Skip to content

Installation & Setup

Step-by-step installation guide for both Windows and Linux / macOS.

Quick path: If you just want to train, skip this page and use local_train.py — it handles everything automatically. See Local Training (Automated).

Prerequisites

Requirement Minimum Recommended Notes
Python 3.9 3.11+ Must be on PATH
Git Any Latest For cloning the repository
GPU (CUDA) Optional RTX 3050 4 GB+ Falls back to CPU gracefully
Kaggle account Required — Free, for dataset download
Disk space ~8 GB ~15 GB Dataset + checkpoints + exports
RAM 8 GB 16 GB+

Installation — Windows

Open Command Prompt or PowerShell (not Administrator).

Step 1 — Install Python 3.11+

Download from python.org/downloads and run the installer.

  • ✅ Check "Add Python to PATH" before clicking Install.
  • ✅ Check "Install for all users" (optional but recommended).

Verify: 

python --version
pip --version

Step 2 — Install Git

Download from git-scm.com/download/win and install with default settings.

Verify: 

git --version

Step 3 — (Optional) Visual C++ Build Tools

Some Python packages compile C extensions. PyTorch does not require this (pre-built wheels). Install only if you encounter build errors with other packages.

Download from: https://visualstudio.microsoft.com/visual-cpp-build-tools/
Select workload: "Desktop development with C++"

Step 4 — Clone the repository

git clone https://github.com/Bharaths31/ML-CB-B-identifier
cd ML-CB-B-identifier

Step 5 — Set up Kaggle credentials

Get your API key: kaggle.com/settings → API → Create New Token

mkdir %USERPROFILE%\.kaggle
copy kaggle.json %USERPROFILE%\.kaggle\kaggle.json

Or skip this step — local_train.py will prompt you interactively on first run.

Step 6 — Run the pipeline

python local_train.py --quarter-data

The script creates .venv, installs dependencies, downloads the dataset, trains the model, and exports it.

Windows: Manual Virtual Environment (Optional)

If you prefer to set up the environment manually without local_train.py:

:: Create virtual environment
python -m venv .venv

:: Activate
.venv\Scripts\activate

:: Upgrade pip
python -m pip install --upgrade pip

:: Install training dependencies
pip install torch>=2.1.0 torchvision>=0.16.0
pip install numpy pandas matplotlib scikit-learn tqdm Pillow requests onnx

After activation, your prompt shows (.venv). All subsequent python and pip commands use the venv.

To deactivate: 

deactivate

Installation — Linux / macOS

Step 1 — Install Python 3.11+

Ubuntu / Debian: 

sudo apt update
sudo apt install python3.11 python3.11-venv python3.11-dev python3-pip -y

Check your default python3 version: 

python3 --version
If it's still an older version, use python3.11 explicitly.

macOS (Homebrew): 

brew install python@3.11

Add to PATH if needed (put in ~/.zshrc or ~/.bashrc): 

export PATH="$(brew --prefix python@3.11)/bin:$PATH"

Fedora / RHEL: 

sudo dnf install python3.11 python3.11-devel -y

Step 2 — Install Git

Ubuntu / Debian: 

sudo apt install git -y

macOS: 

brew install git

Fedora: 

sudo dnf install git -y

Verify: 

git --version

Step 3 — Clone the repository

git clone https://github.com/Bharaths31/ML-CB-B-identifier
cd ML-CB-B-identifier

Step 4 — Set up Kaggle credentials

Get your API key: kaggle.com/settings → API → Create New Token

mkdir -p ~/.kaggle
cp kaggle.json ~/.kaggle/kaggle.json
chmod 600 ~/.kaggle/kaggle.json   # Required — Kaggle CLI checks permissions

Or skip this step — local_train.py will prompt you interactively on first run.

Step 5 — Run the pipeline

python3 local_train.py --quarter-data

Linux / macOS: Manual Virtual Environment (Optional)

# Create virtual environment
python3 -m venv .venv

# Activate
source .venv/bin/activate

# Upgrade pip
pip install --upgrade pip

# Install training dependencies
pip install torch>=2.1.0 torchvision>=0.16.0
pip install numpy pandas matplotlib scikit-learn tqdm Pillow requests onnx

After activation, your prompt shows (.venv). All subsequent python and pip commands use the venv.

To deactivate: 

deactivate

Backbone Weights

The pretrained ImageNet backbone weights are already included in the repository:

efficientnet_lite2.pth   # ~24 MB — used when --backbone lite2 (default)
efficientnet_lite4.pth   # ~50 MB — used when --backbone lite4

No separate download needed. They are loaded automatically by src/train.py at the start of Phase 1.

Dependencies

Training Dependencies (required)

torch>=2.1.0
torchvision>=0.16.0
numpy>=1.24
pandas>=1.5
matplotlib>=3.7
scikit-learn>=1.3
tqdm>=4.66
Pillow>=10.0
requests>=2.31
onnx>=1.16

local_train.py installs these automatically.

Model Tester (test_model.py)

Uses only torch, torchvision, and Pillow — already included above. No extra installation required.

Troubleshooting Installation

Error Cause Fix
'python' is not recognized (Windows) Python not on PATH Re-run Python installer, check "Add to PATH"
python3: command not found (Linux) Python not installed sudo apt install python3.11 -y
pip install build fails Missing MSVC (Windows) Install Visual C++ Build Tools
kaggle: command not found Kaggle CLI not installed Not needed — local_train.py uses the Kaggle REST API directly
Permission denied ~/.kaggle Wrong file permissions chmod 600 ~/.kaggle/kaggle.json
GPU not detected CUDA drivers missing Install NVIDIA drivers + CUDA toolkit, or use --device cpu