If you are like me who does Machine Learning using a think client laptop then this post might serve you. I use my MacBook Pro 2015 to do most of the coding/debugging and data exploration. While, all the heavy computation happens on GPUs in the remote server. Ideally, I wanted a platform with both remote folder mounting or explorer options with a decent IDE that support python programming. I started with Spyder IDE because of its simplicity. Besides, it also has a good support for visualization of native python data types and some advanced data types like numpy and pandas arrays. But, Spyder’s debugging it still not that user friendly. I also tried VSCode editor after installing the “remote-SSH” extension from their applications marketplace. It was easy to setup but the fact that it was abruptly closing SSH sessions and asked to restart the editor constantly frustrated me. Not to mention, the debugger was still not that advanced in VSCode too.
Pycharm is a relatively well built IDE for python programming. Also, I found that their debugging experience is much better than other python IDEs. I recently found out that they also had support for remote programming. I decided to try it and so far I have found it to be reasonably good. Here, I want to share my steps for setting it up in Pycharm.
First you need to create a new project and select the remote ssh interpreter as the option. Also select the remote project location.
Then go to Tools -> Deployment -> Configuration and setup:
<folder from remote machine>
Mappings: ( so when you press the run button Pycharm will run files remotely)
- Local path:
<folder on your laptop>
- Deployment path: /
- Local path:
Also in Tools -> Configuration set it to upload only when
Windows + Sis pressed.
Note: You can make multiple configurations for the same project and easily deploy the same project on any remote machine.
With this, I can run code in my remote machine with files available in both the remote and local machines. I set it up so that the files from local machine are uploaded to the remote machine and not both ways. So now when I save huge model checkpoints, images from dataset, predictions or log files, etc. they only exist on remote machines and are not pulled to the local machine. I can still access any remote files from Tools -> Deployment -> Browse Remote Host.
Now my final setup looks like this:
[ The left file explorer is for my local folder and the right one for remote machine]
A common error: Remote deployment settings not found!!:
Pycharm for some reason complains with the error message saying remote deployment settings for remote debugger not found. You can fix that by adding path mappings from project interpreter settings tab.
A word of warning, if you are tryin to debug in pycharm then rather than always showing all the variables only show the variables you need to watch.
Hope this helps. Cheers !