Hi Fastai members,
I am new to Machine Learning and came across this wonderful website.
If anyone is starting or started recently, we can collaborate and help each other in this Machine Learning journey.
Thanks,
Murali
4 Likes
I’m here
Now I’m on lesson 4
I’m currently past Lesson 3.
Will be happy to help/collaborate.
2 Likes
Hi @Krieker @Brainkite, i have actually in the process of setting up the software to execute the code. Which one you guys are using? Crestle/Paperspace/AWS?
Any suggestions for me.
I am currently going to start week 3 of the course…
How do you solve 'file name too long error ’ while creating the ImageDataBunch object?
Help please…
@cap_rogers how long are your file names? If they’re excessively long is it possible for you to rename them if needed? (You can do this in Jupyter no need to go one by one)
When I printed out the file names,they seemed fine. For eg. ‘PosixPath(’/root/.fastai/data/planet_sample/train/train_11988.jpg’)’. This was one of the files. The path for my training folder is ‘PosixPath(’/root/.fastai/data/planet_sample/train’.
This is the error I am getting
Google Colab is pretty easy to get started and it’s free. Check it out.
Hey all,
I am currently on lesson 2 but am getting pretty excited to try out my own image classification project before I move on to lesson 3.
Anyone tried any personal projects so far?
1 Like
Hi @jianjye, thanks for the suggestion. I was thinking to go with the Crestle. Because it does not need any installation. And i can straight away start working on the Lesson 1 code.
Anyhow i want to know how much time it will take for me to start working on Lesson 1 if i use Google Colab. Do you have any installation steps?
This is the link, i should follow right for Colab - https://course.fast.ai/start_colab.html .
Yup. There wasn’t much to setup really, unless you want to save your models permanently then you might need some lines of extra codes to mount google drive.
I have not used Crestle. If you are happy with it then no reason to switch. In the end they are all Jupyter with minor differences.
@kuruvam I use both AWS and Colab. Think Colab would be better and easier
Hi @kuruvam ,
I started recently and almost done with lesson 1. I have used Google Colab and so far find it easy to use.
Regards,
Chandresh
Hi,
I am also starting lesson 1 and have some error (using Colab as well).
when i try to rename the image files i get the following error, do you know how to solve it ?
Code:
data = ImageDataBunch.from_name_re(path_img, fnames, pat, ds_tfms=get_transforms(), size=224)
data.normalize(imagenet_stats)
Error:
AssertionError Traceback (most recent call last)
in ()
----> 1 data = ImageDataBunch.from_name_re(path_img, fnames, pat, ds_tfms=get_transforms(), size=224)
2 data.normalize(imagenet_stats)
5 frames
/usr/local/lib/python3.6/dist-packages/fastai/vision/data.py in _get_label(fn)
154 if isinstance(fn, Path): fn = fn.as_posix()
155 res = pat.search(str(fn))
–> 156 assert res,f’Failed to find “{pat}” in “{fn}”’
157 return res.group(1)
158 return cls.from_name_func(path, fnames, _get_label, valid_pct=valid_pct, **kwargs)
AssertionError: Failed to find “re.compile(’/([^/]+)_\d.jpg$’)” in “/root/.fastai/data/oxford-iiit-pet/images/beagle_67.jpg”
@Dinas you are missing a + after the d. If you look here: https://www.regextester.com/ it’s a nice way to test regular expressions. Including the ‘+’ before the ‘.jpg’ will select the proper filename.
@muellerzr Thanks a lot !!
and it’s a cool tool for reg exp tests, thanks for sharing 
I was creating an ImageDataBunch for a dataset of flower images. The training folder had about 19K images and its total size was around 900MB. I used colab for it.But the RAM(about 25GB) seems to run out quickly while create the ImageDataBunch object. I tried it for multiple bs values ranging from 16 to 64. I am not able to understand why this is happening. Can anyone shed some light on this?
Hello, fellow July classmates! I also started this month, and I’m about halfway through Lesson 2. I’m using Colab, and I found it reasonably easy to set up just by following the instructions.
(I’ve been fooling around making my own models, using images downloaded from Google Image Search. There’s a Firefox add-on called “Google Images Downloader” that really helps, and then I manually comb through the image files to remove duplicates and weird images, then upload them to my Google Drive and let the Jupyter Notebook access them from there.)