/MNIST_CNN

Yet another convolutional neural network trained on MNIST

Primary LanguageJupyter Notebook

MNIST_CNN

Yet another convolutional neural network trained on MNIST

About

This project was made for the contest, which was a part of Introduction to Machine Learning course in Innopolis University. The model was tested on highly augmented data. The final F1 score of my last model was 0.8924. The commit messages contain scores of each model. The provided code can produce model with different weights, and thus, different score.

Files in this repository

  • IMLProject.ipynb - the notebook that was used to train the model. It can also be found at Google Colab.
  • code.py - standalone python file, should work the same as .ipynb notebook, but it was never actually tested.
  • model.h5.zip - the zip file, containing only model.h5 file - the model itself. It was uploaded using git lfs, so it may be unacessable due to storage limits. But the last version can be found at Google Drive
  • Model is trained on augmented MNIST dataset. Unfortunately, dataset with augmentations was not saved during testing.