random neural nets

Implementations of a random selection of artificial neural net based models and methods.

Python version

Development is done using uv, pinning the python version to the one in the file .python-version.

Installation (on Linux)

pip install random-neural-net-models

or

uv add random-neural-net-models

For package development / notebooks shenanigans:

git clone https://github.com/eschmidt42/random-neural-net-models.git
cd random-neural-net-models
make install-dev-env

Usage

See jupyter notebooks in nbs/ for:

• fastai style learner with tensordict: learner.ipynb
• perceptron: perceptron.ipynb
• backpropagation: backpropagation_rumelhart1986.ipynb
• convolution: convolution_lecun1990.ipynb
• cnn autoencoder:
  • mnist: cnn_autoencoder_fastai2022.ipynb
  • fashion mnist: cnn_autoencoder_fastai2022_fashion.ipynb
• variational autoencoder:
  • dense: dense_variational_autoencoder_fastai2022.ipynb
  • cnn+dense: cnn_variational_autoencoder_fastai2022.ipynb
• optimizers: stochastic_optimization_methods.ipynb
• resnet: resnet_fastai2022.ipynb
• unet:
  • unet_fastai2022.ipynb
  • unet-isbi2012
• diffusion (unet + noise):
  • diffusion_fastai2022_learner.ipynb
  • diffusion_fastai2022_learner_with_attention.ipynb
• mingpt:
  • mingpt_sort.ipynb
  • mingpt_char.ipynb
  • mingpt_adder.ipynb
• transformer: language-model.ipynb
• tokenization: tokenization.ipynb
• tabular:
  • tabular-fastai-classification.ipynb
  • tabular-fastai-classification-with-missingness.ipynb
  • tabular-fastai-classification-with-missingness-and-categories.ipynb
  • tabular-fastai-regression.ipynb
  • tabular-fastai-regression-with-missingness.ipynb
  • tabular-fastai-regression-with-missingness-and-categories.ipynb
  • tabular-variational-auto-encoder.ipynb
  • reusing-vae-for-classification.ipynb

nbs/core contains the notebooks who are guaranteed to function despite refactorings.