def test_save_load():
clf = BayesianRidge(compute_score=True)
X, y = get_traininig_data()
clf.fit(X, y)
y_hat = clf.predict(X)
model = MyCustomModel(clf)
conda = {
"name": "test",
"channels": ["defaults"],
"dependencies": [{"pip": ["scipy", "sklearn"]}]
}
model_save_path = os.path.join(dirname(abspath(__file__)), "AzureMLModel")
local_dependencies = [dirname(abspath(__file__))]
save_generic_model(model, path=model_save_path, conda=conda, local_dependencies=local_dependencies)
df = pd.DataFrame(data=X)
df.columns = df.columns.astype(str)
loaded_generic_model = load_generic_model(model_save_path)
result_df = loaded_generic_model.predict(df)
assert (result_df.to_numpy() == y_hat.reshape(-1, 1)).all()
dfd_path = os.path.join(dirname((abspath(__file__))), "dfd")
os.makedirs(dfd_path, exist_ok=True)
data_save_path = os.path.join(dfd_path, "data.dataset.parquet")
df.to_parquet(data_save_path, engine="pyarrow")
meta_path = os.path.join(dfd_path, "_meta.yaml")
with open(meta_path, "w") as fp:
fp.write("type: DataFrameDirectory\nextension: {}\nformat: Parquet\ndata: data.dataset.parquet")
def test_save_load():
init_params = {
"model_type": "densenet201",
"pretrained": True,
"memory_efficient": False,
"num_classes": 3
}
model = DenseNet(**init_params)
model_save_path = os.path.join(dirname(dirname(abspath(__file__))),
"AzureMLModel")
local_dependencies = [dirname(dirname(abspath(__file__)))]
# Also support list and csv_file
index_to_label = {0: "056.dog", 1: "060.duc", 2: "080.frog"}
save_pytorch_state_dict_model(model,
init_params=init_params,
path=model_save_path,
task_type=TaskType.MultiClassification,
label_map=index_to_label,
local_dependencies=local_dependencies)
loaded_generic_model = load_generic_model(model_save_path)
image_directory = os.path.join(dirname(dirname(abspath(__file__))),
"images")
image_iterator = mock_image_directory_iterator(image_directory)
predict_result = loaded_generic_model.predict(image_iterator)
print(f"predict_result =\n{predict_result}")
loaded_pytorch_model = loaded_generic_model.raw_model
assert isinstance(loaded_pytorch_model, torch.nn.Module)
def test_save_load():
init_params = {"inputSize": 1, "outputSize": 1}
model = LinearRegression(**init_params).to(device)
x_train, y_train = get_training_data()
train(model, x_train, y_train)
model_save_path = os.path.join(dirname(dirname(abspath(__file__))),
"AzureMLModel")
local_dependencies = [dirname(dirname(abspath(__file__)))]
save_pytorch_state_dict_model(model,
init_params=init_params,
path=model_save_path,
local_dependencies=local_dependencies)
loaded_generic_model = load_generic_model(model_save_path)
df = pd.DataFrame({"x": [[10.0], [11.0], [12.0]]})
predict_result = loaded_generic_model.predict(df)
assert predict_result.shape[0] == df.shape[0]
loaded_pytorch_model = loaded_generic_model.raw_model
assert isinstance(loaded_pytorch_model, torch.nn.Module)
dfd_path = os.path.join(dirname(dirname(abspath(__file__))), "dfd")
os.makedirs(dfd_path, exist_ok=True)
data_save_path = os.path.join(dfd_path, "data.dataset.parquet")
df.to_parquet(data_save_path, engine="pyarrow")
meta_path = os.path.join(dfd_path, "_meta.yaml")
with open(meta_path, "w") as fp:
fp.write(
"type: DataFrameDirectory\nextension: {}\nformat: Parquet\ndata: data.dataset.parquet"
)
def __init__(self, model_path, params={}):
logger.info(f"BuiltinScoreModule({model_path}, {params})")
append_score_column_to_output_value_str = params.get(
constants.APPEND_SCORE_COLUMNS_TO_OUTPUT_KEY, None)
self.append_score_column_to_output = isinstance(append_score_column_to_output_value_str, str) and\
append_score_column_to_output_value_str.lower() == "true"
logger.info(
f"self.append_score_column_to_output = {self.append_score_column_to_output}"
)
self.model = load_generic_model(model_path, install_dependencies=True)
logger.info("Generic model loaded")
def main():
model = LinearRegression(1, 1).to(device)
x_train, y_train = get_training_data()
train(model, x_train, y_train)
model_save_path = "./AzureMLModel"
save_pytorch_cloudpickle_model(model,
path=model_save_path,
local_dependencies=["."])
loaded_generic_model = load_generic_model(model_save_path)
df = pd.DataFrame({"x": [[10.0], [11.0], [12.0]]})
predict_result = loaded_generic_model.predict(df)
assert predict_result.shape[0] == df.shape[0]
loaded_pytorch_model = loaded_generic_model.raw_model
assert isinstance(loaded_pytorch_model, torch.nn.Module)
def main():
init_params = {"inputSize": 1, "outputSize": 1}
model = LinearRegression(**init_params).to(device)
x_train, y_train = get_training_data()
train(model, x_train, y_train)
model_save_path = "./AzureMLModel"
local_dependencies = ["."]
save_pytorch_state_dict_model(model,
init_params=init_params,
path=model_save_path,
local_dependencies=local_dependencies)
loaded_generic_model = load_generic_model(model_save_path)
df = pd.DataFrame({"x": [[10.0], [11.0], [12.0]]})
predict_result = loaded_generic_model.predict(df)
assert predict_result.shape[0] == df.shape[0]
loaded_pytorch_model = loaded_generic_model.raw_model
assert isinstance(loaded_pytorch_model, torch.nn.Module)
def main():
X, y = get_training_data()
clf = BayesianRidge(compute_score=True)
clf.fit(X, y)
y_hat = clf.predict(X)
model = MyCustomModel(clf)
model.conda = {
"name": "test",
"channels": ["defaults"],
"dependencies": [{
"pip": ["scipy", "sklearn"]
}]
}
save_generic_model(model, path="./AzureMLModel", local_dependencies=["."])
loaded_generic_model = load_generic_model(path="./AzureMLModel",
install_dependencies=False)
df = pd.DataFrame(data=X)
result_df = loaded_generic_model.predict(df)
print(f"result_df = {result_df}")
assert (result_df.to_numpy() == y_hat.reshape(-1, 1)).all()
from azureml.designer.model.io import load_generic_model
from azureml.studio.core.io.image_directory import ImageDirectory
import torch
import torch.nn as nn
from torchvision import transforms
# from .densenet import DenseNet
from .utils import logger
if __name__ == '__main__':
# Test inference
print("Testing inference.")
loaded_generic_model = load_generic_model(
path='/mnt/chjinche/projects/saved_model')
model = loaded_generic_model.raw_model
# # check predict before save
# state_dict = model.state_dict()
loader_dir = ImageDirectory.load('/mnt/chjinche/data/out_transform_test/')
# to_tensor_transform = transforms.Compose([transforms.ToTensor()])
# model_config = {
# 'model_type': 'densenet201',
# 'pretrained': False,
# 'memory_efficient': True,
# 'num_classes': 3
# }
# new_model = DenseNet(**model_config)
# new_model.load_state_dict(state_dict)
# if torch.cuda.is_available():
# new_model = new_model.cuda()
# if torch.cuda.device_count() > 1:
# new_model = torch.nn.DataParallel(new_model).cuda()
# new_model.eval()
from azureml.designer.model.io import load_generic_model
from azureml.studio.core.io.image_directory import ImageDirectory
import torch
import torch.nn as nn
from torchvision import transforms
# from .densenet import DenseNet
# from .utils import logger
if __name__ == '__main__':
# Test inference
print("Testing inference.")
loaded_generic_model = load_generic_model(
path='/mnt/chjinche/test_data/saved_model_classification')
# model = loaded_generic_model.raw_model
# # check predict before save
# state_dict = model.state_dict()
loader_dir = ImageDirectory.load('/mnt/chjinche/test_data/transform_test/')
# to_tensor_transform = transforms.Compose([transforms.ToTensor()])
# model_config = {
# 'model_type': 'densenet201',
# 'pretrained': False,
# 'memory_efficient': True,
# 'num_classes': 3
# }
# new_model = DenseNet(**model_config)
# new_model.load_state_dict(state_dict)
# if torch.cuda.is_available():
# new_model = new_model.cuda()
# if torch.cuda.device_count() > 1:
# new_model = torch.nn.DataParallel(new_model).cuda()
# new_model.eval()