def init_trainer(self, model, X =None, y = None):
f = tempfile.NamedTemporaryFile(delete=False)
f.write(model)
f.close()
self.X=X
self.y =y
self.params = pb2.SolverParameter()
self.params.net = f.name
set(self.params.test_iter, 10)
self.params.test_interval = 10
self.params.base_lr = 0.01
self.params.momentum = 0.9
self.params.weight_decay = 0.0005
self.params.lr_policy = 'inv'
self.params.gamma = 0.0001
self.params.power = 0.75
self.params.display = 10
self.params.max_iter = 100
self.params.snapshot_after_train = False
f2 = tempfile.NamedTemporaryFile(delete=False)
f2.write(self.params.__str__())
f2.close()
self.solver = caffe.SGDSolver(f2.name)
f2.delete = True
if (X is not None) and (y is not None):
self.create_batches()
# self.solver.set_train_data = types.MethodType(set_train_data, self.solver)
# self.solver.set_test_data = types.MethodType(set_test_data, self.solver)
return self.solver
def init_trainer(self, model, X=None, y=None):
f = tempfile.NamedTemporaryFile(delete=False)
f.write(model)
f.close()
self.X = X
self.y = y
self.params = pb2.SolverParameter()
self.params.net = f.name
set(self.params.test_iter, 10)
self.params.test_interval = 10
self.params.base_lr = 0.01
self.params.momentum = 0.9
self.params.weight_decay = 0.0005
self.params.lr_policy = 'inv'
self.params.gamma = 0.0001
self.params.power = 0.75
self.params.display = 10
self.params.max_iter = 100
self.params.snapshot_after_train = False
f2 = tempfile.NamedTemporaryFile(delete=False)
f2.write(self.params.__str__())
f2.close()
self.solver = caffe.SGDSolver(f2.name)
f2.delete = True
if (X is not None) and (y is not None):
self.create_batches()
# self.solver.set_train_data = types.MethodType(set_train_data, self.solver)
# self.solver.set_test_data = types.MethodType(set_test_data, self.solver)
return self.solver
def norm(self, val):
self.params.norm = set(self.params.norm, val)
def mean(self, val):
self._filler.mean = set(self._filler.mean, val)
def sparse(self, val):
self._filler.sparse = set(self._filler.sparse, val)
def type(self, val):
self._filler.type = set(self._filler.type, val)
def min(self, val):
self._filler.min = set(self._filler.min, val)
def channels(self, val):
set(self.params.channels, val)
def weight_filler(self, val):
set(self.inner_product_param.weight_filler, val)
def width(self, val):
set(self.params.width, val)
def std(self, val):
self._filler.std = set(self._filler.std, val)
def channels(self, val):
set(self.params.channels, val)
def height(self, val):
set(self.params.height, val)
def num(self, val):
set(self.params.num, val)
def width(self, val):
set(self.params.width, val)
def height(self, val):
set(self.params.height, val)
def margin(self, val):
self.params.margin = set(self.params.margin, val)
def num_output(self, val):
self.inner_product_param.num_output = set(self._layer.inner_product_param.num_output, val)
def mean(self, val):
self._filler.mean = set(self._filler.mean, val)
def num(self, val):
set(self.params.num, val)
def sparse(self, val):
self._filler.sparse = set(self._filler.sparse, val)
def weight_filler(self, val):
set(self.inner_product_param.weight_filler, val)
def bias_term(self, val):
set(self.inner_product_param.bias_term, val)
def bias_filler(self, val):
set(self.inner_product_param.bias_filler, val)
def bias_filler(self, val):
set(self.inner_product_param.bias_filler, val)
def num_output(self, val):
self.inner_product_param.num_output = set(
self._layer.inner_product_param.num_output, val)
def value(self, val):
self._filler.value = set(self._filler.value, val)
def bias_term(self, val):
set(self.inner_product_param.bias_term, val)
def max(self, val):
self._filler.max = set(self._filler.max, val)
def type(self, val):
self._filler.type = set(self._filler.type, val)
def std(self, val):
self._filler.std = set(self._filler.std, val)
def value(self, val):
self._filler.value = set(self._filler.value, val)
def min(self, val):
self._filler.min = set(self._filler.min, val)
def max(self, val):
self._filler.max = set(self._filler.max, val)
def margin(self, val):
self.params.margin = set(self.params.margin, val)
def norm(self, val):
self.params.norm = set(self.params.norm, val)
