We Raised $8M Series A to Continue Building Experiment Tracking and Model Registry That “Just Works”

Read more

The Best Comet.ml Alternatives

Comet is one of the most popular tools used by people working on machine learning experiments. It is a self-hosted and cloud-based meta machine learning platform allowing data scientists and teams to track, compare, explain, and optimize experiments and models.

Comet offers a Python library to allow data scientists to integrate their code with Comet and start tracking work in the application. As it’s offered both cloud-hosted and self-hosted, you should be able to manage ML experiments of your entire team.

Comet is converging towards more automated approaches to ML, by adding predictive early stopping (not available with the free version of the software) and announcing neural architecture search (coming in the future).

Some of the Comet most notable features include:

  • Enables sharing work in a team and provides user management
  • Integrates with a bunch of ML libraries
  • Let’s you compare experiments—code, hyperparameters, metrics, predictions, dependencies, system metrics, and more
  • Let’s you visualize samples with dedicated modules for vision, audio, text, and tabular data

And although Comet is a great solution, no tool is perfect for everyone (at least I haven’t heard of such a tool 🙂).

It may be missing some points that are crucial to you and your team. It could be:

  • The lack of certain features: like the inability to log plotly/bokeh plots’
  • Your personal preferences: like the view of the charts in the UI, or the comparison table, or other features specific to your use case;
  • Pricing: maybe you prefer a tool that works in a usage-based pricing model, or that’s open-source;
  • Scalability: maybe it doesn’t satisfy your needs in terms of scalability, and your team runs more and more experiments every month.

Anyhow, there are many other tools available, and to help you find the right fit, we present a list of the best Comet alternatives.

1. Neptune

Neptune is a metadata store for MLOps with the main focus on experiment tracking and model registry. It was built for research and production teams that run a lot of experiments. Its main goal is to make their lives easier and let them do what they really want to do, which is ML (and not manually inserting numbers into excel sheets). 

So, Data Scientists and ML Engineers can use Neptune to log, store, organize, display, compare, and query all your model-building metadata in a single place. This includes metadata such as model metrics and parameters, model checkpoints, images, videos, audio files, data versions, interactive visualizations, and more.

It’s also easy to later share this tracked metadata within the team. Neptune allows you to create workspaces for projects, manage user access, and share links to dashboards with both internal and external stakeholders. 

Neptune – key features:

  • Neptune allows you to log and display model metadata in any structure you want. Whether it is a nested parameter structure for your models, different subfolders for training and validation metrics, or a separate space for packaged models or production artifacts. It’s up to you how you organize it.
  • And then, you can create custom dashboards to combine different metadata types in a preferred way.
  • The pricing of Neptune is usage-based. There’s a comparatively low fixed monthly fee for the whole team (no matter how many people it counts), but other than that, you just pay for what you use.
  • The app can handle thousands of runs and doesn’t slow down when you use it more and more. It scales with your team and the size of your project.

If you want to see Neptune in action, check this live Notebook or this example project (no registration is needed) and just play with it. 

Neptune vs Comet

Neptune and Comet ML both operate in the same market space, however, their feature proposition varies based on both technical and non-technical offerings.

Both of these tools are proprietary software and offer hosted and on-premise setups with different pricing options available to choose from. Neptune offers a usage-based pricing model apart from other fixed offerings in which you can scale up and down depending on the number of experiments and features you’re using. Comet on the other hand has a strict 1-dimensional pricing structure. 

There are also some differences in the features. E.g. If you’re looking for a tool that allows you to track your dataset versions then Neptune is your pick because Comet lacks this feature.

Learn more

💡 See a deep dive into Comet and Neptune differences.

2. TensorBoard

TensorBoard-experiment-tracking
Example dashboard in TensorBoard | Source

TensorBoard is a visualization toolkit for TensorFlow that lets you analyze model training runs. It’s open-source and offers a suite of tools for visualization and debugging of machine learning models.

It allows you to visualize various aspects of machine learning experiments, such as metrics, visualize model graphs, view tensors’ histograms, and more.

It’s a great tool if you’re looking for a Comet alternative to visualize your experiments and dig very deep into them.

TensorBoard – key features:

  • You can log experiments for the whole team in a single place
  • Track experiments that are not based on TensorFlow or deep learning in general
  • Backup whole experimentation history
  • Quickly make reports for project stakeholders
  • Integrate tracking system with other tools from the technological stack
  • Visualization features available

TensorBoard vs Comet

If you’re looking for a tool to visualize your ML model metadata for your project then TensorBoard might be the right choice for you. It’s open-source and its visualizations are quite robust but it runs on a local server so you can’t share it with your team members, unlike Comet.

However, TensorBoard gives you a lot of technical options about visualizing your data to an extent to which Comet doesn’t. For example, it offers smoothening over the metric charts and with images offers a steps (epochs) slider to compare results for different epochs.

3. Guild AI

Guild-AI-experiment-tracking
Example dashboard in Guild AI | Source

Guild AI is an open-source tool used by machine learning engineers and researchers to run, track, and compare experiments. With Guild AI, you can leverage your experiment results to build deeper intuition, troubleshoot issues, and automate model architecture and hyperparameter optimization.

Guild AI is cross-platform and framework independent — you can train and capture experiments in any language using any library. Guild AI runs your unmodified code so you get to use the libraries you want. The tool doesn’t require databases or other infrastructure to manage experiments — it’s simple and easy to use. 

Guild AI – key features:

  • Experiments tracking: any model training, any programming language
  • The automated machine learning process
  • Integrated with any language and library
  • Remote training and backup possibility
  • You can reproduce your results or recreate experiments

Guild AI vs Comet

If you’re looking for an open-source experiment tracking tool that doesn’t require you to change your code then Guild AI would be a good option. As mentioned above you’d have the features of visualization, hyperparameter tuning, and a host of other features. However, since it’s open-source it misses out on key features geared towards scalability and working in teams.

If you’re a team and your priority is sharing the results and scaling to a large number of experiments then unfortunately Guild AI is not suited for it.

May interest you

4. MLflow

MLflow-experiment-tracking
Example dashboard in MLflow | Source

MLflow is an open-source platform that helps manage the whole machine learning lifecycle that includes experimentation, reproducibility, deployment, and a central model registry. 

MLflow is suitable for individuals and for teams of any size. 

The tool is library-agnostic. You can use it with any machine learning library and in any programming language

MLflow comprises four main functions:

  1. MLflow Tracking – an API and UI for logging parameters, code versions, metrics, and artifacts when running machine learning code and for later visualizing and comparing the results
  2. MLflow Projects – packaging ML code in a reusable, reproducible form to share with other data scientists or transfer to production
  3. MLflow Models – managing and deploying models from different ML libraries to a variety of model serving and inference platforms
  4. MLflow Model Registry – a central model store to collaboratively manage the full lifecycle of an MLflow Model, including model versioning, stage transitions, and annotations

MLflow vs Comet

MLflow is another open-source alternative to Comet. It offers the same functionality that Comet does and it also scales well on Big Data. If your team utilizes Apache Spark then MLflow would be a good pick since it scales well with Spark to offer model tracking and model registry on Big data.

But Comet comes with user management features and allows for sharing projects within the team—something that is missing in MLflow. It also offers both, hosted and on-premises setup, while MLflow is only available as an open-source solution that requires you to maintain it on your server. 

5. Weights & Biases

WandB - experiment tracking
Example dashboard in W&B | Source

Weights & Biases a.k.a. WandB is focused on deep learning. Users track experiments to the application with Python library, and – as a team – can see each other’s experiments. It allows them to record experiments and visualize every part of the research. WandB is a hosted service allowing you to backup all experiments in a single place

Wandb – key features:

  • Deals with user management
  • Great UI allows users to visualize, compare, and organize their runs nicely
    Sharing work in a team: multiple features for sharing in a team
  • Integrations with other tools: several open-source integrations available

WandB vs Comet

WandB is a closed-source solution alternative to Comet. It offers very similar features as Comet does barring the integrations with different languages and frameworks. For example, WandB has integrations for fastai and Catalyst for model training while Comet doesn’t. Pricing models of both are also quite different based on requirements, you can check Comet’s here and WandB’s here.

See also

💡 Deep dive into Wandb and Neptune differences.

6. Sacred

Sacred is an open-source tool, developed at the research institution IDSIA (Swiss AI lab). Sacred is a Python library that helps configure, organize, log, and reproduce experiments.

The tool offers a programmatic way to work with configurations. The concept of Observer allows you to track various types of data associated with an experiment.

Moreover, Sacred has automatic seeding – very useful when there is a need to reproduce an experiment.

Sacred – key features:

  • Best for the individual user since sharing the work in a team is not supported
  • Experiments tracking: any model training
  • Integrations with other tools: not supported
  • Bonus: there are few front-ends for Sacred, so you can pick one that suits your needs best. Take a look at this particular integration with Neptune.

Sacred vs Comet

Sacred is another open-source alternative to Comet. If you’re an individual looking for a simple easy to use experiment tracking tool for models built in Python then Sacred would prove a good option. It comes in a standalone pip package with frontend UI depending on other sources, you can view those sources here

Sacred doesn’t support team collaboration so it’s best suited for individual use. If you’re a team working on production-spec projects then Comet would be a better option as compared to Sacred.

Summary

Finding the right tracking tool for your needs always proves fruitful in any Machine Learning project. It helps you achieve desired results faster than any traditional method. Thus, allocating sufficient time and resources to choosing the right tool will in turn end up saving a lot of time and resources. 

Don’t forget to opt for the one that corresponds to your needs, style of work, and gives you enough flexibility to get the most out of your time.

Happy experimenting with your ML projects!


READ NEXT

Hypefactors Case Study: Metadata Management for Teams that Don’t Run Experiments All the Time

7 mins read | Updated October 25th, 2021

Hypefactors is a technology company that works in the media intelligence and reputation tracking domain. They provide an ML-based Public Relations (PR) automation platform, including all the tools to power the PR workflow and visualize the results.

Hypefactors case study
Hypefactors dashboard | Source: Hypefactors

We spoke to the CTO of Hypefactors, Viet Yen Nguyen who leads the technical teams and is responsible for the technical side of the company. He explains this business as: 

“Apple Inc. is worth approximately 2 trillion but if you go to their warehouses, stores, and sum up everything they own you don’t get to 2 trillion. You may get to maybe 60 billion worth of physical goods.

So where is the 1.94 trillion of value that is reflected in the price of the stock? It is the future prospects, the brand potential, the reputation, the intangible things.

We help teams actually measure and track that.”

Hypefactors case study

Viet Yen Nguyen

CTO at Hypefactors

What is the project about?

The data pipelines at Hypefactors monitor the whole media landscape ranging from their social media pipelines to print media, television, radio, etc, to analyze changes in their customers’ brand reputation. This feat is achieved in two phases:

  • 1Getting the data from everywhere
  • 2Enriching the data with ML-based features 

In order to analyze every form of data ranging from image, text, and tabular, they work on a variety of ML problems:

  • NLP classification
  • computer vision (e.g. segmentation)
  • regression for business metrics

As they train and improve many enrichment models using different ML techniques, this naturally involves running many experiments and articulating ways to store the metadata generated by those experiments.

With competent engineers in the data and AI teams, Viet’s team was able to achieve good results with all the components in the pipeline except for Experiment Tracking. Let’s peek into the problems they faced during the process and what measures they took to solve them.

Continue reading ->
The Best ModelDB Alternatives

Best Alternatives to ModelDB

Read more
Experiment tracking Experiment management

15 Best Tools for ML Experiment Tracking and Management

Read more
Best 7 Data Version Control Tools That Improve Your Workflow with Machine Learning Projects

Best 7 Data Version Control Tools That Improve Your Workflow With Machine Learning Projects

Read more
MLflow alternatives

The Best MLflow Alternatives (2022 Update)

Read more