def train_save(ds: Dataset, t_labs: np.ndarray, t_data: np.ndarray,
num_feats: int, num_class: int):
x_data, labels = placeholder(num_feats, num_class)
weight, bias, global_steps = variables(num_feats, num_class)
logits, predict = inferDense(x_data, weight, bias)
acc = calAcc(labels, predict)
train_op, loss = loss_train(labels, logits, global_steps)
tf.summary.scalar('acc', acc)
tf.summary.scalar('loss', loss)
init = tf.global_variables_initializer()
merged = tf.summary.merge_all()
with tf.Session() as sess:
sess.run(init)
writer = tf.summary.FileWriter('logs', sess.graph)
for y_batch, x_batch in ds:
_, gs = sess.run(fetches=[train_op, global_steps],
feed_dict={
x_data: x_batch,
labels: y_batch
})
if gs % 200 == 0:
acc_val, summary = sess.run(fetches=[acc, merged],
feed_dict={
x_data: t_data,
labels: t_labs
})
print(type(summary))
writer.add_summary(summary, gs)
print(acc_val)
writer.close()
def train_supervisor(ds: Dataset, t_labs: np.ndarray, t_data: np.ndarray, num_feats: int, num_class: int):
x_data, labels = placeholder(num_feats, num_class)
weight, bias, global_steps = variables(num_feats, num_class)
logits, predict = inferDense(x_data, weight, bias)
acc = calAcc(labels, predict)
train_op, loss = loss_train(labels, logits, global_steps)
tf.summary.scalar('acc', acc)
tf.summary.scalar('loss', loss)
init = tf.global_variables_initializer()
merged = tf.summary.merge_all()
sv = tf.train.Supervisor(logdir="supervise/", init_op=init, summary_op=merged)
with sv.managed_session() as sess:
sess.run(init)
while not sv.should_stop():
try:
y_batch, x_batch = next(ds)
print(x_batch.shape)
print(y_batch.shape)
_, gs = sess.run(fetches=[train_op, global_steps], feed_dict={
x_data: x_batch, labels: y_batch
})
except StopIteration:
sv.request_stop()
sv.close()
def train_save_model(ds: Dataset, t_labs: np.ndarray, t_data: np.ndarray,
num_feats: int, num_class: int):
x_data, labels = placeholder(num_feats, num_class)
weight, bias, global_steps = variables(num_feats, num_class)
logits, predict = inferDense(x_data, weight, bias)
acc = calAcc(labels, predict)
train_op, _ = loss_train(labels, logits, global_steps)
init = tf.global_variables_initializer()
saved_model_dir = 'saved_model/softmax'
with tf.Session() as sess:
sess.run(init)
writer = tf.summary.FileWriter('logs', sess.graph)
for y_batch, x_batch in ds:
_, gs = sess.run(fetches=[train_op, global_steps],
feed_dict={
x_data: x_batch,
labels: y_batch
})
if gs % 200 == 0:
acc_val = sess.run(fetches=[acc],
feed_dict={
x_data: t_data,
labels: t_labs
})
print(acc_val)
save_model('softmax_saved', sess, x_data, labels, predict, acc)
writer.close()
def restore2(t_labs: np.ndarray, t_data: np.ndarray, num_feats=180, num_class=3):
x_data, labels = placeholder(num_feats, num_class)
weight, bias, global_steps = variables(num_feats, num_class)
logits, pred, acc = infer(x_data, labels, weight, bias)
with tf.Session() as sess:
saver = tf.train.Saver()
saver.restore(sess, save_path='models/softmax')
predict, accuy = sess.run([pred, acc], feed_dict={x_data: t_data, labels: t_labs})
print(accuy)
return iterator, iterator.initializer
if __name__ == '__main__':
# trans_libsvm('data/dna.scale.t', 'data/dna.scale.tf.t')
# trans_libsvm('data/dna.scale.tr', 'data/dna.scale.tf.tr')
tf_files = ['data/dna.scale.tf.t', 'data/dna.scale.tf.tr']
num_feats, num_class, num_epochs, batch_size, num_threads = 180, 3, 1000, 128, 2
libsvmHelper = LibsvmHelper(num_feats, num_class, num_epochs, batch_size,
num_threads)
# label, index, value = libsvmHelper.pipeline(tf_files, tf.get_default_graph())
iterator, initializer = libsvmHelper.ds_iterator(tf_files)
label, index, value = iterator.get_next()
weight, bias, global_steps = variables(num_feats, num_class)
logits, pred = inferSparse(index, value, weight, bias)
train_op, loss = loss_train(tf.one_hot(label, num_class, 1, 0), logits,
global_steps)
# The op for initializing the variables.
init_op = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer(), initializer)
with tf.Session() as sess:
sess.run(init_op)
Writer = tf.summary.FileWriter(logdir='logs', graph=sess.graph)
while True:
try:
_, loss_, gs = sess.run([train_op, loss, global_steps])
if gs % 100 == 0: