Lesson 1 discussion - beginner


(Tash Robinson) #44

Just started lesson 1. I created an account on Crestle and it says select /courses/fastai2 this folder does not exist. Is it moved or how do I get it to show up? It is mentioned that things have been moving around and all information on the current location is on the website… I’m assuming it meant here in the forums, but seach turned up nothing. Is there somewhere else I should be looking?


(Ilia) #45

Hello.

How many images should I have in train dir to find learning rate by lr_find()? I’m asking, because to distinguish men vs. women ~70 images per each category is kind of enough to train the last layer, but not enough to find learning rate by lr_find(): http://dl3.joxi.net/drive/2018/02/18/0011/4037/757701/01/f6ff6494f9.jpg

Thanks.


(Ismaël Koné) #46

Hi, look here:


(Ilia) #47

Thanks, that’s worked out: https://i.imgur.com/QmIIM8L.jpg


(Ismaël Koné) #48

Great!


(YIYU SHI) #49

Hi dortonway, I noticed you were able to train your own data. I am wondering how you did it.

I created a new folder in ‘data’ and created new train and valid folder, and I put my images into the folders. But when I run the model, it always automatically .ipynb_checkpoints, and it broke.

May I ask if you did the same process? Or what should I do to train my own images?

Thank you!!!


(Aditya) #50

It shouldn’t happen now?
Are you on latest build of fast.ai?


(YIYU SHI) #51

I just tried this afternoon and had this issue

Have you solved the problem? I read all your posts but still can not fix it.


(Aditya) #52

Try removing them manually using rm


(pradla) #53

I set up my paperspace and I seem to be having errors in the very beginning (when I am trying to import libraries). The kernel is dieing.

I updated the conda environment but all seems to be upto date.

I am unable to proceed with the rest of the code as it is throwing errors (like torch is not recognized, os is not recognised etc). I thought this might be the main reason.

Can anyone please let me know what is the problem?


(Ilia) #54

Hi, superives.

You can see what I did / my experiments here.


(YIYU SHI) #55

Thanks for the reply. May I ask how you downloaded the images?
I used methods from here (https://github.com/hardikvasa/google-images-download) and moved pictures to new folders. Did you do the same?

Thank you!


(YIYU SHI) #56

Did you just use ‘rm .ipynb_checkpoints’ in the proper folder? I tried it but it said it was a directory so could not be deleted. Do you mind tell me if I made a mistake here? Thank you!


(Ilia) #57

Yes, I’ve used that repo, but I did nothing to .ipynb_checkpoints.


(Aditya) #58

Actually if it’s a directory, then rm -f
To recursively delete a directory,
Warning it will delete everything in that directoryFor you…

Just type man rm or Do a quick google search for more information


(YIYU SHI) #59

Thanks, I used rm -r .ipynb_checkpoints to get rid of this issue


(YIYU SHI) #60

Hey ecdrid,
Did you complete the whole planet classification problem?
I am wondering how long it took you to run TTA after the 256 size pic training.

Thakns


(Aditya) #61

I did it on AWS server, probably 30-45 mins
Can’t recall exactly


(Krisztian Kovacs) #62

I did the same. I used the python tool that @ecdrid showed in the beginning of the wiki thread to download google images; and built a classifier for plants vs animals. It works pretty well, although some downloaded images couldn’t be opened.

If anyone want to see the Jupyter notebook with my code, it’s here on github.