def test_value_error_on_unknown_model_getter():
application = Application()
application.add(InitCommand())
module_name = "foo"
module_file_path = module_name + ".py"
save_path = "foo-config/model.py"
content = """
import torch
def init(**args):
return torch.nn.Conv2d(10, 10, 10)
"""
with open(module_file_path, "a") as the_file:
the_file.write(content)
command = application.find("init")
command_tester = CommandTester(command)
with pytest.raises(ValueError,
match=r".*{fn_name}.*".format(fn_name="get_model")):
command_tester.execute(module_name + " --save=" + save_path)
os.remove(module_file_path)
shutil.rmtree(os.path.dirname(save_path))
def test_value_error_on_unknown_module():
application = Application()
application.add(InitCommand())
command = application.find("init")
command_tester = CommandTester(command)
unknown_module_name = "some_module"
with pytest.raises(ValueError, match=r".*%s.*" % unknown_module_name):
command_tester.execute(unknown_module_name)
def test_init_error_custom_config_with_list():
application = Application()
application.add(InitCommand())
invalid_config = '[{"conf_key": "value"}]'
module_name = "foo"
command = application.find("init")
command_tester = CommandTester(command)
with pytest.raises(ValueError):
command_tester.execute(module_name + " --config '" + invalid_config +
"'")
def test_value_error_on_invalid_config_with_decoding():
application = Application()
application.add(InitCommand())
invalid_config = '{"unclosed_key: 15}'
module_name = "foo"
command = application.find("init")
command_tester = CommandTester(command)
with pytest.raises(ValueError,
match=r".*{reason}.*".format(reason="invalid")):
command_tester.execute(module_name + " --config '" + invalid_config +
"'")
def test_value_error_on_invalid_config_with_single_quotes():
application = Application()
application.add(InitCommand())
invalid_config = "{'output_size': 10}"
module_name = "foo"
command = application.find("init")
command_tester = CommandTester(command)
with pytest.raises(ValueError,
match=r".*{reason}.*".format(reason="double quotes")):
command_tester.execute(module_name + ' --config "' + invalid_config +
'"')
def test_success_load_custom_user_config():
application = Application()
application.add(InitCommand())
custom_config = {
"batch_size": 512,
"device": "cuda:0",
"dataset": "torch.blah",
"learning_rate": 0.2,
}
# set up file path variables
module_name = "tmp_module1_init"
module_file_path = module_name + ".py"
save_path = "tmp-test-save/model.py"
# clean up possible existing files
if os.path.exists(module_file_path):
os.remove(module_file_path)
if os.path.exists(save_path):
shutil.rmtree(os.path.dirname(save_path))
# write model to file path from which we want to import from
content = """
import torch
def get_model(**args):
return torch.nn.Conv2d(10, 10, 10)
"""
with open(module_file_path, "a") as the_file:
the_file.write(content)
command = application.find("init")
command_tester = CommandTester(command)
# Act
command_tester.execute(module_name + " --config {json_config}".format(
json_config=shlex.quote(json.dumps(custom_config))) + " --save=" +
save_path)
assert os.path.exists(save_path)
# Load metadata from saved path
# TODO schema validation
# reader = pkmd.MetadataReader()
# reader.read(os.path.join(os.path.dirname(save_path), 'config.json'))
# Cleanup
os.remove(module_file_path)
shutil.rmtree(os.path.dirname(save_path))
def test_success_init_simple_model():
# Arrange
# set up application with command
application = Application()
application.add(InitCommand())
# set up file path variables
module_name = "bar"
module_file_path = module_name + ".py"
save_path = "bar-config/model.py"
# clean up possible existing files
if os.path.exists(module_file_path):
os.remove(module_file_path)
if os.path.exists(save_path):
shutil.rmtree(os.path.dirname(save_path))
# write model to file path from which we want to import from
content = """
import torch
def get_model(**args):
return torch.nn.Conv2d(10, 10, 10)
"""
with open(module_file_path, "a") as the_file:
the_file.write(content)
# load command and build tester object to act on
command = application.find("init")
command_tester = CommandTester(command)
# Act
command_tester.execute(module_name + " --save=" + save_path)
# Cleanup
os.remove(module_file_path)
shutil.rmtree(os.path.dirname(save_path))
def test_success_init_simple_model():
# Arrange
# set up application with command
application = Application()
application.add(InitCommand())
application.add(TrainCommand())
# set up file path variables
module_name = "module_name_for_training"
module_file_path = module_name + ".py"
save_path = "tmp-train-save-path/model.py"
dataset_module = "tmp_training_dataset"
dataset_module_file_path = dataset_module + ".py"
user_config = {"num_epochs": 3} # Use only few epochs for test
# clean up possible existing files
if os.path.exists(module_file_path):
os.remove(module_file_path)
if os.path.exists(dataset_module_file_path):
os.remove(dataset_module_file_path)
if os.path.exists(save_path):
shutil.rmtree(os.path.dirname(save_path))
# write model to file path from which we want to import from
content_model = """
import torch.nn as nn
import torch.nn.functional as F
class MyModel(nn.Module):
def __init__(self, width: int, height: int):
super(MyModel, self).__init__()
c = 6 # intermediate_channels
self.conv = nn.Conv2d(in_channels=3, out_channels=c, kernel_size=5)
self.fc = nn.Linear((width-5+1)*(height-5+1)*c, 10)
def forward(self, x):
out = F.relu(self.conv(x))
out = out.view(out.size(0), -1)
return F.relu(self.fc(out))
def get_model(**args):
return MyModel(width=32, height=32)
"""
with open(module_file_path, "a") as model_handle:
model_handle.write(content_model)
# write model to file path from which we want to import from
content_dataset = """
import torch
import numpy as np
from torch.utils import data
class MyDataset(data.Dataset):
def __len__(self):
return 200
def __getitem__(self, index):
return torch.rand((3, 32, 32)), np.random.randint(0, 10)
def get_dataset(**args):
return MyDataset()
"""
with open(dataset_module_file_path, "a") as dataset_handle:
dataset_handle.write(content_dataset)
command_init = application.find("init")
init_tester = CommandTester(command_init)
init_tester.execute(module_name + " --save=" + save_path)
command_train = application.find("train")
train_tester = CommandTester(command_train)
train_tester.execute(
save_path
+ " "
+ dataset_module
+ ".get_dataset"
+ " --config {json_config}".format(
json_config=shlex.quote(json.dumps(user_config))
)
)
# Cleanup
os.remove(module_file_path)
os.remove(dataset_module_file_path)
shutil.rmtree(os.path.dirname(save_path))