interpret-community
interpret-community copied to clipboard
interpretml
Fastest way to publish the dashboard on web
It’s often requested to publish the dashboard on the web in order to make it available to users (not in a notebook).
Is it possible? If so, what are the best/fastest ways to do that? Would be great if you could share both an on-premises solution (using open source stuff) and an Azure solution.
@lucazav great question! If you want to share the dashboard with others I'm not sure if there is any easy way to do that right now, even in workspace 2.0 in Azure. This sounds like a new feature request. Also adding @rihorn2 and @mesameki who might have some ideas.
@lucazav I just learned that in workspace 2.0 you can share the dashboard with other team members in the workspace, but you can't share it with anyone external to the workspace, so that might at least be one solution for sharing the visualizations.
The dashboard is shown in the Azure ML portal, under the run history for the run on which you upload the explanation to azure. We are working on a way to have a limited dashboard be sharable as html, but it would not have functionality like what-if and ICE plots that require a model. Also we will have to limit the size of the data at some point.
@lucazav I just learned that in workspace 2.0 you can share the dashboard with other team members in the workspace, but you can't share it with anyone external to the workspace, so that might at least be one solution for sharing the visualizations.
Yep, I know. But more often customers want to self-navigate their data through the model interpretability. An easy way to share the dashboard externally will be a great feature adding!
The dashboard is shown in the Azure ML portal, under the run history for the run on which you upload the explanation to azure. We are working on a way to have a limited dashboard be sharable as html, but it would not have functionality like what-if and ICE plots that require a model. Also we will have to limit the size of the data at some point.
@rihorn2 it's a pity not having all the functionalities available for external access. Do you really want to limit them? Are there technological issues?
I think having this dashboard (with all its features) available, for example, as a Flask widget will be a real contribution to the open source community.
It is a technical limitation as to why a shared dashboard would have to be a limited dashboard. The underlying model is a python object, sharing the client-side html and javascript would allow the user to see any data present at the time the html was copied, but you won't be able to make calls to any model or explainer implemented in python. You could deploy your model and any explainer as a service, but a user should be hesitant to share out their deployed service credentials since they will have to pay for all the compute used as people share their dashboard. So I think the plan at this moment is to enable a sharable read-only view of the dashboard.
It is a technical limitation as to why a shared dashboard would have to be a limited dashboard. The underlying model is a python object, sharing the client-side html and javascript would allow the user to see any data present at the time the html was copied, but you won't be able to make calls to any model or explainer implemented in python. You could deploy your model and any explainer as a service, but a user should be hesitant to share out their deployed service credentials since they will have to pay for all the compute used as people share their dashboard. So I think the plan at this moment is to enable a sharable read-only view of the dashboard.
@rihorn2 it's reasonable. Just keep in mind that a lot of customers I know really would like to deploy model and explainer as a service (and pay for them) in order to have a fully functional dashboard they can use to find insights for their business. So such alternative solution may complete your offer. My 2¢