Scribd
Upload
27 views

AI Updated Debug Tips

The document provides instructions for installing PyTorch on Mac, Linux and Windows. It states that PyTorch can now be installed on Windows using conda. For Mac and Linux, it recommends using conda to install PyTorch=0.1.12. For Windows, it recommends using conda to install pytorch-cpu. It also discusses potential issues if installing the GPU version without the proper driver and provides links for additional installation options and resolving errors.

Uploaded by

Nguyễn Quang Trung
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
27 views

AI Updated Debug Tips

The document provides instructions for installing PyTorch on Mac, Linux and Windows. It states that PyTorch can now be installed on Windows using conda. For Mac and Linux, it recommends using conda to install PyTorch=0.1.12. For Windows, it recommends using conda to install pytorch-cpu. It also discusses potential issues if installing the GPU version without the proper driver and provides links for additional installation options and resolving errors.

Uploaded by

Nguyễn Quang Trung
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 5

How to install PyTorch

With the recent launch of PyTorch support for Windows, the library is supported for Mac, Linux and
now Windows. Please use the following to install PyTorch:

Mac and Linux:

conda install pytorch=0.1.12 -c soumith

You also have pip and building from source as options: https://pytorch.org/previous-versions/

Windows:

conda install -c peterjc123 pytorch-cpu

https://github.com/peterjc123/pytorch-scripts

You can also install the GPU versions but If you do not have a Nvidia driver it’s highly advised to use
CPU versions only so that it installs correctly. If you install the GPU version without the necessary
driver you will most likely see a DLL load error and it can be resolved by installing the CPU version.
In addition, PyTorch recently has moved to v0.4.0. The new version has caused a bug in the Doom
related code and it will show as follows when using the newer version of PyTorch:

TypeError: multinomial() missing 1 required positional arguments: "num_samples"

This can be resolved by using the following solution or downgrading to a PyTorch version that is <
0.4.0:

https://www.udemy.com/artificial-intelligence-az/learn/v4/questions/4319834

To install previous versions of PyTorch you can use the directions here for both conda, pip and building
from source: https://pytorch.org/previous-versions/
How to install Kivy

It’s advisable to install Kivy with the development version for each OS. The development version can
be found within the installation instructions, for example: https://kivy.org/docs/installation/installation-
osx.html.

Using Homebrew with pip¶


You can install Kivy with Homebrew and pip using the following steps:
1. Install the requirements using homebrew:
$ brew install pkg-config sdl2 sdl2_image sdl2_ttf sdl2_mixer gstreamer

2. Install Cython and Kivy using pip:


$ pip install Cython==0.26.1

• To install the development version, use this in the second step:


$ pip install https://github.com/kivy/kivy/archive/master.zip

The development version can be found for each OS. Remember to first install the dependencies.

Windows:

• https://www.udemy.com/artificial-intelligence-az/learn/v4/questions/4544432

• https://kivy.org/docs/installation/installation-windows.html#nightly-wheel-installation

Linux:

• https://kivy.org/docs/installation/installation-linux.html#installation

If you have any trouble with the installation you can also see the modified setups provided by students
in the course that have been successful installing the packages:

Modified Setups with installation instructions:

Mac
• https://www.udemy.com/artificial-intelligence-az/learn/v4/questions/4084834

Ubuntu

• https://www.udemy.com/artificial-intelligence-az/learn/v4/questions/3862592

Windows
Install Gym and Doom

You can find a walkthrough for gym here: https://medium.com/@thelastalias/simple-setup-of-openai-


gym-on-macos-287cf0194ac2

Due to changes with Gym the recommended version to use is v0.7.4. You can install the specific
version with:

pip install gym==0.7.4

In addition it is highly recommended to use a VM if you are on Windows and to install with Ubuntu.

Install the dependencies first:

OSX:

brew install cmake boost boost-python sdl2 swig wget

Ubuntu:

apt-get install -y python-numpy python-dev cmake zlib1g-dev libjpeg-dev xvfb libav-tools xorg-dev
python-opengl libboost-all-dev libsdl2-dev swig

https://github.com/openai/gym#installing-everything

For those not running Python 2.7 gym states that:

We currently support Linux and OS X running Python 2.7 or 3.5. Some users on OSX + Python3 may
need to run

brew install boost-python --with-python3

Install Gym:

pip install gym==0.7.4


Install Doom:

pip install ppaquette-gym-doom

If you run into the non-zero exit installation bug you can try the following modified setups:

Ubuntu

• https://www.udemy.com/artificial-intelligence-az/learn/v4/questions/2695950

Mac

• https://www.udemy.com/artificial-intelligence-az/learn/v4/questions/4088774

You might also like