It was confused experience overall for the first time using FastAI. Currently, If you wanted to learn some courses in its source code, there would be only version 0.7 which is able to support you. Unfortunately, You need to ignore the fastai 1.0.x
in this case.
So I will show you how to install FastAI on Window 10. At the first name, I found it very strange and right now I am used to it.
-
- Make sure you have a Nvidia GPU installed and the drivers are set up already.
-
- If you are a developer, Git should be installed already.
-
- Clone FastAI from Github:
git clone https://github.com/fastai/fastai.git
- Clone FastAI from Github:
-
- Go to the FastAI folder and type:
conda env update
- Go to the FastAI folder and type:
-
- Activate a designed enviroment with a command:
activate fastai
- Activate a designed enviroment with a command:
-
- Add an widget to our jupyter:
jupyter nbextension enable --py widgetsnbextension --sys-prefix
- Add an widget to our jupyter:
-
- This part is quite strange and miserable. You have to to map fastai file in the courses to older version 0.7 outside.
cd courses\dl1
del fastai
mklink /d fastai ..\..\old\fastai
cd ..\..
-
- Run your notebook and checkout whether fastai was installed or not. Command with
jupyter notebook
- Run your notebook and checkout whether fastai was installed or not. Command with
-
- You need to ensure the correlation between versions of Pytorch and FastAI. Otherwise, you would get some errors about lacking API of Pytorch while running FastAI.
fastai 1.0.x
matchespytorch 1.0
fastai 0.7.x
matchespytorch 0.4.1
So please be vigilant about this.
Some experiments:
- Image Recognition
- CNNs
- Overfitting
- Embeddings
- NLP
- RNNs
- CNN Architectures
- Object Detection