AlexNet Image classification

Data

Requirements

The data is acquired from HuggingFace imagenet-1k (needs login).

This project needs some available space of at least:

	Size (GB)
Raw data	333.3
Transformed data	1125.5
Total	1458.8

ETL

From the root of this project, run:

1. python etl/extract.py
2. python etl/transform.py

The local folder dataset should have the following structure:

dataset
└── imagenet_1k_256x256_float32.h5

Running locally with pip

1. Run python -m venv .venv

2. Activate the virtual environment with source .venv/bin/activate

3. Install pip dependencies pip install -r requirements.txt

4. Start a local MLflow server inside a container:

   podman run -d -p 5000:5000 -v ./mlruns:/mlflow/mlruns ghcr.io/mlflow/mlflow mlflow server --backend-store-uri /mlflow/mlruns --default-artifact-root /mlflow/mlruns --host 0.0.0.0

5. Run python src/main.py to train the model

Running locally with Podman/Docker

1. Build the image: podman build -t alex-net:local .
2. Example of run command (Adjust to your GPU config):

   podman run \
   -v "YOUR_DATASET_DIR":/app/dataset \
   -e DATASET_FILE_DIR="/app/dataset/imagenet_1k_256x256_float32.h5" \
   --device nvidia.com/gpu=all \
   --shm-size=512m \
   alex-net:local

Development

When changing the dependencies of pip:

1. Run: pip freeze > requirements.txt