# Simple CNN for knotted x unknotted recognition

The model is trained on downsampled [knots_AF](https://huggingface.co/datasets/EvaKlimentova/knots_AF) dataset where sequences with length < 500 are padded with 'X's.

The architecture is [PENGUINN](https://www.frontiersin.org/articles/10.3389/fgene.2020.568546/full) with modified input.

## Model versions 

(names in format `cnn_{NUMBER_OF_EPOCHS}epochs_{TEST_ACCURACY}.h5`)

**Trained on v1 version of the dataset (it contained duplicites):**

- [`cnn_10epochs_95.h5`](./cnn_10epochs_95.h5)

**Trained on v2 version of the dataset (some unknotted labels were incorrect):**

- [`cnn_5epochs_94.h5`](./cnn_5epochs_94.h5)

- [`cnn_10epochs_94.h5`](./cnn_10epochs_94.h5)

**Trained on the latest version of the dataset:**

- [`cnn_10epochs_latest.h5`](./cnn_10epochs_latest.h5) - accuracy on test: `~95.17 %`