No-Code Machine Learning?
Would anybody be interested in a No-Code Machine Learning platform?
As a user, you'd first be able to upload a numeric dataset, or a collection of images, or a series of text documents, etc. The system would learn from this data then allow you to upload new data to make predictions on!
For example, this may help you predict churn by making predictions based on their user's demographics, usage patterns, payment history, etc.
What do you think?
Another Data Robot?
interesting from implementation standpoint, but who will be willing to employ a system like this in real-life machine learning pipelines? I think it can become useful for a data scientist or an analyst during initial stages of a project to explore data and determine which algo eventually to use.
I am a software engineer in AI space helping productionize data science projects. I think such system would be more useful if it outputs some insights in terms of which algos/perdictors might be useful for a given dataset(s) as opposed to actually predicting something...
But generally, the user would need to do A LOT of work before they can even feed a dataset to such system: data prep / data cleaning, etc, which we all know takes 90% of the time. And you cannot prepare data without actually understanding it...
On the other hand, such tool might be used to make sense of some very noisy sensor data..
overall it is a great idea, but very ambitious... we have to be very specific in terms of who would be the target user for this (i.e. data scientist, data engineer, a business analysts) and what are the use cases.
I appreciate the feedback! Interesting point about how you'd find more value in algo recommendations over actual predictions. That was actually part of our thinking as well.
Yea, it'd be similar to datarobot. We figure there's still plenty of space in their market to carve out our space. Data science is still very much a large corporation with big teams kind of industry, so simplifying the process for small businesses can be a large opportunity. There's also plenty of room to educate and sell to the individual consumer directly. Think people who are interested in data science but don't know where to start.
worth taking a look at runwayml.com
runwayml is pretty cool. definitely in the same world but not sure it's direct competition. seems more geared towards automatically creating art/content based on previous content. thanks for the heads up.
alright. not sure about a complete no-code tool as in for production. perhaps an educational tool would be a better niche for completely no-code. anyway I'm open to hear your thoughts, message me on Twitter same handle.
Yes, if you can make it work. There is a lot of companies attempting to provide this kind of product but that's not to discourage you because they are not really solving it except for in some specific cases. Some have the problem that they only work for some trivial well-defined problems but fail at more difficult ones or ones with a not so clean data set. For example, I've seen one example where one developer without much experience in machine learning tried the ML service, excited about a 99% accuracy but did not question how good or even unrealistic result that was while the training set included the same data as the test set. The question is who you would be creating the product for and what knowledge they bring with them.
The difficult part of machine learning is not coding.
agreed with just about everything you're saying. we're still really early like just finishing up the MVP so not really sure where exactly our product-market fit will land. we're thinking we gear toward small businesses that can't afford a data science team. start with being able to accept messy, tabular data and build into accepting images and text documents as well.