def test_native_module_library_with_in_out(self):
@add_module()
class Test(object):
def __init__(self):
pass
test = Module.create("Test", **{"in": "foo", "out": "bar"})
self.assertIsInstance(test, Test)
def test_native_module_library(self):
@add_module()
class Test(object):
def __init__(self):
pass
test = Module.create("Test")
self.assertIsInstance(test, Test)
def test_native_module_library_with_params(self):
@add_module()
class TestWithParams(object):
def __init__(self, my_param):
self.my_param = my_param
test = Module.create("TestWithParams", my_param="foobar")
self.assertIsInstance(test, TestWithParams)
self.assertEqual(test.my_param, "foobar")
def __init__(self, training_name, data_path, training_results_path):
super().__init__(training_name, data_path, training_results_path)
# Config of the data
self.data_dataset = FashionMNISTDataset
# Config of the model
self.model_model = lambda config: Module.create_from_file(
"deeptech/examples/mnist_model.json",
"MNISTModel",
num_classes=10,
logits=True)
# Config for training
self.training_loss = SparseCrossEntropyLossFromLogits
self.training_optimizer = smart_optimizer(SGD)
self.training_trainer = SupervisedTrainer
self.training_epochs = 10
self.training_batch_size = 32
def __init__(self, training_name, data_path, training_results_path):
super().__init__(training_name, data_path, training_results_path)
# Config of the data
self.data_dataset = lambda split: COCODataset(
split, COCODataset.InputType, FasterRCNNOutput)
self.data_version = 2014
self.data_image_size = (800, 600)
# Config of the model
self.model_categories = [] # Fill from dataset.
self.model_log_delta_preds = False
self.model_model = lambda: Module.create(
"FasterRCNN",
num_classes=len(self.model_categories),
log_deltas=self.model_log_delta_preds)
# Config for training
self.training_loss = self.create_loss
self.training_optimizer = smart_optimizer(SGD, momentum=0.9)
self.training_trainer = SupervisedTrainer
self.training_epochs = 10
self.training_batch_size = 1
self.training_initial_lr = 0.001
def setUp(self) -> None:
add_lib_from_json("tests/json_nets/vgg16_bn.jsonc")
self.module = Module.create("VGG16_bn", logits=True)
self.input_data = torch.from_numpy(
np.zeros((1, 128, 128, 3), dtype=np.float32))
self.result = self.module(self.input_data)