Increasing Transparency and Controllability in Automated Machine Learning

Is there possibility to Integrate ATMSeer a MIT open source project into MindDB Scout UI ,
It may be a great features and differentiator among the existing AutoML products like Rapidminer , Knime ,Auto -Weka, Google AutoML, Watson , Azure ML Studio etc.

Here is link https://github.com/HDI-Project/ATMSeer

2 Likes

Thanks for providing this @Manoj. ATMSeer certainly looks like a good visualization tool for AutoMl. Currently, it works on top of the ATM AutoML system. In the MindsDB Scout we are already providing similar features as ATMSeer (analyze data, analyze the trained model, previews, etc…) also we have visdom used in ligtwood, and constantly upgrading it with new features. If there is some specific feature that you think it’s great to have, please share it with us.

Currently we use Visdom for monitoring the inner-working of model training: https://github.com/mindsdb/lightwood/blob/master/visdom.sh

However the information is rather complex for end users to understand so we haven’t bothered with trying to add it to Scout.

ATMSeer seems to display some of the same information we do with our visdom setup, but obviously more focused around what ATM is doing… I’m not sure adapting it to our machine learning backend (lightwood) is easier than writing something from scratch and/or based on windows.

If it’s easy enough to plug it in and someone wants to make a contribution doing so (or by making the data we display in visdom easier to access by the end user and/or richer) I have no qualms against it, I hardly see as an internal priority atm though. Maybe we’ll come back to it in e.g. 6-12 months.

1 Like