When choosing an electronic part to use in a project—say, a voltage regulator—the number of choices is staggering. Digikey lists about 400,000 results for the search query “voltage regulator”. Additionally, the number of choices that are wrong is very high. If I chose a random one of those 400,000 products to use in my project, I’d almost certainly get something that was useless. However, even after eliminating all the obviously wrong choices (wrong output voltage, wrong function, wrong current limit), there are still likely to be thousands of difficult-to-distinguish options.
A much easier way to make a decision like this would be to answer a series of questions. Given the set of products that match the query “voltage regulator”, I want to answer the minimal number of questions that will return me a unique, single selection.
What do you want to buy?
> voltage regulator
What output voltage?
> 5V
What current limit?
> 1A
Surface mount or through-hole?
> through-hole
Maximum input voltage?
> 15V
Voltage dropout?
> 2V
Buy NCP7805TGOS-ND.
This seems like a problem particularly well suited to machine learning. It’s probably a lot easier for a computer to look at 500 product data sheets, extract key phrases and figure out what the differences between them are than it is for a human to do the same. Dealing with ranges is likely more complicated (Digikey’s search feature does a terrible job of it), but I suspect one can go a long way with just key phrase extraction.
Perhaps this is a good excuse to learn TensorFlow.