Aim 3.4 – Remote Tracking Alpha, Sorting & deleting runs

Aim: AI remote tracking

Special thanks to Geoffrey Chen, gormat, sjakkampudi, osoblanco, Sennevs, gloryVine, krstp and others for continuous feedback and help.

Remote tracking [experimental]

This is a milestone release with lots of anticipated new features and one that’s super-anticipated. Brace yourselves, the collaborative Aim is here one commit at a time!!!

### Ensure Aim version 3.4.x
```sh
$ pip install "aim>=3.4.0"
```

### Initialize the remote server
```sh
$ aim init # (optional)
$ aim server --repo <REPO_PATH>
```
Output:
```
> Server is mounted on 0.0.0.0:53800
> Press Ctrl+C to exit
```
### Start the Aim UI
```
$ aim up --repo <REPO_PATH>
```

### Change only 1 line in your training code
Point your repo to the remote location
```py
aim_run = Run(repo='aim://172.3.66.145:53800')# replace example IP with your tracking server IP/hostname
```

Run, delete and archive: Batch and single

Aim Run delete and archive: batch and single

Ability to stack images on the Images Explorer

Aim: Ability to stack images on the Images Explorer

Text Search

Aim: Text search

Trendline on scatterplots

Now you can render a trendline over your scatterplots on Scatters Explorer. We have enabled linear and Loess trendline.

Thanks to sjakkampudi for the help on this one 🙌.

Aim: Trendline on scatterplots

More Images Features

We have added a number of new tools to the Images Explorer

Display Images by original size

Now you can view & compare the tracked images by their original size. This has been a key missing ingredient for effective images comparison.

Aim: Display images by original size

Align by width

You can also align images by their width. This is especially handy when long images are tracked, such as sound spectograms (40px / 1000px). This feature will allow to effectively compare such images.

Aim: Align by width

Images Sorting

You can also apply sorting of images by any tracked available parameter. In this case I have enabled sorting by step in descending order to see the later generated images in my GAN experiments. A lot smoother than the ones above 🙂

Aim: Images sorting

Learn more

Leave a reply

Your email address will not be published. Required fields are marked *