By snapshotting every detail of every run of a model, dotscience enables you to visualize model behaviour and optimise performance.
100% completely framework agnostic.
Dotscience is totally framework agnostic: use it as your Jupyter environment, or to run any command line scripts for that matter.
Rapid, iterative & evidence-based model development.
Dotscience automatically captures all your model’s input parameters and output metrics, such as accuracy scores.
It snapshots the exact version of the code you ran, as well as the exact version of any data read and written. All this information is stored with the model’s runtime details for perfect reproducibility.
Make better decisions about your experiments.
The Dotscience leaderboard enables you learn from all your model runs. It plots your runs’ performance over time, as well as those of your collaborators.
Interactive visualizations show the effect of any given parameter on any summary statistic produced by the model. This enables you to make an informed decision about the next experiment to try.
A quick primer on dotscience.
Work wherever you want to work.
If you’ve got a compute environment that works for you, simply point dotscience at your hardware.
We’ll track your models where they run, whether its a local machine, a VM, an ML rig or your cluster.
Dotscience versions your data wherever it lives, too: in object storage, like S3, in Spark, or just on your laptop.
No pipeline specs, no data transfer, no walled gardens.
Infrastucture without the devops overhead.
If you want to use the latest top-spec compute resources, dotscience can provision and manage them for you.
We’ll swap out the compute underlying your Jupyter or command-line environment with a cloud instance or cluster of your choice, so you can keep on working in the way that suits you.
For distributed model training, our one-click-cluster will stand up a dedicated Kubernetes cluster and provide an easy-to-use interface. This puts the power of elastic scaling and load balancing into the hands of data science teams, without the need for specialised devops support.
… and any other framework or cloud.
Push and pull lightweight snapshots of code and data with a git-like version control system. Unlike git, Dotscience works on data at scale and is designed for data science.
Stand on the shoulders of giants.
Users can collaborate on projects in Dotscience. View details of your collaborators’ runs & access their outputs. No-one needs to reproduce work or waste money recomputing.
The Dotscience leaderboard shows runs of the same model submitted by multiple users, so you can see your team’s progress and identify promising model versions.
Sign up for our beta
Our beta testers get early access to dotscience for free. Not only can you help shape the product, you can start using it right now in your workflows.