Merge pull request #17 from sgrvinod/0.3.1

0.3.1

Author: sgrvinod

CHANGELOG.md

@@ -1,10 +1,24 @@
 # Change Log
 
+## v0.3.1
+
+### Added
+
+* A [**`pyproject.toml`**
+file](https://github.com/sgrvinod/chess-transformers/blob/main/pyproject.toml) has been added in compliance with [PEP 660](https://peps.python.org/pep-0660/). While the inclusion of a `setup.py` file is not deprecated, its use as a command-line tool, such as in the legacy `setup.py develop` method for performing an editable installation is now deprecated. 
+
+### Changed
+
+* **`chess_transformers.train.datasets.ChessDataset`** was optimized for large datasets. A list of indices for the data split is no longer maintained or indexed in the dataset.
+* The `TRAINING_CHECKPOINT` parameter in each of **`chess_transformers.configs.models`** was set to `None` to reflect the correct conditions for beginning training of a model.
+* Dynamic shape tracing is disabled for the compilation of [*CT-ED-45*](https://github.com/sgrvinod/chess-transformers#ct-ed-45) to prevent memory leaks as seen in [#16](https://github.com/sgrvinod/chess-transformers/issues/16).
+* References to `torch.cuda.amp.GradScaler(...)` have been replaced by `torch.amp.GradScaler(device="cuda", ...)` following its deprecation.
+
 ## v0.3.0
 
 ### Added
 
-* There are 3 new datasets: [ML23c](https://github.com/sgrvinod/chess-transformers#ml23c), [GC22c](https://github.com/sgrvinod/chess-transformers#gc22c), and [ML23d](https://github.com/sgrvinod/chess-transformers#ml23d).
+* There are 3 new datasets: [*ML23c*](https://github.com/sgrvinod/chess-transformers#ml23c), [*GC22c*](https://github.com/sgrvinod/chess-transformers#gc22c), and [*ML23d*](https://github.com/sgrvinod/chess-transformers#ml23d).
 * A new naming convention for datasets is used. Datasets are now named in the format "[*PGN Fileset*][*Filters*]". For example, *LE1222* is now called [*LE22ct*](https://github.com/sgrvinod/chess-transformers#le22ct), where *LE22* is the name of the PGN fileset from which this dataset was derived, and "*c*", "*t*" are filters for games that ended in checkmates and games that used a specific time control respectively.
 * [*CT-EFT-85*](https://github.com/sgrvinod/chess-transformers#ct-eft-85) is a new trained model with about 85 million parameters.
 * **`chess_transformers.train.utils.get_lr()`** now accepts new arguments, `schedule` and `decay`, to accomodate a new learning rate schedule: exponential decay after warmup.

README.md

@@ -4,7 +4,7 @@
 
 <h1 align="center"><i>Chess Transformers</i></h1>
 <p align="center"><i>Teaching transformers to play chess</i></p>
-<p align="center"> <a href="https://github.com/sgrvinod/chess-transformers/releases/tag/v0.3.0"><img alt="Version" src="https://img.shields.io/github/v/tag/sgrvinod/chess-transformers?label=version"></a> <a href="https://github.com/sgrvinod/chess-transformers/blob/main/LICENSE"><img alt="License" src="https://img.shields.io/github/license/sgrvinod/chess-transformers?label=license"></a></p>
+<p align="center"> <a href="https://github.com/sgrvinod/chess-transformers/releases/tag/v0.3.1"><img alt="Version" src="https://img.shields.io/github/v/tag/sgrvinod/chess-transformers?label=version"></a> <a href="https://github.com/sgrvinod/chess-transformers/blob/main/LICENSE"><img alt="License" src="https://img.shields.io/github/license/sgrvinod/chess-transformers?label=license"></a></p>
 <br>
 
 *Chess Transformers* is a library for training transformer models to play chess by learning from human games. 

chess_transformers/configs/models/CT-E-20.py

@@ -94,8 +94,8 @@
     pathlib.Path(__file__).parent.parent.parent.resolve() / "checkpoints" / NAME
 )  # folder containing checkpoints
 TRAINING_CHECKPOINT = (
-    NAME + ".pt"
-)  # path to model checkpoint to resume training, None if none
+    None  # path to model checkpoint (NAME + ".pt") to resume training, None if none
+)
 CHECKPOINT_AVG_PREFIX = (
     "step"  # prefix to add to checkpoint name when saving checkpoints for averaging
 )

chess_transformers/configs/models/CT-ED-45.py

@@ -50,7 +50,7 @@
 N_MOVES = 10  # expected maximum length of move sequences in the model, <= MAX_MOVE_SEQUENCE_LENGTH
 DISABLE_COMPILATION = False  # disable model compilation?
 COMPILATION_MODE = "default"  # mode of model compilation (see torch.compile())
-DYNAMIC_COMPILATION = True  # expect tensors with dynamic shapes?
+DYNAMIC_COMPILATION = False  # expect tensors with dynamic shapes?
 SAMPLING_K = 1  # k in top-k sampling model predictions during play
 MODEL = ChessTransformer  # custom PyTorch model to train
 
@@ -94,8 +94,8 @@
     pathlib.Path(__file__).parent.parent.parent.resolve() / "checkpoints" / NAME
 )  # folder containing checkpoints
 TRAINING_CHECKPOINT = (
-    NAME + ".pt"
-)  # path to model checkpoint to resume training, None if none
+    None  # path to model checkpoint (NAME + ".pt") to resume training, None if none
+)
 CHECKPOINT_AVG_PREFIX = (
     "step"  # prefix to add to checkpoint name when saving checkpoints for averaging
 )

chess_transformers/configs/models/CT-EFT-20.py

@@ -94,8 +94,8 @@
     pathlib.Path(__file__).parent.parent.parent.resolve() / "checkpoints" / NAME
 )  # folder containing checkpoints
 TRAINING_CHECKPOINT = (
-    NAME + ".pt"
-)  # path to model checkpoint to resume training, None if none
+    None  # path to model checkpoint (NAME + ".pt") to resume training, None if none
+)
 CHECKPOINT_AVG_PREFIX = (
     "step"  # prefix to add to checkpoint name when saving checkpoints for averaging
 )

chess_transformers/configs/models/CT-EFT-85.py

@@ -93,7 +93,9 @@
 CHECKPOINT_FOLDER = str(
     pathlib.Path(__file__).parent.parent.parent.resolve() / "checkpoints" / NAME
 )  # folder containing checkpoints
-TRAINING_CHECKPOINT = None  # path to model checkpoint to resume training, None if none
+TRAINING_CHECKPOINT = (
+    None  # path to model checkpoint (NAME + ".pt") to resume training, None if none
+)
 AVERAGE_STEPS = {491000, 492500, 494000, 495500, 497000, 498500, 500000}
 CHECKPOINT_AVG_PREFIX = (
     "step"  # prefix to add to checkpoint name when saving checkpoints for averaging

chess_transformers/train/datasets.py

@@ -34,19 +34,15 @@ def __init__(self, data_folder, h5_file, split, n_moves=None, **unused):
         # Open table in H5 file
         self.h5_file = tb.open_file(os.path.join(data_folder, h5_file), mode="r")
         self.encoded_table = self.h5_file.root.encoded_data
+        self.split = split
 
         # Create indices
-        # TODO: optimize by using a start_index and not a list of indices
         if split == "train":
-            self.indices = list(range(0, self.encoded_table.attrs.val_split_index))
+            self.first_index = 0
         elif split == "val":
-            self.indices = list(
-                range(
-                    self.encoded_table.attrs.val_split_index, self.encoded_table.nrows
-                )
-            )
+            self.first_index = self.encoded_table.attrs.val_split_index
         elif split is None:
-            self.indices = list(range(0, self.encoded_table.nrows))
+            self.first_index = 0
         else:
             raise NotImplementedError
 
@@ -56,33 +52,41 @@ def __init__(self, data_folder, h5_file, split, n_moves=None, **unused):
         if n_moves is not None:
             # This is the same as min(MAX_MOVE_SEQUENCE_LENGTH, n_moves)
             self.n_moves = min(
-                len(self.encoded_table[self.indices[0]]["moves"]) - 1, n_moves
+                len(self.encoded_table[self.first_index]["moves"]) - 1, n_moves
             )
         else:
-            self.n_moves = len(self.encoded_table[self.indices[0]]["moves"]) - 1
+            self.n_moves = len(self.encoded_table[self.first_index]["moves"]) - 1
 
     def __getitem__(self, i):
-        turns = torch.IntTensor([self.encoded_table[self.indices[i]]["turn"]])
+        turns = torch.IntTensor([self.encoded_table[self.first_index + i]["turn"]])
         white_kingside_castling_rights = torch.IntTensor(
-            [self.encoded_table[self.indices[i]]["white_kingside_castling_rights"]]
+            [self.encoded_table[self.first_index + i]["white_kingside_castling_rights"]]
         )  # (1)
         white_queenside_castling_rights = torch.IntTensor(
-            [self.encoded_table[self.indices[i]]["white_queenside_castling_rights"]]
+            [
+                self.encoded_table[self.first_index + i][
+                    "white_queenside_castling_rights"
+                ]
+            ]
         )  # (1)
         black_kingside_castling_rights = torch.IntTensor(
-            [self.encoded_table[self.indices[i]]["black_kingside_castling_rights"]]
+            [self.encoded_table[self.first_index + i]["black_kingside_castling_rights"]]
         )  # (1)
         black_queenside_castling_rights = torch.IntTensor(
-            [self.encoded_table[self.indices[i]]["black_queenside_castling_rights"]]
+            [
+                self.encoded_table[self.first_index + i][
+                    "black_queenside_castling_rights"
+                ]
+            ]
         )  # (1)
         board_position = torch.IntTensor(
-            self.encoded_table[self.indices[i]]["board_position"]
+            self.encoded_table[self.first_index + i]["board_position"]
         )  # (64)
         moves = torch.LongTensor(
-            self.encoded_table[self.indices[i]]["moves"][: self.n_moves + 1]
+            self.encoded_table[self.first_index + i]["moves"][: self.n_moves + 1]
         )  # (n_moves + 1)
         length = torch.LongTensor(
-            [self.encoded_table[self.indices[i]]["length"]]
+            [self.encoded_table[self.first_index + i]["length"]]
         ).clamp(
             max=self.n_moves
         )  # (1), value <= n_moves
@@ -99,7 +103,14 @@ def __getitem__(self, i):
         }
 
     def __len__(self):
-        return len(self.indices)
+        if self.split == "train":
+            return self.encoded_table.attrs.val_split_index
+        elif self.split == "val":
+            return self.encoded_table.nrows - self.encoded_table.attrs.val_split_index
+        elif self.split is None:
+            return self.encoded_table.nrows
+        else:
+            raise NotImplementedError
 
 
 class ChessDatasetFT(Dataset):
@@ -175,12 +186,11 @@ def __len__(self):
         elif self.split == "val":
             return self.encoded_table.nrows - self.encoded_table.attrs.val_split_index
         elif self.split is None:
-            self.encoded_table.nrows
+            return self.encoded_table.nrows
         else:
             raise NotImplementedError
 
 
-
 if __name__ == "__main__":
     # Get configuration
     parser = argparse.ArgumentParser()

chess_transformers/train/train.py

@@ -5,7 +5,7 @@
 import torch.backends.cudnn as cudnn
 
 from tqdm import tqdm
-from torch.cuda.amp import GradScaler
+from torch.amp import GradScaler
 from torch.utils.data import DataLoader
 from torch.utils.tensorboard import SummaryWriter
 
@@ -96,7 +96,7 @@ def train_model(CONFIG):
     criterion = criterion.to(DEVICE)
 
     # AMP scaler
-    scaler = GradScaler(enabled=CONFIG.USE_AMP)
+    scaler = GradScaler(device=DEVICE, enabled=CONFIG.USE_AMP)
 
     # Find total epochs to train
     epochs = (CONFIG.N_STEPS // (len(train_loader) // CONFIG.BATCHES_PER_STEP)) + 1

pyproject.toml

@@ -0,0 +1,44 @@
+[build-system]
+requires = ["setuptools >= 64"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "chess-transformers"
+version = "0.3.1"
+description = "Teaching transformers to play chess."
+authors = [{ name = "Sagar Vinodababu", email = "sgrvinod@gmail.com" }]
+maintainers = [{ name = "Sagar Vinodababu", email = "sgrvinod@gmail.com" }]
+readme = "README.md"
+requires-python = ">=3.6.0"
+dependencies = [
+    "beautifulsoup4==4.12.3",
+    "chess==1.10.0",
+    "colorama==0.4.5",
+    "ipython==8.17.2",
+    "Markdown==3.3.4",
+    "py_cpuinfo==9.0.0",
+    "regex==2024.7.24",
+    "scipy==1.13.1",
+    "setuptools==69.0.3",
+    "tables==3.9.2",
+    "tabulate==0.9.0",
+    "torch==2.4.0",
+    "tqdm==4.64.1",
+    "tensorboard==2.18.0",
+]
+license = { text = "MIT License" }
+keywords = ["transformer", "chess", "pytorch", "deep learning", "chess engine"]
+classifiers = [
+    "Development Status :: 3 - Alpha",
+    "Intended Audience :: Science/Research",
+    "License :: OSI Approved :: MIT License",
+    "Programming Language :: Python :: 3.12",
+    "Topic :: Scientific/Engineering :: Artificial Intelligence",
+]
+
+[project.urls]
+homepage = "https://github.com/sgrvinod/chess-transformers"
+source = "https://github.com/sgrvinod/chess-transformers"
+changelog = "https://github.com/sgrvinod/chess-transformers/blob/main/CHANGELOG.md"
+releasenotes = "https://github.com/sgrvinod/chess-transformers/releases"
+issues = "https://github.com/sgrvinod/chess-transformers/issues"

setup.py

@@ -1,43 +1,3 @@
 from setuptools import setup, find_packages
 
-with open("README.md", mode="r", encoding="utf-8") as readme_file:
-    readme = readme_file.read()
-
-
-setup(
-    name="chess-transformers",
-    version="0.3.0",
-    author="Sagar Vinodababu",
-    author_email="sgrvinod@gmail.com",
-    description="Chess Transformers",
-    long_description=readme,
-    long_description_content_type="text/markdown",
-    license="MIT License",
-    url="https://github.com/sgrvinod/chess-transformers",
-    download_url="https://github.com/sgrvinod/chess-transformers",
-    packages=find_packages(),
-    python_requires=">=3.6.0",
-    install_requires=[
-        "beautifulsoup4==4.12.3",
-        "chess==1.10.0",
-        "colorama==0.4.5",
-        "ipython==8.17.2",
-        "Markdown==3.3.4",
-        "py_cpuinfo==9.0.0",
-        "regex==2024.7.24",
-        "scipy==1.13.1",
-        "setuptools==69.0.3",
-        "tables==3.9.2",
-        "tabulate==0.9.0",
-        "torch==2.4.0",
-        "tqdm==4.64.1",
-    ],
-    classifiers=[
-        "Development Status :: 3 - Alpha",
-        "Intended Audience :: Science/Research",
-        "License :: OSI Approved :: MIT License",
-        "Programming Language :: Python :: 3.9",
-        "Topic :: Scientific/Engineering :: Artificial Intelligence",
-    ],
-    keywords="transformer networks chess pytorch deep learning",
-)
+setup(packages=find_packages())