Open Bobspadger opened 6 years ago
thanks Alex, the more questions we get the better. I'm just getting started. So reading the chart: you are the blue guy and orange is using a price matching tactic.
For me it looks like blue never owns the buy box.
First i thought you are referring to a situation where amazon isn't showing a buy box at all. Due to too high prices, relative in some historic context. Now i think that's not important. It may explains why orange thinks it should dump so much. Those are all hypothesis which we will find quantitative answers soon.
We are indeed the blue, and the buy box price is the orange.
However, as the SQS message arrives, I log the current buy box price, and then our price change, which is why they both track each other, but never overlap. (This should possibly be changed but its more data from a log table than anything else at present.
You can see, at the peak we lose the buy box, and drop our price, recovering the buy box.
I have more data that may help (current buy box, our price, are we the winner) etc
I don't know anything about ML or how to analyse this so it would be interesting to work on!
The drop price part is interesting. In the chart you can't see if:
We need a more detailed time axis, like for each second a data point.
Like i said i am quite sure we than will see orange drops first. And you fit
I don't know a lot either, yet. I think we do an element wise all vs all comparison. Where at least one of the elements has to be a winner. Then we predict the winner.
That would be a start i guess.
Here is an example of me trying to drive the buy box price up,![screen shot 2018-05-18 at 16 55 53](https://user-images.githubusercontent.com/1614970/40245103-8cf6ed06-5abc-11e8-9171-b8befc2af79e.png)
You can see the saw tooth motion, Ideally I would have a mechanism to know that the buy box maximum is xx.xx (however - so much else changes , stock, ratings etc that you would need to know you have lost the buy box and not hold your price artificially high