parser = argparse.ArgumentParser(description='Compare to Tesseract OCR')
parser.add_argument('-m',
'--model',
metavar='m',
type=str,
default='model.pth')
parser.add_argument('input_dir', type=str, help='Input directory')
args = parser.parse_args()
device = torch.device('cpu')
dataset = data.MyDataset(args.input_dir)
net = model.MyNet().to(device)
net.load_state_dict(torch.load(args.model, map_location='cpu'))
net.eval()
torch.no_grad()
time_sum_tess = 0
time_sum_mynet = 0
accuracy_sum_tess = 0.0
accuracy_sum_mynet = 0.0
accuracy_worst_tess = float('inf')
accuracy_worst_mynet = float('inf')
num_total = 0
import pandas as pd
import torch
import model
import processData
if __name__ == "__main__":
dp = processData.DataProcessing()
train_x, train_y = dp.load_data('data/train.csv')
test_x = dp.load_data('data/test.csv')
model = model.MyNet()
model.train(train_x, train_y, 100)
model.write_result(test_x)
def train_on_cpu():
#merged = tf.summary.merge_all()
#print(merged)
p1 = tf.placeholder(tf.float32, [None, 20, 40, 3])
p2 = tf.placeholder(tf.float32, [None, 8, 10])
p3 = tf.placeholder(tf.float32, [None, 10])
label = tf.placeholder(tf.float32, [None, 1])
y_pred = m.MyNet(p1, p2, p3)
#mse = tf.reduce_sum((label -y_pred)**2)
mse = tf.losses.mean_squared_error(label, y_pred)
mse_train_summary = tf.summary.scalar('train mse', mse)
mse_test_summary = tf.summary.scalar('test mse', mse)
mse_train_hist = tf.summary.histogram('train_mse_hist', mse)
mse_test_hist = tf.summary.histogram('test_mse_hist', mse)
#mse = tf.nn.l2_loss(label,y_pred)
train_opti = tf.train.AdamOptimizer(1e-4).minimize(mse)
saver = tf.train.Saver(max_to_keep=10)
sess.run(tf.global_variables_initializer())
# Compute for epochs.
all_train_steps = 0
all_test_steps = 0
if C.RESTORE_MODEL:
saver.restore(sess, tf.train.latest_checkpoint(C.LOAD_MODEL_PATH))
# search for checkpoint file
print("模型开始增量训练====>")
graph = tf.get_default_graph()
for i in range(C.EPOCHS):
train_next_element = get_data(C.TRAIN_RECODER_PATH)
test_next_element = get_data(C.TEST_RECODER_PATH)
train_bacth_per_epoch = 0
while True:
try:
train_bacth_per_epoch += 1
all_train_steps += 1
train_mse_loss = get_loss(train_next_element, p1, p2, p3,
label, mse, train_opti,
all_train_steps, train_writer,
mse_train_summary, mse_train_hist)
print("Train Epoch {} Batch {} All steps {}: "
"MSE ==> {}".format(i + 1, train_bacth_per_epoch,
all_train_steps, train_mse_loss))
if all_train_steps % C.SAVE_FREQUENCY == 0:
print("model is saved at all train step {}".format(
all_train_steps))
saver.save(sess,
C.REGTRSSION_MODEL_SAVE_PATH,
global_step=C.MAX_STEP)
except tf.errors.OutOfRangeError:
break
test_loss = 0
test_batchs = 0
while True:
try:
all_test_steps += 1
test_mse_loss = get_loss(test_next_element, p1, p2, p3, label,
mse, train_opti, all_test_steps,
test_writer, mse_test_summary,
mse_test_hist)
test_loss += test_mse_loss
test_batchs += 1
print("Test batch {} : MSE ==> {}".format(
test_batchs, test_mse_loss))
#tf.summary.histogram('test_mse_loss_steps', test_mse_loss)
except tf.errors.OutOfRangeError:
break
test_loss_ave = test_loss / test_batchs
print("Epoch {} : All Test Data Average MSE {}".format(
i + 1, test_loss_ave))
test_writer.close()
train_writer.close()
def train_on_gpu():
with tf.device("/cpu:0"):
global_step = tf.train.get_or_create_global_step()
tower_grads = []
p1 = tf.placeholder(tf.float32, [None, 20, 40, 3])
p2 = tf.placeholder(tf.float32, [None, 8, 10])
p3 = tf.placeholder(tf.float32, [None, 10])
label = tf.placeholder(tf.float32, [None, 1])
train_opti = tf.train.AdamOptimizer(1e-4)
saver = tf.train.Saver(max_to_keep=10)
with tf.variable_scope(tf.get_variable_scope()):
for i in range(num_gpus):
#with tf.device(assign_to_device('/gpu:{}'.format(i), ps_device='/cpu:0')):
with tf.device('/gpu:{}'.format(i)):
print(
"assign_to_device====>>>>",
assign_to_device('/gpu:{}'.format(i),
ps_device='/cpu:0'))
_x1 = p1[i * C.BATCH_SIZE:(i + 1) * C.BATCH_SIZE]
_x2 = p2[i * C.BATCH_SIZE:(i + 1) * C.BATCH_SIZE]
_x3 = p3[i * C.BATCH_SIZE:(i + 1) * C.BATCH_SIZE]
_y = label[i * C.BATCH_SIZE:(i + 1) * C.BATCH_SIZE]
y_pred = m.MyNet(_x1, _x2, _x3)
tf.get_variable_scope().reuse_variables()
mse = tf.losses.mean_squared_error(_y, y_pred)
grads = train_opti.compute_gradients(mse)
tower_grads.append(grads)
grads = average_gradients(tower_grads)
train_opti = train_opti.apply_gradients(grads)
sess.run(tf.global_variables_initializer())
# Compute for epochs.
all_train_steps = 0
all_test_steps = 0
if C.RESTORE_MODEL:
saver.restore(sess, tf.train.latest_checkpoint(C.LOAD_MODEL_PATH))
#search for checkpoint file
print("模型开始增量训练====>")
graph = tf.get_default_graph()
for i in range(C.EPOCHS):
train_next_element = get_data_gpu(C.TRAIN_RECODER_PATH,
num_gpus * C.BATCH_SIZE)
test_next_element = get_data(C.TEST_RECODER_PATH)
train_bacth_per_epoch = 0
while True:
try:
train_bacth_per_epoch += 1
all_train_steps += 1
mse_train_summary = tf.summary.scalar('train mse', mse)
mse_test_summary = tf.summary.scalar('test mse', mse)
mse_train_hist = tf.summary.histogram(
'train_mse_hist', mse)
mse_test_hist = tf.summary.histogram('test_mse_hist', mse)
train_mse_loss = get_loss(train_next_element, p1, p2, p3,
label, mse, train_opti,
all_train_steps, train_writer,
mse_train_summary,
mse_train_hist)
if all_train_steps % 500 == 0:
print("Train Epoch {} Batch {} All steps {}: "
"MSE ==> {}".format(i + 1, train_bacth_per_epoch,
all_train_steps,
train_mse_loss))
if all_train_steps % C.SAVE_FREQUENCY == 0:
# print("model is saved at all train step {}".format(all_train_steps))
saver.save(sess,
C.REGTRSSION_MODEL_SAVE_PATH,
global_step=C.MAX_STEP)
except tf.errors.OutOfRangeError:
break
test_loss = 0
test_batchs = 0
while True:
try:
all_test_steps += 1
test_mse_loss = get_loss(test_next_element, p1, p2, p3,
label, mse, train_opti,
all_test_steps, test_writer,
mse_test_summary, mse_test_hist)
test_loss += test_mse_loss
test_batchs += 1
# print("Test batch {} : MSE ==> {}".format(test_batchs, test_mse_loss))
# tf.summary.histogram('test_mse_loss_steps', test_mse_loss)
except tf.errors.OutOfRangeError:
break
test_loss_ave = test_loss / test_batchs
print("Epoch {} : All Test Data Average MSE {}".format(
i + 1, test_loss_ave))
test_writer.close()
train_writer.close()