fastai

The fastai library simplifies training fast and accurate neural nets using modern best practices. See the fastai website to get started. The library is based on research in to deep learning best practices undertaken at fast.ai, and includes "out of the box" support for vision, text, tabular, and collab (collaborative filtering) models. For brief examples, see the examples folder; detailed examples are provided in the full documentation. For instance, here's how to train an MNIST model using resnet18 (from the vision example):

untar_data(MNIST_PATH)
data = image_data_from_folder(MNIST_PATH)
learn = ConvLearner(data, tvm.resnet18, metrics=accuracy)
learn.fit(1)

Note for course.fast.ai students

If you are using fastai for any course.fast.ai course, please do NOT install fastai from pip or conda using the instructions below; the instructions below are for fastai v1, but the courses use fastai 0.7. For the courses, you should simply follow the instructions in the course (i.e. clone this repo, cd to it, and conda env update), and the notebooks will work (there is a symlink to old/fastai/, which is fastai 0.7, in each course notebook directory).

Note: If you want to learn how to use fastai v1 from its lead developer, Jeremy Howard, he will be teaching it in the Deep Learning Part I course at the University of San Francisco from Oct 22nd, 2018.

Conda Install

To install fastai with pytorch-nightly + CUDA 9.2 simply run:

conda install -c pytorch -c fastai fastai pytorch-nightly cuda92

If your setup doesn't have CUDA support remove the cuda92 above (in which case you'll only be able to train on CPU, not GPU, which will be much slower). For different versions of the CUDA toolkit, you'll need to install the appropriate CUDA conda package based on what you've got installed on your system (i.e. instead of cuda92 in the above, pick the appropriate option for whichever toolkit version you have installed; to see a list of options type: conda search "cuda*" -c pytorch).

NB: We are currently using a re-packaged torchvision in order to support pytorch-nightly, which is required for using fastai.

If your system doesn't have CUDA, you can install the CPU-only torch build:

conda install -c pytorch -c fastai fastai pytorch-nightly==1.0.0.dev20180928=py3.6_cpu_0

PyPI Install

Please execute the following 3 steps:

First install the nightly pytorch build, e.g. for CUDA 9.2:
```
pip install torch_nightly -f https://download.pytorch.org/whl/nightly/cu92/torch_nightly.html
```
If you have a different CUDA version find the right build here. Choose Preview/Linux/Pip/python3.6|python3.7 and your CUDA version and it will give you the correct install instruction.

Next, install a custom torchvision build, that is built against torch_nightly.

pip install --index-url https://test.pypi.org/simple/ --extra-index-url https://pypi.org/simple/ torchvision==0.2.1.post2

Now you can install fastai. Note, that this is a beta test version at the moment, please report any issues:
```
pip install fastai
```
On some setups, the last pip command still tries to get torch-0.4.1. If that happens to you, you may try:
```
pip uninstall torchvision fastai
pip install --no-deps torchvision
pip install fastai
```

Is My System Supported?

Python: You need to have python 3.6 or higher
Operating System:

Since fastai-1.0 relies on pytorch-1.0, you need to be able to install pytorch-1.0 first.

As of this moment pytorch.org's pre-1.0.0 version (torch-nightly) supports:

Platform GPU CPU

linux binary binary

mac source binary

windows source source

Legend: binary = can be installed directly, source = needs to be built from source.

This will change once pytorch 1.0.0 is released and installable packages made available for your system, which could take some time after the official release is made. Please watch for updates here.

If your system is currently not supported, please consider installing and using the very solid "v0" version of fastai. Please see the instructions.

Platform	GPU	CPU
linux	binary	binary
mac	source	binary
windows	source	source

Developer Install

First, follow the instructions above for either PyPi or Conda. Then remove the fastai package (pip uninstall fastai or conda uninstall fastai) and replace it with a pip editable install:

git clone https://github.com/fastai/fastai
cd fastai
tools/run-after-git-clone
pip install -e .[dev]

You can test that the build works:

jupyter nbconvert --execute --ExecutePreprocessor.timeout=600 --to notebook examples/tabular.ipynb

Please refer to CONTRIBUTING.md and develop.md for more details.

Copyright

Copyright 2017 onwards, fast.ai, Inc. Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. A copy of the License is provided in the LICENSE file in this repository.