model: Use auto args and config

Author: aghinsa

dffml/base.py

@@ -3,6 +3,7 @@
 they follow a similar API for instantiation and usage.
 """
 import abc
+import copy
 import inspect
 import argparse
 import contextlib
@@ -22,7 +23,7 @@ def get_args(t):
 
 
 from .util.cli.arg import Arg
-from .util.data import traverse_config_set, traverse_config_get
+from .util.data import traverse_config_set, traverse_config_get, type_lookup
 
 from .util.entrypoint import Entrypoint
 
@@ -114,6 +115,8 @@ def mkarg(field):
     # HACK For detecting dataclasses._MISSING_TYPE
     if "dataclasses._MISSING_TYPE" not in repr(field.default):
         arg["default"] = field.default
+    if "dataclasses._MISSING_TYPE" not in repr(field.default_factory):
+        arg["default"] = field.default_factory()
     if field.type == bool:
         arg["action"] = "store_true"
     elif inspect.isclass(field.type):
@@ -140,18 +143,20 @@ def convert_value(arg, value):
     if value is None:
         # Return default if not found and available
         if "default" in arg:
-            return arg["default"]
+            return copy.deepcopy(arg["default"])
         raise MissingConfig
 
-    # TODO This is a oversimplification of argparse's nargs
     if not "nargs" in arg:
         value = value[0]
     if "type" in arg:
+        type_cls = arg["type"]
+        if type_cls == Type:
+            type_cls = type_lookup
         # TODO This is a oversimplification of argparse's nargs
         if "nargs" in arg:
-            value = list(map(arg["type"], value))
+            value = list(map(type_cls, value))
         else:
-            value = arg["type"](value)
+            value = type_cls(value)
     if "action" in arg:
         if isinstance(arg["action"], str):
             # HACK This accesses _pop_action_class from ArgumentParser
@@ -205,7 +210,7 @@ def config(cls):
     """
     Decorator to create a dataclass
     """
-    datacls = dataclasses.dataclass(eq=True, init=True, frozen=True)(cls)
+    datacls = dataclasses.dataclass(eq=True, init=True)(cls)
     datacls._fromdict = classmethod(_fromdict)
     datacls._replace = lambda self, *args, **kwargs: dataclasses.replace(
         self, *args, **kwargs

dffml/model/model.py

@@ -10,6 +10,7 @@
 from typing import AsyncIterator, Tuple, Any, List, Optional, NamedTuple, Dict
 
 from ..base import (
+    config,
     BaseConfig,
     BaseDataFlowFacilitatorObjectContext,
     BaseDataFlowFacilitatorObject,
@@ -27,7 +28,8 @@ class ModelNotTrained(Exception):
     pass
 
 
-class ModelConfig(BaseConfig, NamedTuple):
+@config
+class ModelConfig:
     directory: str
     features: Features
 
@@ -72,43 +74,12 @@ class Model(BaseDataFlowFacilitatorObject):
     various machine learning frameworks or concepts.
     """
 
+    CONFIG = ModelConfig
+
     def __call__(self) -> ModelContext:
         # If the config object for this model contains the directory property
         # then create it if it does not exist
         directory = getattr(self.config, "directory", None)
         if directory is not None and not os.path.isdir(directory):
             os.makedirs(directory)
         return self.CONTEXT(self)
-
-    @classmethod
-    def args(cls, args, *above) -> Dict[str, Arg]:
-        cls.config_set(
-            args,
-            above,
-            "directory",
-            Arg(
-                default=os.path.join(
-                    os.path.expanduser("~"), ".cache", "dffml"
-                )
-            ),
-        )
-        cls.config_set(
-            args,
-            above,
-            "features",
-            Arg(
-                nargs="+",
-                required=True,
-                type=Feature.load,
-                action=list_action(Features),
-            ),
-        )
-
-        return args
-
-    @classmethod
-    def config(cls, config, *above) -> BaseConfig:
-        return ModelConfig(
-            directory=cls.config_get(config, above, "directory"),
-            features=cls.config_get(config, above, "features"),
-        )

tests/test_cli.py

@@ -33,7 +33,7 @@
 from dffml.util.entrypoint import entry_point
 from dffml.util.asynctestcase import AsyncTestCase
 from dffml.util.cli.cmds import ModelCMD
-
+from dffml.base import config
 from dffml.cli import Merge, Dataflow, Train, Accuracy, Predict, List
 
 from .test_df import OPERATIONS, OPIMPS
@@ -94,6 +94,14 @@ def mktempfile(self):
         return self._stack.enter_context(non_existant_tempfile())
 
 
+@config
+class FakeConfig:
+    features: Features
+    directory: str = os.path.join(
+        os.path.expanduser("~"), ".cache", "dffml", "test_cli", "fake"
+    )
+
+
 class FakeFeature(Feature):
 
     NAME: str = "fake"
@@ -134,6 +142,7 @@ async def predict(self, repos: AsyncIterator[Repo]) -> AsyncIterator[Repo]:
 class FakeModel(Model):
 
     CONTEXT = FakeModelContext
+    CONFIG = FakeConfig
 
 
 def feature_load(loading=None):