class CachedModel(Model):
def __init__(self,
inputs,
outputs,
name,
verbose,
version_date=None,
proj_dir=DEFAULT_DIR,
**model_params):
if version_date is None:
version = str(date.today())
else:
version = str(version_date)
self.dir = model_dir = os.path.join(proj_dir, 'models', version, name)
model_file = os.path.join(model_dir, 'model.h5')
# Check I'm already supposed to exist
if os.path.exists(model_file):
if verbose:
print('Loading Cached Model: models/' + version + '/' + name)
# Load myself
self.model = load_model(model_file)
else:
# I don't exist yet
if verbose:
print('Creating new model...')
if not os.path.exists(model_dir):
os.makedirs(model_dir)
self.model = Model(inputs=inputs,
outputs=outputs,
name=name,
**model_params)
# Save myself
if verbose:
print('Saving to models/' + version + '/' + name)
self.model.save(model_file)
model_attrs = list(self.model.__dict__.keys())
for attr in model_attrs:
self.__setattr__(attr, self.model.__getattribute__(attr))
def compile(self, verbose=1, **kwargs):
self.model.compile(**kwargs)
model_file = os.path.join(self.dir, 'model.h5')
if verbose:
self.model.save(model_file)
print('Cached to: \n %s' % model_file)
def fit(self, *args, **kwargs):
verbose = 1
force = False
if 'force' in list(kwargs.keys()):
force = kwargs.pop('force')
if 'verbose' in kwargs.keys():
verbose = kwargs['verbose']
weights_dir = os.path.join(self.dir, 'trained_weights')
weights_file = os.path.join(weights_dir, 'weights.h5')
if force:
if verbose:
print('Forced retraining...')
self.model.fit(*args, **kwargs)
elif os.path.exists(weights_file):
if verbose:
print('loading cached weights from ' + weights_file)
self.model.load_weights(weights_file)
else:
self.model.fit(*args, **kwargs)
if not os.path.exists(weights_dir):
os.makedirs(weights_dir)
if verbose:
print('caching weights to: \n ' + weights_file)
self.model.save_weights(weights_file)
