Loading...

{{notif_text}}

SocialHelpouts is now CutShort! Read about it here
Who pays how much? Be informed with this salary report on Indian startups.
Why join channels?
Learn from peers
Discuss and share learning resources with the top professionals across the world
Open business or job opportunities
Earn reputation points to get consulting projects, attract talent or land jobs.
Accelerate your growth
Grow your network and get exclusive deals from our learning partners.
signup now
Nipun Soni asked a question

At the MVP stage in a company, data is not abundant. How does one go about making machine learning models at this stage?

At the MVP stage where data is not available in abundance, how do you cope up with analyzing and predicting data. If you are able to build and predict with this small dataset, will the algorithm still hold relevance or will there need to be changes when there is a sudden inflow of complex data points. 

answer
submitting answer...
submit
No answers yet. Be the first one to answer!
2 answers
Samim Ekram A am an python developer with interest in AI
as soon as the data increases you will surely face complexity in analyzing it and the current algorithms will hold no/less relevance. You didn't mention what kind of data you are working with but as far as I know if working in text analysis,video,image or audio there is data available in abundance to train the machine learning model. anyhow try building a model which is flexible , do some research on the type of data you are working on, read some research paper and see how the complexity changes and design the algo accordingly
Loading comments...
Meher Vamsi ML/NLP Engineer

Building a ML model without the rite data is like buying shoes without knowing which size. (Sorry, poor anology :P) . Having said that, there are multiple things you need to consider. 

1. If it is early stage and you have very little data, more often than not you may not even need a ML model. For example, if you are a E-comm company with few products and few users, you can not build a reccommendation engine and you do not have to also. You can make do with good old if-else conditions or hand crafted rules. 

2. If your product does require you to show some Proof of concept that you can pull off a particular model; than try to get some data that looks similar to your problem and develop a model on it. Say if you are into Data science consulting and you want to impress a new Digital marketing client with your swanky new click through rate than show that your model can do well on this criteo dataset.
3. As far as I know, there is no staright forward way to starting with a simple model on small data and gradually increasing the complexity as the data gets complex. It is much like building multiple models with multiple datasets. 

4. You may want to try to genreate a fake dataset if you know what kind of characteristics your features you would be having. For ex: predicting the grade of a student; given features like attendace, marks in various subjects etc. (Simple example, but you get the idea!) 

I think we can help out much better if you can tell what kind of problem is that you are solving. :) 

Loading comments...
To view all answers to this question, join this channel
join this channel
Awesome! You have connected your Facebook account. Like us on Facebook to stay updated.