Kaggle Notebook Internet Offでpytorch_tabnetを使う方法 (2023年12月時点)

はじめに

Kaggle Typingコンペ（Code competition）にてInternet offでのTabNetのインストールに困ったので解決策を残しておきます。

問題

TabNetのpackage依存関係でエラーが起きていました。
古いバージョンのTabNet (4.0以前) は要件に1.2 <= torch < 2.0を含みますが、Kaggle Notebook環境ではtorch==2.0.0となっており、以下のエラーが生じていました。

!pip install --no-index --find-links /kaggle/input/pytorch-tabnet/pytorch_tabnet-4.0-py3-none-any.whl pytorch_tabnet
from pytorch_tabnet.tab_model import TabNetRegressor

Looking in links: /kaggle/input/pytorch-tabnet/pytorch_tabnet-4.0-py3-none-any.whl
Processing /kaggle/input/pytorch-tabnet/pytorch_tabnet-4.0-py3-none-any.whl
Requirement already satisfied: numpy<2.0,>=1.17 in /opt/conda/lib/python3.10/site-packages (from pytorch_tabnet) (1.24.3)
Requirement already satisfied: scikit_learn>0.21 in /opt/conda/lib/python3.10/site-packages (from pytorch_tabnet) (1.2.2)
Requirement already satisfied: scipy>1.4 in /opt/conda/lib/python3.10/site-packages (from pytorch_tabnet) (1.11.4)
INFO: pip is looking at multiple versions of pytorch-tabnet to determine which version is compatible with other requirements. This could take a while.
ERROR: Could not find a version that satisfies the requirement torch<2.0,>=1.2 (from pytorch-tabnet) (from versions: none)
ERROR: No matching distribution found for torch<2.0,>=1.2
---------------------------------------------------------------------------
ModuleNotFoundError                       Traceback (most recent call last)
Cell In[6], line 3
      1 # install TabNet
      2 get_ipython().system('pip install --no-index --find-links /kaggle/input/pytorch-tabnet/pytorch_tabnet-4.0-py3-none-any.whl pytorch_tabnet')
----> 3 from pytorch_tabnet.tab_model import TabNetRegressor

ModuleNotFoundError: No module named 'pytorch_tabnet'

環境
Python 3.10.12
pytorch-tabnet 4.0
torch 2.0.0
numpy 1.24.3
scikit_learn 1.2.2
scipy 1.11.4
tqdm 4.66.1

解決策

pywheelsにあるpytorch-tabnetの最新版wheelファイル (4.1.0) を使用するとうまくいきました。確認したところ、pytorch-tabnet version4.1.0からはtorch >= 2.0もサポートされていました。

手順

1. TabNetのwheelファイルを下記よりダウンロード
   https://www.piwheels.org/project/pytorch-tabnet/
2. wheelファイルをdatasetとしてアップロード
3. pipでインストール

以下はwheelファイルを/kaggle/input/pytorch-tabnetに入れたときのコマンドです

!pip install --no-index --find-links /kaggle/input/pytorch-tabnet/pytorch_tabnet-4.1.0-py3-none-any.whl pytorch_tabnet
from pytorch_tabnet.tab_model import TabNetRegressor

Looking in links: /kaggle/input/pytorch-tabnet/pytorch_tabnet-4.1.0-py3-none-any.whl
Processing /kaggle/input/pytorch-tabnet/pytorch_tabnet-4.1.0-py3-none-any.whl
Requirement already satisfied: numpy>=1.17 in /opt/conda/lib/python3.10/site-packages (from pytorch_tabnet) (1.24.3)
Requirement already satisfied: scikit_learn>0.21 in /opt/conda/lib/python3.10/site-packages (from pytorch_tabnet) (1.2.2)
Requirement already satisfied: scipy>1.4 in /opt/conda/lib/python3.10/site-packages (from pytorch_tabnet) (1.11.4)
Requirement already satisfied: torch>=1.3 in /opt/conda/lib/python3.10/site-packages (from pytorch_tabnet) (2.0.0+cpu)
Requirement already satisfied: tqdm>=4.36 in /opt/conda/lib/python3.10/site-packages (from pytorch_tabnet) (4.66.1)
Requirement already satisfied: joblib>=1.1.1 in /opt/conda/lib/python3.10/site-packages (from scikit_learn>0.21->pytorch_tabnet) (1.3.2)
Requirement already satisfied: threadpoolctl>=2.0.0 in /opt/conda/lib/python3.10/site-packages (from scikit_learn>0.21->pytorch_tabnet) (3.2.0)
Requirement already satisfied: filelock in /opt/conda/lib/python3.10/site-packages (from torch>=1.3->pytorch_tabnet) (3.12.2)
Requirement already satisfied: typing-extensions in /opt/conda/lib/python3.10/site-packages (from torch>=1.3->pytorch_tabnet) (4.5.0)
Requirement already satisfied: sympy in /opt/conda/lib/python3.10/site-packages (from torch>=1.3->pytorch_tabnet) (1.12)
Requirement already satisfied: networkx in /opt/conda/lib/python3.10/site-packages (from torch>=1.3->pytorch_tabnet) (3.1)
Requirement already satisfied: jinja2 in /opt/conda/lib/python3.10/site-packages (from torch>=1.3->pytorch_tabnet) (3.1.2)
Requirement already satisfied: MarkupSafe>=2.0 in /opt/conda/lib/python3.10/site-packages (from jinja2->torch>=1.3->pytorch_tabnet) (2.1.3)
Requirement already satisfied: mpmath>=0.19 in /opt/conda/lib/python3.10/site-packages (from sympy->torch>=1.3->pytorch_tabnet) (1.3.0)
Installing collected packages: pytorch_tabnet
Successfully installed pytorch_tabnet-4.1.0

おわりに

バージョン管理の大切さを身をもって学びました。