[New Integration]: An App to Democratize Machine Learning using MindsDB 👍👍

[parthiv11]

Is there an existing integration?

• [X] I have searched the existing integrations.

Use Case

As we know MindsDB has focused on Democratize Machine Learning. With this idea in mind had made a very simple and clean user interface on their website, through which anyone can utilize the power of ML. My idea is to make an app with the magic of ML powered by MindsDB API at the backend, with this app mindsDB will reach more users and make them aware of the Power of ML.

Just think how easy it would become to get to know about data and also train a model based on that data and query it in according our use-case 👍 👍

Motivation

To democratize Machine Learning, first of all, it should be made accessible to everyone and also easy to use interface. Don't you think it would be great to train our data using an app with some checkboxes, dropdowns, and clicks of buttons.

Also, app can show graph based on input data according to which type of graph the user wants. 👍👍❤❤❤✌✌

Implementation

I am not a professional designer just a student, but i have tried to make some design which will at least an idea how our app will change the ML world..

Talking about backend mindsDB API best according to me

Anything else?

If you want this app or You have liked this idea then upvote me with 👍 sign. any suggestion then please feel free to drop here

👍

[andy-brainome]

To reflect and summarize what I just read in order to judge my understanding - the proposal is to build an app store/play store style App that interfaces with a mindsdb service, user selects data sources, mindsdb does something, and the app graphs the output.

[parthiv11]

To reflect and summarize what I just read in order to judge my understanding - the proposal is to build an app store/play store style App that interfaces with a mindsdb service, user selects data sources, mindsdb does something, and the app graphs the output.

Yes @andy-brainome you get it right but it can do everything that mindsDB can support like model training , predictions, graph for visualization etc.

Thank you

[andy-brainome]

To reflect and summarize what I just read in order to judge my understanding - the proposal is to build an app store/play store style App that interfaces with a mindsdb service, user selects data sources, mindsdb does something, and the app graphs the output.

Yes @andy-brainome you get it right but it can do everything that mindsDB can support like model training , predictions, graph for visualization etc.

Thank you

My take away is that quite a bit more vision will be necessary to bridge the gap between "everything... etc." design statements and a concrete workflow with specific goals and use cases.

I'm very new to the mindsdb ecology and am unfamiliar with any successful applications of the technology. Nevertheless, I can see two benefits to this proposal over the SQL web interface I've encountered.

A) a GUI definitely opens up the field of possible users who don't want to know/use SQL for wrangling data into AI predictors. The GUI design needs to be narrowly defined and strongly opinionated in order to have a hope of reaching completion in a reasonable timeframe. Most users have a hard time relating to generic designs because it requires them to learn the product before closing the gap between features and their needs. E.g. Atlassian jira solved one problem space amazingly well before being repurposed into other spaces. Had they required users to define their processes before it could be useful, it would have the same traction as most of its competitors.

B) defining this as a mobile app leads me to wonder what benefits are accrued as compared to more conventional delivery methods. It is tied to a web service which precludes any benefits of working offline. Is there a strong desire for data scientist/ engineers to build on the go while commuting, at the family picnic, or at a ball game? Is the small screen really the most efficient medium for labeling instances and applying DS tricks to the data sets. Whatever the case, this app will be an add-on to the heavy lifting necessary to upload/define data sets and start solving the problem domain.

[parthiv11]

@andy-brainome According to me features in an app should be

• user selects data sources
• MindsDB can show graphs for understanding Dataset (if the user wants)
• Train the model using Data
• predict using the trained model

All these features with a nice GUI

[andy-brainome]

I love the simplicity of your workflow but I hardly find it compelling and I have doubts about the usefulness of it beyond a single adhoc use.

1) Data set selection - this step could be as simple as picking a table/view/adhoc query with filters and ordering. Mindsdb has the power of every db behind this step. Exactly how much db complexity is exposed to the user will make or break the scope and odds of completing a project of this scale.

2) Graphs and stuff - lovely to see but how does the user turn the pretty/ugly pictures into concrete metrics and actionable changes. Perhaps they feed back into step 1 until the labels are comprehensive, the distributions are balanced, and the feature engineering reaches steady state. Just showing some simple graphs without influencing the outcomes is meaningless.

3) Running AutoML or other ML work flow(s) is the magic behind the curtain here. The uninitiated will not understand why the model is not built instantaneously like my Google search queries. Providing clear guidance on how long to wait until complete is crucial for setting user expectations.

4) predictor validation - the touchstone of our process. Requires adequate domain knowledge to judge the value in this model vs. starting over. Balancing accuracy with generalization with model size is a judgement call that a generic tool can never make.

All these steps are mechanically correct. What is missing is a compelling storyline that converts users from abstract entities into agents of change.

And that leads me to my final observation on this design document: where does the anti-entropy ordering this tool generates convert from adhoc ephemeral to permanent lasting change? All of this effort needs to be more than just shuffling data into code. It needs to transform data into knowledge and ultimately wisdom.

Sorry to go all big picture on you but these tools need to be sticky for them to really succeed and that means it must satisfy an existential need or generate cash. Never missing a step while dancing is fine but great dancing is provoking an emotional response from everyone present.

On Fri, Jun 10, 2022, 4:06 AM Parthiv Makwana @.***> wrote:

@andy-brainome https://github.com/andy-brainome According to me features in an app should be

• user selects data sources
• MindsDB can show graphs for understanding Dataset (if the user wants)
• Train the model using Data
• predict using the trained model

All these features with a nice GUI

— Reply to this email directly, view it on GitHub https://github.com/mindsdb/mindsdb/issues/2314#issuecomment-1152245781, or unsubscribe https://github.com/notifications/unsubscribe-auth/AOTAKQDJ7ZF6WI7OYC3XON3VOMOVBANCNFSM5YHTM6AA . You are receiving this because you were mentioned.Message ID: @.***>

[parthiv11]

@andy-brainome What are your suggestions?

[andy-brainome]

@andy-brainome What are your suggestions?

I am not sure what you are asking for. I have offered several hundred words of hopefully constructive criticism.

Assuming you are looking for guidance on how to proceed. It depends on your goals. IMO, there are two paths you can follow that are not mutually exclusive.

A) You can code an app that does the 4 steps you envision and let the world judge if you have a minimum viable product.

B) You can test your design and assumptions by observing an "incidental data scientist with an iPhone" (your target demographic) trying to make ML work for them but not having the math skills or training that a trained DS has.

Answer these questions for yourself:

Would this tool be enough for a CFO or business analyst to enter a Kaggle competition?

How would they do feature engineering with this tool and would they need to do that to build a successful model?

[parthiv11]

@andy-brainome You know about InfluxDB. If yes then go to issue #2328 and upvote it.

If no then trust me it should be in list. It should be in intergration with mindsDB.

Thank you in advance

[ZoranPandovski]

Closing for now until we open source our GUI.