# 110th place effort

I'd like to share a brief overview of the work I've done to get to the 110th place. That was close to the medal zone, in which I actually spent a few weeks, but the competition really went wild on the last ~10 days and I couldn't spend much time on it, so at the end I didn't earn any medal, but still learnt a lot.
Overview:
Used the data preprocessed in a few different image sizes ( (256,512,3), (512,512,3)…. ) and max length (282, 282/2).
Uploaded all the preprocessed datasets to my personal account in Google Cloud Storage, which allowed me to use TPU in Colab Pro.
Trained some Densenet pytorch models using GPU (oof, that took some time) and some tensorflow models using TPU. The tensorflow models differed from each other mostly in hyperparameters choices and the dataset used.
Encoders tested: Densenet-169, EfficientNet-B0/B2/B4, EfficientNetV2-B2, and other
[timm models](https://github.com/rwightman/pytorch-image-models/tree/master/timm)
Decoders tested: LSTM with attention, Transformer with attention
I've got the most effective results by manually tuning the learning rate scheduler
Post-processed ~10 submissions using
[@kirderf](https://www.kaggle.com/kirderf)
[notebook](https://www.kaggle.com/kirderf/basic-postp-with-different-submissions/)
RDKit normalization using
[@nofreewill](https://www.kaggle.com/nofreewill)
[notebook](https://www.kaggle.com/nofreewill/normalize-your-predictions)
I guess my solution lacked a voting system or a more careful look at the samples predicted to be in the medal zone (top 100).
Thanks for everyone who competed and shared works and ideas!