The MLP Arena
Pony Talk => Pony Corral => Topic started by: ChibiMango on March 03, 2021, 08:26:47 PM
-
I already posted about this in the offtopic pony thread, but I think I'll make it it's own thread.
One of my hobbies outside of MLP is machine learning. To practice coding my own AI, I'm trying to create one that guesses the price of ponies! Hopefully the long-term goal would be to use it as a tool to price sales, but right now it's far away from there.
It's kind of a pain to collect data for this thing, so here's how you can help!
- Have a pony that you own and know the price of. It can be one you're selling or one you bought for yourself recently enough that you remember how much you paid.
- Identify the pony! While I'd prefer to know the exact set the pony came from, just the name is fine.
- Make a copy of this spreadsheet, then fill it out with details about the pony. It'll generate a code that'll let the AI know what kind of condition the pony is in!
https://docs.google.com/spreadsheets/d/1zNwxM6SXYM4RXn3gAjUONDb7k-x5VhMpbhz7bHTPbtM/edit?usp=sharing (https://docs.google.com/spreadsheets/d/1zNwxM6SXYM4RXn3gAjUONDb7k-x5VhMpbhz7bHTPbtM/edit?usp=sharing) - Post the price, identity, code, and any extra information here! I'll see if I can get it added to my dataset to teach the AI!
An example of one of the AI's latest guesses: it guessed that this Valenshy is worth $9.61 just now. It's actually worth $14.98, so that's a loss of $5.37. Still needs some work, but it's a start!
visitors can't see pics , please register or login
-
This seems like a great and impressive idea !
I'm trying to think of more or different criterias to describe the pony's condition, such as :
- Damaged by acne cream
- Sunburnt
- Chewed body
- Missing body parts
- Plasticizer leakage
- Faulty electronics
- Faulty mechanical parts
- Smoke damage (cigarette smell/yellowing)
- Rotten hair
- Faded hair (Fading Pink,...)
- Unsealed head
- De-tinselled
- De-flocked
- Damaged flocking
- Restored
- Rehair
- Mint/MIB/MOC
- Damaged card/Box
Maybe I will come up with more. I hope that helps !
Have a nice day !
Rose.
-
This is a really good idea. People used to regularly respond to price check requests on the Corral, which was a great resource. I'm not so sure what people do nowadays, maybe it's shifted to Facebook which I don't have.
-
Very interesting! I'm looking forward to seeing how your AI progresses.
-
Thanks for all the feedback!
I've messed around with things a bit more by cleaning up the dataset a bit to take care of outliers (in this case, ridiculously cheap or especially pricey ponies), and I've also tweaked a few training settings. Specifically, I've set it to create a backup file every time a stat called 'validation loss' hits a new record for lowness. That means that the best versions of the model will be preserved for me to save!
These changes have definitely helped. The newest guess for our testing Valenshy is $15.91, MUCH closer than before! :biggrin:
This seems like a great and impressive idea !
I'm trying to think of more or different criterias to describe the pony's condition, such as :
- Damaged by acne cream
- Sunburnt
- Chewed body
- Missing body parts
- Plasticizer leakage
- Faulty electronics
- Faulty mechanical parts
- Smoke damage (cigarette smell/yellowing)
- Rotten hair
- Faded hair (Fading Pink,...)
- Unsealed head
- De-tinselled
- De-flocked
- Damaged flocking
- Restored
- Rehair
- Mint/MIB/MOC
- Damaged card/Box
Maybe I will come up with more. I hope that helps !
Have a nice day !
Rose.
Thank you! I'll look into adding some of these! Some of them, like MIB, might be a little tricky to add because it'd require me to add a new category to the dataset, which would require a bit of an overhaul of the system I've got going on, but I'll see what I can do!
-
interesting.
So you only manually input the prices?