# Setup Guide for DeepPurpose (Windows/CUDA 11.8)

This guide reproduces the current working environment for `deeppurpose002.py` on Windows 10/11 with an NVIDIA GPU.

## 1. Prerequisites
- **Python 3.10.x** (Required for compatibility with DGL and DeepPurpose)
- **NVIDIA GPU** with drivers installed.

## 2. Create Virtual Environment
Open PowerShell/Terminal in your project folder:
```powershell
python -3.10 -m venv .venv
.\.venv\Scripts\Activate
```

## 3. Install PyTorch (CUDA 11.8)
Install the specific CUDA-enabled version of PyTorch:
```powershell
pip install torch==2.7.1+cu118 torchvision==0.22.1+cu118 torchaudio==2.7.1+cu118 --index-url https://download.pytorch.org/whl/cu118
```
*(Note: If 2.7.1+cu118 is not available in the future, check pytorch.org for the compatible LTS version matching DGL)*

## 4. Install Deep Graph Library (DGL)
DGL must match the CUDA version:
```powershell
pip install dgl==2.2.1+cu118 -f https://data.dgl.ai/wheels/cu118/repo.html
pip install dgllife
```

## 5. Install Core Dependencies
DeepPurpose and PyTDC (Therapeutics Data Commons):
```powershell
pip install DeepPurpose PyTDC
pip install git+https://github.com/bp-kelley/descriptastorus
pip install pandas numpy matplotlib prettytable tensorboard lifelines scikit-learn
```

## 6. Windows-Specific Fixes (Crucial)

### Fix A: PyTDC Dependency Issues (`tiledbsoma`, `cellxgene_census`)
PyTDC tries to install `cellxgene_census`, which depends on `tiledbsoma`. On Windows, `tiledbsoma` often fails to build.
**Solution:** If `pip install PyTDC` fails, manually install dependencies excluding the problematic ones, or create mock files if the code doesn't actually use them (PyTDC generally works without them for standard datasets like DAVIS/KIBA).

### Fix B: DGL GraphBolt DLL Error
DGL 2.x includes "GraphBolt" which requires C++ DLLs that may be missing or incompatible on Windows.
**Error:** `FileNotFoundError: ... graphbolt_pytorch_X.X.X.dll`
**Solution:**
1.  **Code Fix (Recommended):** Add this line to the **very top** of your Python script (before `import torch` or `import dgl`):
    ```python
    import os
    os.environ["DGL_DISABLE_GRAPHBOLT"] = "1"
    ```
2.  **Library Fix (Manual):** If the error persists, edit `.venv\Lib\site-packages\dgl\graphbolt\__init__.py` and comment out all imports to prevent it from loading.

## 7. Running the Script
```powershell
python deeppurpose002.py
```
To force usage of a specific GPU encoding:
```powershell
python deeppurpose002.py --drug_enc DGL_GCN --target_enc CNN --epochs 10
```