neptune.ai Pros and Cons

We have been using Neptune for more than a year and it fulfills all our needs in terms of experiment tracking. Neptune is also easy to integrate and use, making it an easy decision to go for. Review collected by and hosted on G2.com.

Currently Neptune lacks functionalities (for our usage) for data monitoring and model versioning. Nevertheless, we got nice feedback that these topics are in their roadmap. I'm looking forward to see the next steps. Review collected by and hosted on G2.com.

Neptune stores all the runs’ results that later can be gathered into a single dataframe. It is extremely helpful to have all the experiments in one, accessible place. This, combined with run tagging, creates a powerful combination for results analysis.

I recommend neptune for ML projects. It is very beginner friendly and easy to introduce in any ML framework. I’ve successfully used it with keras, pytorch, and optuna. I think it is a great tool that is a staple when it comes to experiment management. Review collected by and hosted on G2.com.

- You cant't add notes no neptune runs which requires separte tool for writing notes

- GUI compare runs at certain time stamp (run A after 10th epoch vs run B after 20th epoch)

- no nan values handling! Review collected by and hosted on G2.com.

Super easy to use and integrates neatly into our model experiment lifecycle. We deploy very often and across several different projects, and without a proper organization tool things can get messy quite quickly. Neptune has helped us keep everything neatly organized, easily reproducable, and helps keep everyone on the team up-to-date with the latest advancements. First thing we do whenever we start a new project is to ensure we have Neptune tracking set up, which helps us iterate quickly.

The main draw for us were the customizable dashboards for experiment comparison, as we have many metrics to keep track of, and we haven't found an alternative that can satisfy our customization needs like Neptune. Having all relevant metrics grouped together and easily comparing several experiments has been great. Review collected by and hosted on G2.com.

Some more convenience features would be great. For example the ability to replicate custom dashboards across projects, or turning custom views directly into new dashboards. Also something like being able to batch add several columns using regex would be super useful. Review collected by and hosted on G2.com.

Neptune is a tool that manages machine learning experiments. Wehn you are doing hyperparameter optimization and prototyping, it is important to have a tool that keep track of the different experiments. Neptune can do exactly that.

It is easy to analyze a single run and e.g. check how the loss is progressing. You can also compare runs to see what changed and select the best. Uploading model checkpoints can help you to go back to specific runs even a long time later Review collected by and hosted on G2.com.

It takes some time to learn how to make the most out of neptune. Review collected by and hosted on G2.com.

It keeps improving

Easy to integrate into projects

Customer support team is responsive and very considerate of suggestions Review collected by and hosted on G2.com.

At the momemnt I am very happy with neptune. It has improved a lot over time. Review collected by and hosted on G2.com.

We are Neptune.ai customers since 2021 and we had a great experience with the platform so far.

Our main use is experiment tracking of a large number of runs for several different projects, where multiple researchers and engineers contribute to each project.

The Runs view is really what makes it superior compared to every other experiment tracking platform. The way we can deal with hundreds of hyperparameters and metrics across thousands of runs really makes it the perfect experiment tracking plaform.

Integration with Lightning and Keras is extremely easy, and customization is flexible when necessary. We love the native tracking of hardware resources.

As users of the on-prem version, deployment on K8s was always smooth and well supported by the extremely competent support staff. We have seen the tool evolving in these years, becoming better and better at every release. Review collected by and hosted on G2.com.

Honestly we don't have much to report and we look forward to seeing the Report functionality being introuced soon! Review collected by and hosted on G2.com.

The most valuable thing for us is the ability to upload a wide range of data types, such as summary plots at the end of a run or generated samples throughout model training. It allows us to use Neptune to present and share results with colleagues, all in one place, without having to manage files locally. Before Neptune, we used to log all the experimental results to a file located on our server. It was difficult to visualize, share and discuss results. We also struggled to keep track of different versions of experiments. So, we would absolutely recommend Neptune. Review collected by and hosted on G2.com.

The integrations with other tools like PyTorch Lightning are occassionally buggy, but this is a minor issue. Review collected by and hosted on G2.com.

Neptune has great integration for several frameworks, which makes working with it seamless.

UI is uncluttered and usable. Review collected by and hosted on G2.com.

Capabilities of creating dynamic plots are there but could be more sophisticated. Review collected by and hosted on G2.com.

Being able to track a ton of metrics and easily share them with teammates. The UI is very smooth and responsible, even when using large image datasets. It was also very easy to integrate within pytorch lightning. The Neptune-AI is fantastic and replies very quickly whenever we need support. Review collected by and hosted on G2.com.

Would be great to have more plotting options in UI (e.g., bar plots or histograms) Review collected by and hosted on G2.com.

The python sdk is easy to use, the UI is fast, and the project is actively maintained, new features are added regularly without breaking changes. The enterprise pricing plan is also cheaper than alternatives. Review collected by and hosted on G2.com.

It's missing a couple quality of life features Review collected by and hosted on G2.com.