3W-unsupervised-learning

This project sets up a reproducible Conda environment for running unsupervised learning experiments using:

• MOMENT (time-series foundation model)
• 3W Toolkit (Petrobras 3W dataset utilities)

The setup is intentionally strict: Conda manages all dependencies, while pip is used only to install project code (without pulling dependencies).

---

Prerequisites

Before starting, make sure you have:

• Conda installed (Miniconda or Anaconda)
• Git installed
• A Linux or macOS environment (tested on Linux)

You can verify Conda with:

conda --version

---

Repository layout

The 3W repository must be cloned one directory above this project.

Example directory structure

projects/
├── 3W/
│   └── (ThreeWToolkit source code)
│
├── 3W-unsupervised-learning/
│   ├── environment.yml
│   ├── setup.sh
│   └── README.md

Clone repositories

cd projects

git clone --branch v.1.60.0 --single-branch https://github.com/petrobras/3W.git

git clone <THIS_REPOSITORY_URL>

---

Environment setup

The entire setup is automated via setup.sh.

What the script does

1. Creates the Conda environment from environment.yml
2. Activates the environment
3. Installs project code (3W and momentfm) using pip --no-deps

Run setup

From inside 3W-unsupervised-learning:

bash setup.sh

This will create and configure the Conda environment named:

3W-unsupervised

---

Activating the environment

After installation:

conda activate 3W-unsupervised

---

Verifying the installation

You can quickly test that everything is working:

python - << 'EOF'
import torch
import momentfm
import ThreeWToolkit

print("Setup successful")
print("Torch version:", torch.__version__)
print("CUDA available:", torch.cuda.is_available())
EOF

---

Notes

• Do not install project dependencies with pip
• All numerical, CUDA, and ML dependencies are managed by Conda
• Pip is used strictly to install source code
• This avoids dependency conflicts (e.g., NumPy, CUDA, PyTorch)

---

Troubleshooting

If you encounter ModuleNotFoundError:

• Add the missing package to environment.yml
• Update the environment using:

conda env update -f environment.yml

---