Example MNIST Tensorflow

This example shows how to:

• Define a CNN by subclassing the tf.keras.Model class
• Train the model (from scratch or loading an existing model)
• Visualize training history
• Load & use a dataset from Tensorflow & Hugging Face
• Load & Save models using SavedModel & SafeTensors formats
• Inference

Index

• Setup Virtual environment with venv
• Training
• Inference
• Visualization
• Resources

Virtual Environment & Dependencies

Create Virtual environment & install dependencies

python3.11 -m venv venv
source venv/bin/activate
pip install --upgrade pip
pip install -r requirements.txt

Activate Virtual environment

source venv/bin/activate

Deactivate Virtual environment

deactivate

Training

--save_freq=0 if you don't want to save the model

Training from Scratch

python train.py --epochs=10 --save_freq=10 --batch_size=64

Training loading previous version

python train.py --epochs=10 --save_freq=5 --load --batch_size=64

Inference

Loads all images inside the selected folder and makes predictions.

images inside sample's folder have different sizes

python inference.py --img_folder="samples"

Visualization

Resources

• Datasets

  • MNIST database
  • Hugging Face Datasets Docs
  • Hugging Face MNIST Dataset
  • Tensorflow MNIST Dataset

• Save & Load Models

  • SafeTensors
  • SavedModel Tensorflow

• Model Definition

  • tf.keras.Model