def train(opts):
opts.cfg['pretrained'] = root + opts.cfg['pretrained']
now = datetime.datetime.now().strftime('%Y_%m_%d-%H_%M_%S')
opts.cfg['utc_time'] = now
opts.out = get_log_dir(opts.cfg['output_dir'], 1, opts.cfg, root)
#print(opts.cfg)
kwargs = {'num_workers': opts.cfg['n_workers']} if cuda else {}
train_loader = torch.utils.data.DataLoader(
Cloud_Data(class_names,
root,
args.image_type,
image_set='train',
backbone=opts.cfg['backbone'],
transform=True),
batch_size=opts.cfg['batch_size'],
shuffle=True,
**kwargs)
val_loader = torch.utils.data.DataLoader(Cloud_Data(
class_names,
root,
args.image_type,
image_set='val',
backbone=opts.cfg['backbone'],
transform=True),
batch_size=opts.cfg['batch_size'],
shuffle=False,
**kwargs)
data_loader = [train_loader, val_loader]
trainer = Trainer(data_loader, opts)
start_epoch = 0
start_iteration = 0
'''
if opts.cfg['pretrained']:
checkpoint = torch.load(opts.cfg['pretrained'])
start_epoch = checkpoint['epoch']
start_iteration = checkpoint['iteration']
'''
trainer.epoch = start_epoch
trainer.iteration = start_iteration
trainer.Train()
val_loader = torch.utils.data.DataLoader(
Cloud_Data(
class_names,
root,
args.image_type,
image_set='val',
backbone=opts.cfg['backbone'],
transform=True),
batch_size=opts.cfg['batch_size'],
shuffle=False,
**kwargs
)
data_loader=[train_loader,val_loader]
trainer = Trainer(data_loader, opts)
start_epoch = 0
start_iteration = 0
'''
if opts.cfg['pretrained']:
checkpoint = torch.load(opts.cfg['pretrained'])
start_epoch = checkpoint['epoch']
start_iteration = checkpoint['iteration']
'''
trainer.epoch = start_epoch
trainer.iteration = start_iteration
trainer.Train()
from neuralnetwork import NeuralNetwork
from trainer import Trainer
import matplotlib.pyplot as plt
import pickle
network = pickle.load(open("test.boi", "rb"))
ashketchum = Trainer(4, "adadelta", learningRate=0.05, delta=0.9)
ashketchum.addData(np.matrix([[0], [0]]), np.matrix([[0]]))
ashketchum.addData(np.matrix([[0], [1]]), np.matrix([[0]]))
ashketchum.addData(np.matrix([[1], [0]]), np.matrix([[0]]))
ashketchum.addData(np.matrix([[1], [1]]), np.matrix([[1]]))
errors = []
ashketchum.Train(network)
errors.append(ashketchum.error)
plt.show()
axes = plt.gca()
axes.set_xlim(0, len(errors))
axes.set_ylim(0, errors[0])
line, = axes.plot(range(len(errors)), errors)
while ashketchum.error > 0.025:
ashketchum.Train(network)
errors.append(ashketchum.error)
#print ashketchum.error
axes.set_xlim(0, len(errors))
axes.set_ylim(0, max(errors))
line.set_xdata(range(len(errors)))
