# 6th place solution fastai

Here are the write-up and code for my solution!
Blog post:
[https://link.medium.com/Kv5kyHjcIX](https://link.medium.com/Kv5kyHjcIX)
Code:
[https://github.com/mnpinto/audiotagging2019](https://github.com/mnpinto/audiotagging2019)
Summary:
* Models: xresnets
* Image size: 256x256
* Mixup sampling from a uniform distribution
* Horizontal and Vertical Flip as new labels (total 320 labels)
* Compute loss only for samples with F2 score (with a threshold of 0.2) less than 1.
* Noisy data: ~3500 "good noisy samples" used the same way as curated data
* TTA: Slice clips each 128px in the time axis (no overlap), generate predictions for each slice and compute the
max
for each class.
* Final submissions: 1) Average of 2 models: public LB 0.742, private LB 0.74620; 2) Average of 6 models: public LB 0.742, private LB 0.75421.
Additional observations:
* I found better results when using random crops of 128x128 and rescale to 256x256, comparing to random crops of 128x256 and rescale to 256x256, I was expecting the opposite.
* I wonder why max_zoom=1.5 works; I would not expect so.
Acknowledgements:
[@daisukelab](/daisukelab)
thanks for the code to generate the mel spectrograms! Thanks to everyone that contributed in the discussions or with kernels. And finally, thanks to the organizers for this great competition!