No no, thats fine. Check how do you do id column and then take a look at this line:
df.index = [i.split('.jpg')[0].split('/')[-1] for i in data.test_dl.dataset.fnames]
2 Likes
ahhh sure enough when I look at data.test_dl.dataset.fnames the order is different from the submission file! Iāll have to resort the predictions it seemsā¦thanks!!
2 Likes
Sure. That would be great.
Please send me your kaggle ids, I will send invites
Did anyone get less than .207 as a loss in this comp? I reached it, but when I added augmentation it made it worse and reached 2.18 (Though my local validation is done better with 89% accuracy). I lost here, now I am trying to find what makes the difference between my local validation and the test set.
Thanks @jeremy for the wonderful start, without your Mooc I would not even submit one entry, Now I am on top 28%.
Nice job!
Which comp @jamesrequa ?
My kaggle id: jamesrequa
Awesome!!!
Dog breedā¦https://www.kaggle.com/c/dog-breed-identification
@sermakarevich also from our fast.ai class is ranked even higher at 8th place 
Congrats!
Great work James.
Congrats to you for an even better score!
Haha mine was supposed to be 1 submission that somehow turned into 4ā¦thanks for your help with sorting out that part
1 Like
Did your issue just end up being that your filenames werenāt matched up to the images?
@KevinB Yep! Always remember to look at data.test_dl.dataset.fnames to make sure the sorting of the test files and hence also the predictions ordering is what you expected them to be!
Thatās burned me twice and it was exactly how you were describing. Training awesome, Validation awesome, testing absolute garbage.I was using kerasā flow_from_directory and I was able to just set shuffle=false to fix it, but I havenāt dug into these libraries enough to know if there is an equivalent.
https://www.kaggle.com/c/invasive-species-monitoring/discussion/38165 Ok I am not crazy - I seen you here Thanks for this sharing, it helped me a lot.
1 Like
Aha that was a fun compā¦Glad you were able to get something out of it! And look now you are teaching me
1 Like
