Class imbalance in this dataset

mdribeiro / DeepCFD

DeepCFD: Efficient Steady-State Laminar Flow Approximation with Deep Convolutional Neural Networks

MIT License

204 stars 65 forks source link

Class imbalance in this dataset #1

Open airlexi opened 4 years ago

airlexi commented 4 years ago

Hey guys, I was exploring your code and found a huge class imbalance in the dataset. Your testing split contains 2samples which contribute almost everything to the MSE. Leaving those two out would reduce your achieved MSE of 2.04 to around 0.3. The CFD simulation of those samples looks totally different than the rest. In total there are about 6 or 7 of those samples in your dataset. Since I guess this is not intended I wanted to let you know :) Best regards Alex

mdribeiro commented 4 years ago

Hello Alex, Thank you so much for letting us know about this. We are currently working on to improve the dataset, creating more diverse samples (e.g. moving the obstacles around the domain and increasing the obstacle size variability), as well as correcting some bugs in the data generation pipeline that have been found, such as the issue you mentioned about. We hope to be able to provide the new data, as well as the new trained model results soon. Best and stay safe! :) Mateus

airlexi commented 4 years ago

That sounds nice :) I will stay tuned!