This is why "machine learning" is still flawed:

"Stations you might like:"
- Lullabies & Bedtime Music
- Instrumental Lullabies
- Rockabye Baby
- Toddler Time
- Twinkle Twinkle Little Rock Star
- The Wiggles

Perhaps I should get my kids their own account. Perhaps the algorithm should now that if I'm looking for music after 9:30 on my phone, and not the Alexa, my kids are already asleep.

Its not a problem with machine learning as such, just that they never plugged time into the algorithm.

If they had it would have taken that into account.
So its an error in design and parameter selection.

The most common one when any ML or AI solution performed worse than hoped.

stuck in the wrong cluster again?

@pythonInRelay thats a bit harsh ;)

I have a similar problem with streaming services where we share one in the family so I can totally relate.

But my kids are older so time would not help me :P

But its something to keep in mind if we ever our self get involved in ML.

Think outside of the box, pitch wild ideas and test them.

Any parameter that proves no real value can be dropped but do not try to second guess to much.

We know to little of how our own minds work and @devphobe question is a good example on how a service might jump ahead of the competition :)

@Voxera yes, but I'm sure that adding time in would disrupt somebody else who wants non-time based responses. One day I presume compute will be cheap enough that different ML models can be used for different users.

Just teach your kids to like industrial black metal. Problem solved.

@electrineer I like the way you think

ML is easy, data gathering, sorting, tagging, planning the whole system is difficult. ML are easy after that, they learn themselves.

@Gregozor2121 Anybody can teach a student that's willing to learn. Teaching a student a path of reasoning to the correct answer is hard.