rasbt / scipy2022-talk

scipy2022-talk

Slides: https://sebastianraschka.com/pdf/slides/2022-07-scipy-corn.pdf

Using the Code

Step 1: Install the requirements

git clone https://github.com/rasbt/scipy2022-talk.git
cd scipy2022-talk
conda create -n coral-pytorch python=3.8
conda activate coral-pytorch
pip install -r requirements.txt
python -m spacy download en_core_web_sm

Step 2: Run the code

MLP with CORN loss

cd src

python main_mlp.py \
--batch_size 16 \
--data_path ../datasets/ \
--learning_rate 0.01 \
--mixed_precision true \
--num_epochs 40 \
--num_workers 3 \
--output_path ./cement_strength \
--loss_mode corn

MLP with cross entropy loss

python main_mlp.py \
...
--loss_mode crossentropy

More examples

• PyTorch Hub for loading pre-trained models: https://github.com/rasbt/ord-torchhub
• Tutorials for using the various ordinal regression models with CNNs, RNNs, and MLPs: https://github.com/Raschka-research-group/coral-pytorch
• The CORN paper repository with detailed experiment logs: https://github.com/Raschka-research-group/corn-ordinal-neuralnet

Interactive Demo

You can try an interactive version at https://bit.ly/3aCgSeG.

(The source code for this interactive demo is available at https://github.com/rasbt/ord-torchhub/tree/main/app.)

References

• Xintong Shi, Wenzhi Cao, and Sebastian Raschka Deep Neural Networks for Rank-Consistent Ordinal Regression Based On Conditional Probabilities. https://arxiv.org/abs/2111.08851