/"
net = NN(H1=20, H2=20, d=4, p=0.5)
lr = 1e-2
# data
x_tr, y_tr, x_te, y_te = prepare_data(data_dir, mode='numpy')
x_tr = make_tensor(x_tr, 'x_tr')
y_tr = make_tensor(y_tr, 'y_tr')
x_te = make_tensor(x_te, 'x_te')
y_te = make_tensor(y_te, 'y_te')
pred_te = net.predict(x_te)
r2_te = r2(pred_te, y_te)
mse_te = mse(pred_te, y_te)
decay = (100, 0.2)
n_epoch = 300
batch_size = 200
sample = tf.train.slice_input_producer([x_tr, y_tr])
x_batch, y_batch = tf.train.batch(sample, batch_size)
pred = net.predict(x_batch)
loss = mse(pred, y_batch)