Co author here - I make the API and the backend.
The numbers come from a large sample of exchanges, with several million new datapoint added every day to cover every crypto we find important.
It is more of a hobby for me, but I hope you like it!
The numbers come from a large sample of exchanges, with several million new datapoint added every day to cover every crypto we find important.
It is more of a hobby for me, but I hope you like it!
If you ever decide you want to expand to more than top 50 coins, then I'd love to help in any way I can - particularly with their API - feel free to email me: "james" at $my_hackernews_username ".nz"
We are on #1 page!! Big surprise !! Now I have to reply to all questions but HN tell me 'you are posting too much'
I must fix bugs first, but I will be happy to email you!! I love cryptopia and it's approach!! Our email is in the faq: cryptomarketplot@protonmail.com
There is more on the website than the eye can see now. We will unveil new features soon.
Expanding beyond top #50 is something we want. We want to support more coins, and there is work in progress on that.
But after noticing what coins it will expose, we almost changed our mind and decide to just be top20.
After hard thinking, we decided to spend much more effort on methods to flags suspicious coins. Because there are many small promising coins that can benefit from exposure.
It is almost there. More algorithms to make!
If you say you made it for hobby, how are you trying to replace Coinmarketcap?
I hope we grow and replace coinmarketcap. But that will take hard work, and lots of luck.
We have algorithms to spot anomalies. We want to offer extremely reliable data.
Disclose the method and source
"average price on last calendar"
Average is a bad function to minimize error between expected values and actual values. Use median or median of the 5%-95% percentiles.
We need to upgrade the FAQ to be more specific with the methods we use.
To be quick: The price come from many exchange. We exclude suspicious points. Then we do aggregation function like the average.
Averaging may not technically be the best method, but it is simple, and it has a wide support. And with the number of data points we have, we assume the Law of Large Number apply.
Still it could be interesting to use the median. I need to think more about it.
Null hypothesis mandates you make no such assumption. Even if you did, a gaussian distribution assumes variable independence which markets are most certainly not. Even still if the variables were independent, you surely don't have enough samples of prices to report an accurate mean.
But you would be surprised to know how many samples we have at any moment!
We would like to offer historical API to paying customers. We have a lot of data.
But we want to retain customers. We do not want them to leave after dumping our data. At keep keep paying us :-)
So we came to the conclusion: you get data from when you start to pay us, and we add a little extra (last 2 months) but no more. It is in the FAQ.
Historical API, we must think more about that.