minor change

myresources/crocodile/deeplearning.py

@@ -3,20 +3,21 @@
 dl
 """
 
+import numpy as np
+import numpy.typing as npt
+import pandas as pd
+from tqdm import tqdm
+
 from crocodile.matplotlib_management import ImShow
 from crocodile.core import List as L, Struct as S, Base
 from crocodile.file_management import P, Save, PLike, Read
 from crocodile.meta import Experimental
 
-import numpy as np
-import numpy.typing as npt
-import pandas as pd
-from abc import ABC
-from typing import TypeVar, Type, Any, Optional, Union, Callable, Literal, TypeAlias
 import enum
-from tqdm import tqdm
 import copy
+from abc import ABC
 from dataclasses import dataclass, field
+from typing import TypeVar, Type, Any, Optional, Union, Callable, Literal, TypeAlias, Protocol
 
 
 @dataclass
@@ -69,10 +70,33 @@ class Device(enum.Enum):
 SubclassedBaseModel = TypeVar("SubclassedBaseModel", bound='BaseModel')
 
 
+HPARAMS_SUBPATH: str = 'metadata/hyperparameters' # location within model directory where this will be saved.
 PACKAGE: TypeAlias = Literal['tensorflow', 'torch']
 PRECISON = Literal['float64', 'float32', 'float16']
 
 
+class HyperParams(Protocol):
+ # ================== General ==============================
+ name: str
+ root: P
+ pkg_name: PACKAGE
+
+ # ===================== Data ==============================
+ seed: int
+ shuffle: bool
+ precision: PRECISON
+
+ # ===================== Training ==========================
+ test_split: float
+ learning_rate: float
+ batch_size: int
+ epochs: int
+
+
+def get_hp_save_dir(hp: HyperParams):
+ return (P(hp.root) / hp.name).create()
+
+
 @dataclass
 class HParams:
  # ===================== Data ==============================
@@ -96,7 +120,8 @@ class HParams:
  subpath: str = 'metadata/hyperparameters' # location within model directory where this will be saved.
 
  def save(self):
- subpath = self.subpath
+ # subpath = self.subpath
+ subpath = HPARAMS_SUBPATH
  save_dir = self.save_dir
  self_repr = str(self)
 
@@ -137,7 +162,7 @@ class DataReader:
  subpath = P("metadata/data_reader")
  """This class holds the dataset for training and testing.
  """
- def __init__(self, hp: SubclassedHParams, # type: ignore
+ def __init__(self, hp: HyperParams, # type: ignore
  specs: Optional[Specs] = None,
  split: Optional[dict[str, Any]] = None) -> None:
  # split could be Union[None, 'npt.NDArray[np.float64]', 'pd.DataFrame', 'pd.Series', 'list[Any]', Tf.RaggedTensor etc.
@@ -149,13 +174,15 @@ def __init__(self, hp: SubclassedHParams, # type: ignore
  # self.df_handler = df_handler
  def save(self, path: Optional[str] = None, **kwargs: Any) -> None:
  _ = kwargs
- base = (P(path) if path is not None else self.hp.save_dir).joinpath(self.subpath).create()
+ base = (P(path) if path is not None else get_hp_save_dir(self.hp)).joinpath(self.subpath).create()
  try: data: dict[str, Any] = self.__getstate__()
  except AttributeError: data = self.__dict__
  Save.pickle(path=base / "data_reader.DataReader.dat.pkl", obj=data)
  Save.pickle(path=base / "data_reader.DataReader.pkl", obj=self)
  @classmethod
- def from_saved_data(cls, path: Union[str, P], hp: SubclassedHParams, # type: ignore
+ def from_saved_data(cls, path: Union[str, P],
+ # hp: SubclassedHParams, # type: ignore
+ hp: HyperParams,
  **kwargs: Any):
  path = P(path) / cls.subpath / "data_reader.DataReader.dat.pkl"
  data: dict[str, Any] = Read.pickle(path)
@@ -400,7 +427,7 @@ def fit(self, viz: bool = True, weight_name: Optional[str] = None,
  self.history.append(copy.deepcopy(hist.history)) # it is paramount to copy, cause source can change.
  if viz:
  artist = self.plot_loss()
- artist.fig.savefig(str(self.hp.save_dir.joinpath(f"metadata/training/loss_curve.png").append(index=True).create(parents_only=True)))
+ artist.fig.savefig(str(get_hp_save_dir(self.hp).joinpath(f"metadata/training/loss_curve.png").append(index=True).create(parents_only=True)))
  return self
 
  def switch_to_sgd(self, epochs: int = 10):
@@ -473,7 +500,7 @@ def load_weights(self, directory: PLike) -> None:
  self.model.load_weights(path) # .expect_partial()
  def summary(self):
  from contextlib import redirect_stdout
- path = self.hp.save_dir.joinpath("metadata/model/model_summary.txt").create(parents_only=True)
+ path = get_hp_save_dir(self.hp).joinpath("metadata/model/model_summary.txt").create(parents_only=True)
  with open(str(path), 'w', encoding='utf-8') as f:
  with redirect_stdout(f): self.model.summary()
  return self.model.summary()
@@ -568,10 +595,10 @@ def save_class(self, weights_only: bool = True, version: str = 'v0', strict: boo
  """
  self.hp.save() # goes into the meta path.
  self.data.save() # goes into the meta path.
- Save.pickle(obj=self.history, path=self.hp.save_dir / 'metadata/training/history.pkl', verbose=True, desc="Training History") # goes into the meta path.
- try: Experimental.generate_readme(self.hp.save_dir, obj=self.__class__, desc=desc)
+ Save.pickle(obj=self.history, path=get_hp_save_dir(self.hp) / 'metadata/training/history.pkl', verbose=True, desc="Training History") # goes into the meta path.
+ try: Experimental.generate_readme(get_hp_save_dir(self.hp), obj=self.__class__, desc=desc)
  except Exception as ex: print(ex) # often fails because model is defined in main during experiments.
- save_dir = self.hp.save_dir.joinpath(f'{"weights" if weights_only else "model"}_save_{version}')
+ save_dir = get_hp_save_dir(self.hp).joinpath(f'{"weights" if weights_only else "model"}_save_{version}')
  if weights_only: self.save_weights(save_dir.create())
  else:
  self.save_model(save_dir)
@@ -598,12 +625,14 @@ def save_class(self, weights_only: bool = True, version: str = 'v0', strict: boo
  'module_path_rh': module_path_rh,
  'cwd_rh': P.cwd().collapseuser().as_posix(),
  }
- Save.json(obj=specs, path=self.hp.save_dir.joinpath('metadata/code_specs.json').str, indent=4)
- print(f'SAVED Model Class @ {self.hp.save_dir.as_uri()}')
- return self.hp.save_dir
+ Save.json(obj=specs, path=get_hp_save_dir(self.hp).joinpath('metadata/code_specs.json').str, indent=4)
+ print(f'SAVED Model Class @ {get_hp_save_dir(self.hp).as_uri()}')
+ return get_hp_save_dir(self.hp)
 
  @classmethod
- def from_class_weights(cls, path: PLike, hparam_class: Optional[Type[SubclassedHParams]] = None, data_class: Optional[Type[SubclassedDataReader]] = None,
+ def from_class_weights(cls, path: PLike,
+ hparam_class: Optional[Type[SubclassedHParams]] = None,
+ data_class: Optional[Type[SubclassedDataReader]] = None,
  device_name: Optional[Device] = None, verbose: bool = True):
  path = P(path)
  if hparam_class is not None:
@@ -676,7 +705,7 @@ def from_path(path_model: PLike, **kwargs: Any) -> 'SubclassedBaseModel': # typ
 
  def plot_model(self, dpi: int = 150, strict: bool = False, **kwargs: Any): # alternative viz via tf2onnx then Netron.
  import keras
- path = self.hp.save_dir.joinpath("metadata/model/model_plot.png")
+ path = get_hp_save_dir(self.hp).joinpath("metadata/model/model_plot.png")
  try:
  keras.utils.plot_model(self.model, to_file=str(path), show_shapes=True, show_layer_names=True, show_layer_activations=True, show_dtype=True, expand_nested=True, dpi=dpi, **kwargs)
  print(f"Successfully plotted the model @ {path.as_uri()}")

myresources/crocodile/deeplearning_template.py

@@ -22,7 +22,7 @@ class HParams(dl.HParams):
  subpath: str = 'metadata/hyperparameters' # location within model directory where this will be saved.
  name: str = field(default_factory=lambda: "model-" + randstr(noun=True))
  root: P = P.tmp(folder="tmp_models")
- pkg_name = 'tensorflow'
+ pkg_name: dl.PACKAGE = 'tensorflow'
  # device_name: Device=Device.gpu0
  # ===================== Data ==============================
  seed: int = 234