iris – deployable machine-learning notebooks
iris /

what's iris?

iris is a simple python notebook that lets you quickly build, share, and deploy machine learning models.

You select a dataset, write a train() function and a predict() function, and then iris provides you with a unique URL to get predictions.

get started

Let's see how easy it is deploy your first model with iris!

step 1
Sign in to your iris profile using GitHub.
step 2
Click the create model button and select a dataset:
Classify iris plants into three species in this classic dataset
721 pokemon with stats and types
step 3
Write a train() function, which gets the dataset as a pandas dataframe df:

and a predict() function, which gets adf of samples from the endpoint and the le and clf returned from train():

step 4
Press cmd + enter to train your model.
step 5
Make POST requests to get predictions!



iris provides you with easy access to many clean datasets at the click of a button (or upload your own!).

simple and minimal

iris is only two functions, train()and predict(). No more lines-upon-lines of proprietary solutions.

automatic pickling, unpickling, and cloud storage

iris automatially pickles, unpickles, and stores your models' files in the cloud, so you no longer need to fix difficult-to-debug pickling errors.

immediate deployments

Every model you write in iris is instantly deployed with a REST endpoint so you can integrate it into your app.

shareable and reproducible

Each model gets its own shareable page, with a static environment, so you can share and reproduce the results.

frequently asked questions

Can this be used in production?

Absolutely. Here's everything we've built in iris to make it production-ready for your app.

Fast predictions

We've made our predictions just as fast as deploying your own model microservice in your own cluster 🏎.

High availability

iris is committed to high uptime. We've built our infrastructure on Kubernetes and Google Cloud.

Secure code and data storage

iris employs the latest security measures to make sure no one but your team has access to your data and code.

Transparent and shareable

All of the model code and data you write with iris is completely visible in an easy-to-read notebook. Your code isn't hidden behind some magical API.

Why not Jupyter notebooks?

They're not deployable

You can't deploy a Jupter notebook into a server or microservice without a lot of work.

iris automatically gives you a unique endpoint to share or integrate into your app.

Duplicated effort

Every time you write a Jupyter notebook, you need to write code to load your data, split the data into test/train sets, calculate the accuracy and confusion matrix, and then search around for your train/predict functions to copy into a script for deployment.

iris takes care of all of this for you so you can just click which dataset you want, and write your train()predict() functions.

They encourage proprietary solutions

iris notebooks have the same structure across all models, so you know exactly where where training and where predictions occur, no matter which model you're looking at.

Can I upload my own datasets?

Yes! You can upload your own CSV and use it within iris.

Can I make my models and datasets private?

Yes! If you upgrade to the Developer plan, you can make both models and datasets private

Can I take my models out of iris later?

Yes--iris is just python. You can easily copy your model code out of iris later if you choose to.