def train(self):
# train conv stack layer by layer
for i, stack in enumerate(self.conv_stack):
if self.checkpoint_file != '':
model = load(self.checkpoint_file)
self.net = FastNet(self.learning_rate, self.image_shape, self.n_out, initModel=model)
# delete softmax layer
self.net.del_layer()
self.net.del_layer()
# for i in range(len(self.fc_params)):
# self.net.del_layer()
self.net.disable_bprop()
layerParam = stack + [self.fakefc_param, self.softmax_param]
self.net.append_layers_from_dict(layerParam)
self.init_data_provider()
self.scheduler.reset()
self.scheduler.set_level(i)
self.test_outputs = []
self.train_output = []
AutoStopTrainer.train(self)
# train fc layer
for i, stack in enumerate(self.fc_stack):
model = load(self.checkpoint_file)
self.net = FastNet(self.learning_rate, self.image_shape, self.n_out, initModel=model)
self.net.del_layer()
self.net.del_layer()
self.net.disable_bprop()
if i == len(self.fc_stack) - 1:
layerParam = stack + [self.softmax_param]
else:
layerParam = stack + [self.fakefc_param, self.softmax_param]
self.net.append_layers_from_dict(layerParam)
self.init_data_provider()
self.scheduler.reset()
self.scheduler.set_level(i)
self.test_outputs = []
self.train_output = []
AutoStopTrainer.train(self)
model = load(self.checkpoint_file)
self.test_id += 1
self.net = FastNet(self.learning_rate, self.image_shape, self.n_out, initModel=model)
self.test_range = self.origin_test_range
self.init_data_provider()
self.scheduler = Scheduler(self)
self.num_epoch /= 2
AutoStopTrainer.train(self)
def train(self):
MiniBatchTrainer.train(self)
for i, cate in enumerate(self.range_list):
self.set_category_range(cate)
self.num_batch = self.curr_epoch = self.curr_batch = 0
self.curr_minibatch = 0
self.num_minibatch = self.train_minibatch_list[i]
model = load(self.checkpoint_file)
layers = model['model_state']['layers']
for l in layers:
if l['type'] == 'fc':
l['weight'] = None
l['bias'] = None
l['weightIncr'] = None
l['biasIncr'] = None
fc = layers[-2]
fc['outputSize'] = cate
self.learning_rate = self.learning_rate_list[i]
self.net = FastNet(self.learning_rate, self.image_shape, self.n_out, init_model = model)
self.net.clear_weight_incr()
MiniBatchTrainer.train(self)
def _get_next_batch(self):
self.get_next_index()
if self.curr_batch_index == 0:
random.shuffle(self.batch_range)
self.curr_epoch += 1
self.curr_batch = self.batch_range[self.curr_batch_index]
# print self.batch_range, self.curr_batch
filename = os.path.join(self.data_dir, 'data_batch_%d' % self.curr_batch)
data = util.load(filename)
img = data['data'] - self.batch_meta['data_mean']
self.labels = np.array(data['labels'])
self.data = np.require(img, requirements='C', dtype=np.float32)
def _get_next_batch(self):
self.get_next_index()
if self.curr_batch_index == 0:
random.shuffle(self.batch_range)
self.curr_epoch += 1
self.curr_batch = self.batch_range[self.curr_batch_index]
# print self.batch_range, self.curr_batch
filename = os.path.join(self.data_dir,
'data_batch_%d' % self.curr_batch)
self.data = util.load(filename)
return BatchData(self.data['data'] - self.batch_meta['data_mean'],
np.array(self.data['labels']), self.curr_epoch,
self.curr_batch)
def _get_next_batch(self):
self.get_next_index()
if self.curr_batch_index == 0:
random.shuffle(self.batch_range)
self.curr_epoch += 1
self.curr_batch = self.batch_range[self.curr_batch_index]
# print self.batch_range, self.curr_batch
filename = os.path.join(self.data_dir, 'data_batch_%d' % self.curr_batch)
self.data = util.load(filename)
return BatchData(self.data['data'] - self.batch_meta['data_mean'],
np.array(self.data['labels']),
self.curr_epoch,
self.curr_batch)
def _get_next_batch(self):
self.get_next_index()
if self.curr_batch_index == 0:
random.shuffle(self.batch_range)
self.curr_epoch += 1
self.curr_batch = self.batch_range[self.curr_batch_index]
# print self.batch_range, self.curr_batch
filename = os.path.join(self.data_dir,
'data_batch_%d' % self.curr_batch)
data = util.load(filename)
img = data['data'] - self.batch_meta['data_mean']
self.labels = np.array(data['labels'])
self.data = np.require(img, requirements='C', dtype=np.float32)
def _get_next_batch(self):
self.get_next_index()
if self.curr_batch_index == 0:
random.shuffle(self.batch_range)
self.curr_epoch += 1
self.curr_batch = self.batch_range[self.curr_batch_index]
filename = os.path.join(self.data_dir + '.%s' % self.curr_batch)
data_dic = util.load(filename)
#data = np.concantenate([data[self.data_name] for data in data_list], axis = 1)
#labels = np.concatenate([np.array( data['labels'].tolist() ) for data in data_list])
data = data_dic[self.data_name].transpose()
labels = data_dic['labels']
self.labels = labels
data = np.require(data, requirements='C', dtype=np.float32)
self.data = data
def _get_next_batch(self):
self.get_next_index()
if self.curr_batch_index == 0:
random.shuffle(self.batch_range)
self.curr_epoch += 1
self.curr_batch = self.batch_range[self.curr_batch_index]
filename = os.path.join(self.data_dir + '.%s' % self.curr_batch)
data_dic = util.load(filename)
#data = np.concantenate([data[self.data_name] for data in data_list], axis = 1)
#labels = np.concatenate([np.array( data['labels'].tolist() ) for data in data_list])
data = data_dic[self.data_name].transpose()
labels = data_dic['labels']
self.labels = labels
data = np.require(data, requirements='C', dtype=np.float32)
self.data = data
def train(self):
AutoStopTrainer.train(self)
if self.layerwised:
for i in range(len(self.n_filters) - 1):
next_n_filter = [self.n_filters[i + 1]]
next_size_filter = [self.size_filters[i + 1]]
model = load(self.checkpoint_file)
self.net = FastNet(self.learning_rate, self.image_shape, 0, initModel=model)
self.net.del_layer()
self.net.del_layer()
self.net.disable_bprop()
self.net.add_parameterized_layers(next_n_filter, next_size_filter, self.fc_nouts)
self.init_data_provider()
self.scheduler = Scheduler(self)
self.test_outputs = []
self.train_outputs = []
AutoStopTrainer.train(self)
def __init__(self, data_dir='.', batch_range=None):
self.data_dir = data_dir
self.meta_file = os.path.join(data_dir, 'batches.meta')
self.curr_batch_index = 0
self.curr_batch = None
self.curr_epoch = 1
self.data = None
if os.path.exists(self.meta_file):
self.batch_meta = util.load(self.meta_file)
else:
print 'No default meta file \'batches.meta\', using another meta file'
if batch_range is None:
self.batch_range = self.get_batch_indexes()
else:
self.batch_range = batch_range
random.shuffle(self.batch_range)
def __init__(self, data_dir='.', batch_range=None):
self.data_dir = data_dir
self.meta_file = os.path.join(data_dir, 'batches.meta')
self.curr_batch_index = 0
self.curr_batch = None
self.curr_epoch = 1
self.data = None
if os.path.exists(self.meta_file):
self.batch_meta = util.load(self.meta_file)
else:
print 'No default meta file \'batches.meta\', using another meta file'
if batch_range is None:
self.batch_range = self.get_batch_indexes()
else:
self.batch_range = batch_range
random.shuffle(self.batch_range)
def __init__(self, checkpoint_dir, test_id):
self.checkpoint_dir = checkpoint_dir
if not os.path.exists(self.checkpoint_dir):
os.system('mkdir -p \'%s\'' % self.checkpoint_dir)
self.test_id = test_id
self.regex = re.compile('^test%d-(\d+)$' % self.test_id)
cp_pattern = self.checkpoint_dir + '/test%d-*' % self.test_id
cp_files = glob.glob(cp_pattern)
if not cp_files:
self.checkpoint = None
self.checkpoint_file = None
else:
self.checkpoint_file = sorted(cp_files, key=os.path.getmtime)[-1]
util.log('Loading from checkpoint file: %s', self.checkpoint_file)
self.checkpoint = util.load(self.checkpoint_file)
def __init__(self, checkpoint_dir, test_id):
self.checkpoint_dir = checkpoint_dir
if not os.path.exists(self.checkpoint_dir):
os.system('mkdir -p \'%s\'' % self.checkpoint_dir)
self.test_id = test_id
self.regex = re.compile('^test%d-(\d+)$' % self.test_id)
cp_pattern = self.checkpoint_dir + '/test%d-*' % self.test_id
cp_files = glob.glob(cp_pattern)
if not cp_files:
self.checkpoint = None
self.checkpoint_file = None
else:
self.checkpoint_file = sorted(cp_files, key=os.path.getmtime)[-1]
util.log('Loading from checkpoint file: %s', self.checkpoint_file)
self.checkpoint = util.load(self.checkpoint_file)
def train(self):
MiniBatchTrainer.train(self)
for i, group in enumerate(self.num_group_list):
self.set_num_group(group)
self.num_batch = self.curr_epoch = self.curr_batch = 0
self.curr_minibatch = 0
self.num_minibatch = self.train_minibatch_list[i]
model = load(self.checkpoint_file)
layers = model['model_state']['layers']
fc = layers[-2]
fc['outputSize'] = group
fc['weight'] = None
fc['bias'] = None
fc['weightIncr'] = None
fc['biasIncr'] = None
self.learning_rate = self.learning_rate_list[i]
self.net = FastNet(self.learning_rate, self.image_shape, self.n_out, init_model = model)
self.net.clear_weight_incr()
MiniBatchTrainer.train(self)
def __init__(self, checkpoint, show_filters, channels = 3): self.checkpoint = checkpoint self.model = util.load(self.checkpoint) self.layers = self.model['layers'] self.show_filters = show_filters self.channels = channels
param_dict['factor'] = factor[0]
else:
param_dict['factor'] = factor
learning_rate = util.string_to_float_list(args.learning_rate)
if len(learning_rate) == 1:
param_dict['learning_rate'] = learning_rate[0]
else:
param_dict['learning_rate'] = learning_rate
param_dict['batch_size'] = args.batch_size
param_dict['checkpoint_dir'] = args.checkpoint_dir
trainer = args.trainer
cp_pattern = param_dict['checkpoint_dir'] + '/test%d' % param_dict['test_id']
cp_files = glob.glob('%s*' % cp_pattern)
if not cp_files:
util.log('No checkpoint, starting from scratch.')
param_dict['init_model'] = Parser(args.param_file).get_result()
else:
cp_file = sorted(cp_files, key=os.path.getmtime)[-1]
util.log('Loading from checkpoint file: %s', cp_file)
param_dict['init_model'] = util.load(cp_file)
trainer = get_trainer_by_name(trainer, param_dict, args)
util.log('start to train...')
trainer.train()
#trainer.predict(['pool5'], 'image.opt')
def __init__(self, checkpoint, show_filters, channels = 3): self.checkpoint = checkpoint self.model = util.load(self.checkpoint) self.layers = self.model['model_state']['layers'] self.show_filters = show_filters self.channels = channels