Nshims
New Member
- Joined
- May 28, 2014
- Messages
- 227
- Points
- 18
Well, here we are with a first test version,
this can be strange because it is already version 4.45.2,
but that has a simple explanation.
The first number represents the generation of the algorithm used,
after comes the number of tests, learning process and updates,
the last one is just the number of pokemons featured in the current version.
So we can say that it is the 4th algorithm that I tested
and that are only 2 pokemons available (Growlithe and Staryu),
i will always relate to the first evo but, all evo line is considered.
The main engine is a perceptron, that take 8 data set as input (shiny status, IVs and hidden power) and give as output
4 categories each one represent the probability of the pokemon being expensive,
more specifically price sets (1k - 100k, 101k - 1m, 1.001m - 100m, 100.001m +)
plus a certain impossible set that distinguishes non-belonging to a given category.
4 categories each one represent the probability of the pokemon being expensive,
more specifically price sets (1k - 100k, 101k - 1m, 1.001m - 100m, 100.001m +)
plus a certain impossible set that distinguishes non-belonging to a given category.
You can download a copy and see how it works,
it's a simple interface, you just have to enter the data and press predict.
3 separately trained models with different initial conditions of give approximate prices,
after an average is calculated and a set of possible prices goes from -10% to + 10%.
Download
I noticed that avast sometimes sends the file for further analysis, this can cause a runtime error,
you just have to wait for the analysis to finish.
Note: for the moment this is a beta version, the small number of data means that there are still a lot of dead spots,
I have not tested all the spacing but a difference of 0 - 2 ivs can still give good prices.
it's a simple interface, you just have to enter the data and press predict.
3 separately trained models with different initial conditions of give approximate prices,
after an average is calculated and a set of possible prices goes from -10% to + 10%.
Download
I noticed that avast sometimes sends the file for further analysis, this can cause a runtime error,
you just have to wait for the analysis to finish.
Note: for the moment this is a beta version, the small number of data means that there are still a lot of dead spots,
I have not tested all the spacing but a difference of 0 - 2 ivs can still give good prices.
Now the hardest parts start, collecting data and setting prices.
Data collection and price definition are two steps that are linked by the fact that they improve the quality of predictions.
But this data must be very well chosen so that the algorithm can find the real motivations behind the prices.
With that said, if you want to help in one way or another, here are the instructions:
To help with data collection you must send me screenshots with the small frame of the pokemons
in adition if you have time you can create an excel file or a csv.
it is important that the database consists of real pokemons that have been captured in game.
Because theoretically all evs are equiprobable, but in practice we see only a small part.
Small frame example:
CSV example:
Shiny|Lvl|HP|ATK|DEF|SP ATK|SP DEF|SPD|Hidden Power|
If you want to help with price checking, you can contact me by pm or in the game to discuss it.
As I usually program on Linux, I had to use MSYS | MinGW to create a more comfortable environment,
if you have skills in windows development, I would be happy to ask you some questions,
pm me in game or here for more information.
To help with data collection you must send me screenshots with the small frame of the pokemons
in adition if you have time you can create an excel file or a csv.
it is important that the database consists of real pokemons that have been captured in game.
Because theoretically all evs are equiprobable, but in practice we see only a small part.
Small frame example:
CSV example:
Shiny|Lvl|HP|ATK|DEF|SP ATK|SP DEF|SPD|Hidden Power|
If you want to help with price checking, you can contact me by pm or in the game to discuss it.
As I usually program on Linux, I had to use MSYS | MinGW to create a more comfortable environment,
if you have skills in windows development, I would be happy to ask you some questions,
pm me in game or here for more information.
Thanks for price check:
Kingstone
2020/07/06
Added new hidden power system
Added hidden power power calculation system
Added Eevee data (Beta)
2020/04/10
Added accuracy detection system
Fixed bug related to IVs normalization
2020/04/09
Added Scyther / Scizor
Fixed classification bug
Fixed set currency bug
Added new hidden power system
Added hidden power power calculation system
Added Eevee data (Beta)
2020/04/10
Added accuracy detection system
Fixed bug related to IVs normalization
2020/04/09
Added Scyther / Scizor
Fixed classification bug
Fixed set currency bug
Last edited: