Lesson 1 In-Class Discussion

I see your name on line 291

@jeremy would be really cool if you gave a tutorial about your development tools and workflow:

  1. The IDEs you use (and what plugins, looks like you like vim)
  2. Your notebook plugins
  3. How you debug, prototype and handle working with AWS vs. locally

Anything else you found useful about your workflow.

6 Likes

Crestle is too slow here.

That’s a different guy, we just share the name. My forum id is merajat. :slight_smile:

Tons of people trying to use it at the same time. Not surprising there is a slow-down.

2 Likes

Rajat, we have to ping Jeremy after class.

yeah we will do that, thanks

I’ve had to rewrite some of the code to make it work in python 3.5. Some format strings are a little weird. Particularly in ‘fastai/conv_learner.py’ and 'fastai/learner.py’
Anything that looks like

f'{self.foo}_{self.bar}'

I’ve needed to change to

'{}_{}'.format(self.foo, self.bar)
2 Likes

What’s the intuition behind the fact that linear functions stringed together with non linear functions can approximate any type of function. Meaning why can deep learning a universal approximation function?

what do in crestle. I created account before today so I did git clone https://github.com/fastai/fastai.git. Now what do I have to follow up to upon this jupyter notebook jeremy has opened?

Yes.
then go to fastai/courses/dl1/lesson1.ipynb

You can use python version 3.6 and avoid having to rewrite them

1 Like

I’m not using Crestle, but I had an issue with:

os.makedirs('/cache/tmp', exist_ok=True)

I needed to create the /cache directory (in linux) and chmod 777 /cache

sudo mkdir /cache
sudo chmod 777 /cache
1 Like

How do we refresh Crestle to get the latest fast.ai course data

resnet34, does that mean vgg has fallen out of favour?

If you ignore that line. It may just work.

1 Like

Old fashion (-:

2 Likes

Since I don’t have disk space in my /tmp partition, I created a new folder in a larger disk, and replaced all the ‘os.makedirs’ lines in the notebook

PATH = "/media/NEMO/DATA/fastai/data/dogscats/"
os.makedirs(PATH, exist_ok=True)
os.makedirs('/media/NEMO/DATA/fastai/data/dogscats/models', exist_ok=True)
# after the ln -s lines
os.makedirs('/media/NEMO/DATA/fastai/cache/tmp', exist_ok=True)
!ln -fs /cache/tmp {PATH}

can anyone help with this error: Cant find a solution:

9 from functools import partial
10 from collections import Iterable, Counter
—> 11 from isoweek import Week
12 from pandas_summary import DataFrameSummary
13

ImportError: No module named ‘isoweek’

@Jeremy
Is it always better to use a library- fastailib or sklearn for pre-processing the data?
Labelling the classes, etc. Or is writing functions suggested?