I had several deep learning and NLP classes with Comet. This software is recommended by the class. The overall usage is pretty good, with easy setup and grading, and their intuitive interface.

Great plotting tool, fast speed, and detailed documentation

It is a bit hard to manage when there are too many experiment trails.

It is a good software that helped me keep track of my Deep Learning experiments.

Experiment tracking and collaboration is accessible on this platform Also easy to upload files

Limited file size upload. It throws errors for bigger file size

The reports feature which enabled one to embed different panels

At first the learning process of how the software works was steep.

We have really enjoyed using comet. It's been really easy to setup and makes experiment tracking and gaining insight into the model's performance so much better. The support has also been really good, especially when interacting through the Slack Connect channel. Unfortunately, the business has thought about moving away from the product due to the high cost of licenses.

Comet is really easy to setup and provides really useful visualizations without a ton of boilerplate code. A simple log_asset() or log_metric() call can go a long way into providing insight into how the model is training or performing with comet automatically generating graphs or confusion matrices or viewing images and predictions online.

Occasionally there are inconsistencies with what is offered in the Comet UI and what is available through the python SDK. This was observed when trying to automatically upload and tag models in the model registry. The price is also becoming a factor on whether or not we move away from comet.

I needed a tool that would help me in keeping track of my experiments. I got a whole set of tools that is perfect for my ML research

The ease of integration with existing pipelines and workflow is fantastic. I also love the new panels UI, it cuts down all configuration to a minimum

The only thing which annoys me is the chaos in custom view settings. Most of the time i create some view, and when I want to reuse it later i have to create a whole new copy of it, which leads to constantly increasing the number of views.

Easy to use and integrate, can track data from any device, can also check from mobile phone on the go

Data organisation of old trials could be done better

Comet ML provides an integrated platform to track experiments, log assets/metrics and also provide various visualizations. It is a great tool to maintain and reproduce models

The documentation could be improved and also the panels needs to be improved for accessibility

I have explored bunch of other experimentation platform for ML projects but Comet is my most favourite tool by far as it provides easy integration with major ML frameworks such as Scikit-learn, Pytorch, Tensorflow, Keras, and so on. In addition to that, it supports multiple programming languages.

I don't feel any major features Comet is missing as most of the experimentation in ML projects already fulfilled.

Comet is fairly easy to set up and use, and has the ability to store different types of results. The panel system makes it easy to compare across hyperparameters and the community panels are well developed.

The UI can be slow as more experiments are added an occasionally there are minor bugs which are manageable.

Comet is a good tool that is great in machine learning and deep learning.

sadly, Comet software has no free trial.

Has been seamless.

It is an easy alternative to replace local experiment tracking that makes it convenient to check on live/previous experiments from any device.

Saving model can be a hassle, especially when size becomes large.

Ease of use and usability within a team.

Team size is limited on academic plan makes it hard to organize