Note: this is for FastAi 0.7 which is pretty much replaced at this point by the fully rewritten FastAi 1.0
After spending 12 hours here is how I managed to have fastai 0.7 environment and Deep Learning lesson1 jupyter notebook working.
Please familiarize yourself with the following post and the associated issues that may appear before attempting the steps that worked for me.
- Install Visual Studio Community 2017 C++ Desktop Development edition. (C++ dependencies needed for python 3)
- Install Anaconda with latest version of Python 3.7. Install Git for W10. Restart.
- In Anaconda Prompt clone the fastai git.
- Add fiona fix to environment.yml in fastai folder. This worked for me.
Howto: installation on Windows
Can also try this - but I have not tried it:
Installing fastai on Windows 10: Failed building wheel for fiona
- In Anaconda Prompt: go to fastai clone and execute conda env update. Then activate fastai.
- If pip error: update pip from 10.1 to 18.
7: same as Jeremy’s HowTo:
- make symlink to old\fastai (so that you are using fastai 0.7)
mklink /d fastai …\old\fastai
- if kernel not found when running jupyter notebook (i had that problem): then change the path in the Anaconda3\envs\fastai\share\jupyter\kernels\python3\kernel.json to Anaconda3/envs/fastai/python
- when running lesson1.pynb if module not found error:
check in your fastai folder to see if the old\fastai folder has all files in it.
mine did not, so i manually copied the old\fastai folder from the fastai git to my fastai folder. (somehow files get lost in that folder during conda env update??)
Any comments regarding these steps and what worked for you, or did not, are welcome.