REF: 重构5.5

torch_lib/core/__init__.py

@@ -1,8 +1,8 @@
 from typing import Any, Dict, Optional, Union, TypeVar
-from torch_lib.callback.dataset import DataParser, DataProvider
-from torch_lib.callback.metrics import M_SEQ
-from torch_lib.callback.run import R_SEQ
-from torch_lib.utils import MultiConst, get_device, type_cast, MethodChaining, InvocationDebug
+from torch_lib.callback.dataset import ConstantDataProvider, DataParser, DataProvider
+from torch_lib.callback.metrics import M_SEQ, MetricCallbackExecutor
+from torch_lib.callback.run import R_SEQ, RunCallbackExecutor
+from torch_lib.utils import NOTHING, MultiConst, get_device, type_cast, MethodChaining, InvocationDebug, is_nothing, logger
 from torch_lib.utils.type import NUMBER
 from torch_lib.context import Context
 from torch.utils.data import DataLoader
@@ -32,28 +32,36 @@ def train(
         self,
         train_dataset: DATASET,
         total_epochs: int = 1,
-        eval_dataset: DATASET = None,
-        run_callbacks: R_SEQ = None,
+        eval_dataset: DATASET = NOTHING,
+        run_callbacks: R_SEQ = NOTHING,
         log_option = None  # TODO: log system design
     ):
+        self._build_total_epochs(total_epochs)
+        self._build_run_callback_exec(run_callbacks)
+        self._build_dataset(train_dataset, 'train')
+        self._build_dataset(eval_dataset, 'eval')
         self.ctx.build.train(self.ctx)
 
     @InvocationDebug('ModelProxy.Predict')
     def predict(
         self,
         dataset: DATASET,
-        run_callbacks: R_SEQ = None,
+        run_callbacks: R_SEQ = NOTHING,
         log_option = None  # TODO: log system design
     ):
+        self._build_run_callback_exec(run_callbacks)
+        self._build_dataset(dataset, 'eval')
         self.ctx.build.predict(self.ctx)
 
     @InvocationDebug('ModelProxy.Eval')
     def eval(
         self,
         dataset: DATASET,
-        run_callbacks: R_SEQ = None,
+        run_callbacks: R_SEQ = NOTHING,
         log_option = None  # TODO: log system design
     ):
+        self._build_run_callback_exec(run_callbacks)
+        self._build_dataset(dataset, 'eval')
         self.ctx.build.eval(self.ctx)
 
     @InvocationDebug('ModelProxy.Summary')
@@ -73,7 +81,11 @@ def build(
         lr_decay_options: Optional[Dict] = None,
         data_parser: Optional[DataParser] = None
     ) -> MP:
-        pass
+        self._build_loss(loss_func)
+        self._build_metric_callback_exec(metric_callbacks)
+        self._build_data_parser(data_parser)
+        self._build_optimizer(optimizer, learning_rate, optimizer_options)
+        self._build_lr_decay(lr_decay, lr_decay_options)
 
     @InvocationDebug('ModelProxy.TrainBuilder')
     @MethodChaining
@@ -188,3 +200,52 @@ def build_eval(self) -> MP:
             # end callback
             handler.End()
         ])
+
+    @InvocationDebug('ModelProxy._build_loss')
+    def _build_loss(self, loss_func):
+        if loss_func is not None:
+            self.ctx.build.loss_func = loss_func if is_nothing(loss_func) is False else NOTHING
+
+    @InvocationDebug('ModelProxy._build_metric_callback_exec')
+    def _build_metric_callback_exec(self, metric_callbacks):
+        if metric_callbacks is not None:
+            self.ctx.build.metric_callback_exec = MetricCallbackExecutor(metric_callbacks) if is_nothing(metric_callbacks) is False else NOTHING
+
+    @InvocationDebug('ModelProxy._build_data_parser')
+    def _build_data_parser(self, data_parser):
+        if data_parser is not None:
+            self.ctx.build.data_parser = data_parser if is_nothing(data_parser) is False else NOTHING
+
+    @InvocationDebug('ModelProxy._build_run_callback_exec')
+    def _build_run_callback_exec(self, run_callbacks):
+        if run_callbacks is not None:
+            self.ctx.build.run_callback_exec = RunCallbackExecutor(run_callbacks) if is_nothing(run_callbacks) is False else NOTHING
+
+    @InvocationDebug('ModelProxy._build_optimizer')
+    def _build_optimizer(self, optimizer, learning_rate, optimizer_options):
+        if optimizer is not None:
+            if isinstance(optimizer, Optimizer):
+                self.ctx.build.optimizer = optimizer
+
+    @InvocationDebug('ModelProxy._build_lr_decay')
+    def _build_lr_decay(self, lr_decay, lr_decay_options):
+        pass
+
+    @InvocationDebug('ModelProxy._build_total_epochs')
+    def _build_total_epochs(self, total_epochs):
+        self.ctx.epoch.total = total_epochs if isinstance(total_epochs, int) else NOTHING
+
+    @InvocationDebug('ModelProxy._build_dataset')
+    def _build_dataset(self, dataset, mode: str):
+        if dataset is not None:
+            if is_nothing(dataset):
+                dataset = NOTHING
+            else:
+                dataset = dataset if isinstance(dataset, DataProvider) else ConstantDataProvider(dataset)
+
+            if mode == 'train':
+                self.ctx.build.train_provider = dataset
+            elif mode == 'eval':
+                self.ctx.build.eval_provider = dataset
+            else:
+                logger.warn('_build_dataset mode not supported.')

torch_lib/log/__init__.py

@@ -2,11 +2,13 @@
 
 
 color_dict = {
-    'r': 31,
-    'g': 32,
-    'y': 33,
-    'b': 34,
-    'w': 38
+    'r': 31,  # red
+    'g': 32,  # green
+    'y': 33,  # yellow
+    'b': 34,  # blue
+    'p': 35,  # purple
+    'c': 36,  # cyan
+    'w': 38  # white
 }
 
 info_prefix = '[TORCH_LIB INFO]'

torch_lib/utils/__init__.py

@@ -75,6 +75,9 @@ def __setattr__(self, *_):
     def __setitem__(self, *_):
         pass
 
+    def __len__(self):
+        return 0
+
     def __iter__(self):
         return self