deep00 0.0.0__tar.gz

deep00-0.0.0/PKG-INFO

@@ -0,0 +1,26 @@
+Metadata-Version: 2.1
+Name: deep00
+Version: 0.0.0
+Summary: Ultra simple deep learning tool
+Home-page: https://github.co.jp/
+Author: bib_inf
+Author-email: contact.bibinf@gmail.com
+License: CC0 v1.0
+Classifier: Programming Language :: Python :: 3
+Classifier: Topic :: Software Development :: Libraries
+Classifier: License :: CC0 1.0 Universal (CC0 1.0) Public Domain Dedication
+Description-Content-Type: text/markdown
+
+English description follows Japanese.
+
+---
+
+### 概要
+超簡単に深層学習が使えるツールです。
+
+説明は執筆中です
+
+---
+
+### Overview
+description is under construction.

deep00-0.0.0/README.md

@@ -0,0 +1,13 @@
+English description follows Japanese.
+
+---
+
+### 概要
+超簡単に深層学習が使えるツールです。
+
+説明は執筆中です
+
+---
+
+### Overview
+description is under construction.

deep00-0.0.0/deep00/test.py

@@ -0,0 +1,307 @@
+"""
+deep00 - シンプルな深層学習回帰ツール (PyTorch)
+
+Usage:
+	model = deep00("reg", neurons=[7, 7], activation="relu")
+	model.fit(train_x, train_y)
+	pred_y = model(test_x)
+"""
+
+import numpy as np
+import torch
+import torch.nn as nn
+import torch.optim as optim
+from torch.utils.data import TensorDataset, DataLoader
+
+
+class _RegressionNet(nn.Module):
+	"""全結合ニューラルネットワーク（回帰用）"""
+
+	_ACTIVATIONS = {
+		"relu": nn.ReLU,
+		"sigmoid": nn.Sigmoid,
+		"tanh": nn.Tanh,
+		"leaky_relu": nn.LeakyReLU,
+		"elu": nn.ELU,
+		"selu": nn.SELU,
+		"gelu": nn.GELU,
+	}
+
+	def __init__(self, in_dim: int, out_dim: int, neurons: list[int], activation: str):
+		super().__init__()
+		act_cls = self._ACTIVATIONS.get(activation)
+		if act_cls is None:
+			raise ValueError(
+				f"未対応の活性化関数: '{activation}'  "
+				f"対応一覧: {list(self._ACTIVATIONS.keys())}"
+			)
+
+		layers: list[nn.Module] = []
+		prev = in_dim
+		for n in neurons:
+			layers.append(nn.Linear(prev, n))
+			layers.append(act_cls())
+			prev = n
+		layers.append(nn.Linear(prev, out_dim))  # 出力層（活性化なし）
+
+		self.net = nn.Sequential(*layers)
+
+	def forward(self, x: torch.Tensor) -> torch.Tensor:
+		return self.net(x)
+
+
+class deep00:
+	"""
+	シンプルな深層学習回帰モデル。
+
+	Parameters
+	----------
+	task : str
+		タスク種別。現在は "reg"（回帰）のみ対応。
+	neurons : list[int]
+		中間層のニューロン数一覧。例: [7, 7]
+	activation : str
+		活性化関数名。デフォルトは "relu"。
+		対応: relu, sigmoid, tanh, leaky_relu, elu, selu, gelu
+	epochs : int
+		エポック数。デフォルト 100。
+	batch_size : int
+		ミニバッチサイズ。デフォルト 32。
+	lr : float
+		学習率。デフォルト 1e-3。
+	optimizer : str
+		最適化手法。"adam" または "sgd"。デフォルト "adam"。
+	weight_decay : float
+		L2正則化の強さ。デフォルト 0。
+	early_stopping : int or None
+		検証損失が改善しないエポック数で学習を打ち切る。
+		None の場合は早期終了なし。デフォルト None。
+	val_ratio : float
+		早期終了用の検証データ割合（0〜1）。デフォルト 0.1。
+	verbose : bool
+		学習経過を表示するか。デフォルト True。
+	seed : int or None
+		乱数シード。デフォルト None。
+	device : str or None
+		"cpu" / "cuda" / None（自動選択）。
+	"""
+
+	def __init__(
+		self,
+		task: str = "reg",
+		*,
+		neurons: list[int] | None = None,
+		activation: str = "relu",
+		epochs: int = 100,
+		batch_size: int = 32,
+		lr: float = 1e-3,
+		optimizer: str = "adam",
+		weight_decay: float = 0.0,
+		early_stopping: int | None = None,
+		val_ratio: float = 0.1,
+		verbose: bool = True,
+		seed: int | None = None,
+		device: str | None = None,
+	):
+		if task != "reg":
+			raise ValueError(f"未対応のタスク: '{task}'（現在は 'reg' のみ）")
+
+		self.task = task
+		self.neurons = neurons if neurons is not None else [32, 32]
+		self.activation = activation
+		self.epochs = epochs
+		self.batch_size = batch_size
+		self.lr = lr
+		self.optimizer_name = optimizer
+		self.weight_decay = weight_decay
+		self.early_stopping = early_stopping
+		self.val_ratio = val_ratio
+		self.verbose = verbose
+		self.seed = seed
+
+		if device is None:
+			self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+		else:
+			self.device = torch.device(device)
+
+		self._model: _RegressionNet | None = None
+		self._train_loss_history: list[float] = []
+		self._val_loss_history: list[float] = []
+
+	# ------------------------------------------------------------------
+	# 内部ユーティリティ
+	# ------------------------------------------------------------------
+	def _to_tensor(self, arr: np.ndarray) -> torch.Tensor:
+		return torch.as_tensor(arr, dtype=torch.float32, device=self.device)
+
+	def _build_model(self, in_dim: int, out_dim: int) -> _RegressionNet:
+		model = _RegressionNet(in_dim, out_dim, self.neurons, self.activation)
+		return model.to(self.device)
+
+	def _make_optimizer(self, model: _RegressionNet) -> optim.Optimizer:
+		if self.optimizer_name == "adam":
+			return optim.Adam(model.parameters(), lr=self.lr, weight_decay=self.weight_decay)
+		elif self.optimizer_name == "sgd":
+			return optim.SGD(model.parameters(), lr=self.lr, weight_decay=self.weight_decay)
+		else:
+			raise ValueError(
+				f"未対応のoptimizer: '{self.optimizer_name}'  対応: adam, sgd"
+			)
+
+	# ------------------------------------------------------------------
+	# fit
+	# ------------------------------------------------------------------
+	def fit(self, train_x: np.ndarray, train_y: np.ndarray) -> "deep00":
+		"""
+		モデルを学習する。
+
+		Parameters
+		----------
+		train_x : np.ndarray, shape (N, in_dim)
+		train_y : np.ndarray, shape (N, out_dim)
+
+		Returns
+		-------
+		self
+		"""
+		if self.seed is not None:
+			torch.manual_seed(self.seed)
+			np.random.seed(self.seed)
+
+		x = self._to_tensor(train_x)
+		y = self._to_tensor(train_y)
+		in_dim = x.shape[1]
+		out_dim = y.shape[1]
+
+		self._model = self._build_model(in_dim, out_dim)
+		opt = self._make_optimizer(self._model)
+		criterion = nn.MSELoss()
+
+		# --- 検証データ分割（早期終了が有効な場合）---
+		use_val = self.early_stopping is not None and self.val_ratio > 0
+		if use_val:
+			n = x.shape[0]
+			idx = np.random.permutation(n)
+			val_n = max(1, int(n * self.val_ratio))
+			val_idx, trn_idx = idx[:val_n], idx[val_n:]
+			val_x, val_y = x[val_idx], y[val_idx]
+			x, y = x[trn_idx], y[trn_idx]
+
+		dataset = TensorDataset(x, y)
+		loader = DataLoader(dataset, batch_size=self.batch_size, shuffle=True)
+
+		best_val_loss = float("inf")
+		patience_counter = 0
+		best_state = None
+		self._train_loss_history = []
+		self._val_loss_history = []
+
+		for epoch in range(1, self.epochs + 1):
+			# --- 学習 ---
+			self._model.train()
+			epoch_loss = 0.0
+			count = 0
+			for bx, by in loader:
+				opt.zero_grad()
+				pred = self._model(bx)
+				loss = criterion(pred, by)
+				loss.backward()
+				opt.step()
+				epoch_loss += loss.item() * bx.shape[0]
+				count += bx.shape[0]
+			epoch_loss /= count
+			self._train_loss_history.append(epoch_loss)
+
+			# --- 検証 ---
+			if use_val:
+				self._model.eval()
+				with torch.no_grad():
+					val_pred = self._model(val_x)
+					val_loss = criterion(val_pred, val_y).item()
+				self._val_loss_history.append(val_loss)
+
+				if val_loss < best_val_loss:
+					best_val_loss = val_loss
+					patience_counter = 0
+					best_state = {k: v.clone() for k, v in self._model.state_dict().items()}
+				else:
+					patience_counter += 1
+			else:
+				val_loss = None
+
+			# --- ログ ---
+			if self.verbose and (epoch % max(1, self.epochs // 10) == 0 or epoch == 1):
+				msg = f"Epoch {epoch:>5d}/{self.epochs}  train_loss={epoch_loss:.6f}"
+				if val_loss is not None:
+					msg += f"  val_loss={val_loss:.6f}"
+				print(msg)
+
+			# --- 早期終了 ---
+			if use_val and patience_counter >= self.early_stopping:
+				if self.verbose:
+					print(f"Early stopping at epoch {epoch} (patience={self.early_stopping})")
+				break
+
+		# ベストモデル復元
+		if best_state is not None:
+			self._model.load_state_dict(best_state)
+
+		return self
+
+	# ------------------------------------------------------------------
+	# 推論（__call__ で呼び出し可能）
+	# ------------------------------------------------------------------
+	def __call__(self, x: np.ndarray) -> np.ndarray:
+		"""
+		予測を行う。
+
+		Parameters
+		----------
+		x : np.ndarray, shape (N, in_dim)
+
+		Returns
+		-------
+		np.ndarray, shape (N, out_dim)
+		"""
+		return self.predict(x)
+
+	def predict(self, x: np.ndarray) -> np.ndarray:
+		if self._model is None:
+			raise RuntimeError("先に fit() を実行してください。")
+		self._model.eval()
+		with torch.no_grad():
+			pred = self._model(self._to_tensor(x))
+		return pred.cpu().numpy()
+
+	# ------------------------------------------------------------------
+	# 便利メソッド
+	# ------------------------------------------------------------------
+	@property
+	def loss_history(self) -> dict[str, list[float]]:
+		"""学習損失の履歴を返す。"""
+		h = {"train": self._train_loss_history}
+		if self._val_loss_history:
+			h["val"] = self._val_loss_history
+		return h
+
+	def save(self, path: str) -> None:
+		"""モデルの重みを保存する。"""
+		if self._model is None:
+			raise RuntimeError("モデルがまだ構築されていません。")
+		torch.save(self._model.state_dict(), path)
+
+	def load(self, path: str, in_dim: int, out_dim: int) -> None:
+		"""保存済みの重みを読み込む。"""
+		self._model = self._build_model(in_dim, out_dim)
+		self._model.load_state_dict(torch.load(path, map_location=self.device))
+
+	def summary(self) -> str:
+		"""モデル構造の概要を返す。"""
+		if self._model is None:
+			return "モデル未構築（fit() 実行前）"
+		lines = [str(self._model)]
+		total = sum(p.numel() for p in self._model.parameters())
+		trainable = sum(p.numel() for p in self._model.parameters() if p.requires_grad)
+		lines.append(f"Total params: {total:,}")
+		lines.append(f"Trainable params: {trainable:,}")
+		return "\n".join(lines)

deep00-0.0.0/deep00.egg-info/PKG-INFO

@@ -0,0 +1,26 @@
+Metadata-Version: 2.1
+Name: deep00
+Version: 0.0.0
+Summary: Ultra simple deep learning tool
+Home-page: https://github.co.jp/
+Author: bib_inf
+Author-email: contact.bibinf@gmail.com
+License: CC0 v1.0
+Classifier: Programming Language :: Python :: 3
+Classifier: Topic :: Software Development :: Libraries
+Classifier: License :: CC0 1.0 Universal (CC0 1.0) Public Domain Dedication
+Description-Content-Type: text/markdown
+
+English description follows Japanese.
+
+---
+
+### 概要
+超簡単に深層学習が使えるツールです。
+
+説明は執筆中です
+
+---
+
+### Overview
+description is under construction.

deep00-0.0.0/deep00.egg-info/SOURCES.txt

@@ -0,0 +1,9 @@
+README.md
+setup.py
+deep00/__init__.py
+deep00/test.py
+deep00.egg-info/PKG-INFO
+deep00.egg-info/SOURCES.txt
+deep00.egg-info/dependency_links.txt
+deep00.egg-info/requires.txt
+deep00.egg-info/top_level.txt

deep00-0.0.0/deep00.egg-info/dependency_links.txt

@@ -0,0 +1 @@
+

deep00-0.0.0/deep00.egg-info/requires.txt

@@ -0,0 +1 @@
+ezpip

deep00-0.0.0/deep00.egg-info/top_level.txt

@@ -0,0 +1 @@
+deep00

deep00-0.0.0/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+

deep00-0.0.0/setup.py

@@ -0,0 +1,29 @@
+
+from setuptools import setup
+# 公開用パッケージの作成 [ezpip]
+import ezpip
+
+# 公開用パッケージの作成 [ezpip]
+with ezpip.packager(develop_dir = "./_develop_deep00/") as p:
+	setup(
+		name = "deep00",
+		version = "0.0.0",
+		description = "Ultra simple deep learning tool",
+		author = "bib_inf",
+		author_email = "contact.bibinf@gmail.com",
+		url = "https://github.co.jp/",
+		packages = p.packages,
+		install_requires = ["ezpip"],
+		long_description = p.long_description,
+		long_description_content_type = "text/markdown",
+		license = "CC0 v1.0",
+		classifiers = [
+			"Programming Language :: Python :: 3",
+			"Topic :: Software Development :: Libraries",
+			"License :: CC0 1.0 Universal (CC0 1.0) Public Domain Dedication"
+		],
+		# entry_points = """
+		# 	[console_scripts]
+		# 	py6 = py6:console_command
+		# """
+	)