Skip to the content.


If you are familiar with how to predict and you have noticed the zstreams that are created, you might consider some more advanced functionality. The /copula API serves two purposes:

The following family of streams illustrates the pattern.

Type of stream Examples
Regular #1 c5_bitcoin.json
Regular #2 c5_ethereum.json
Regular #3 c5_cardano.json
Z-scores stream #1 z1~c5_ethereum~3555
Bivariate copula #1,#2 z2~c5_bitcoin~c5_ethereum~3555
Trivariate copula (70s) z3~c5_bitcoin~c5_cardano~c5_ethereum~70
Trivariate copula (1 hr) z3~c5_bitcoin~c5_cardano~c5_ethereum~3555

To reiterate there are NO MULTIVARIATE streams on However there are univariate streams whose values represent a point in \(R^2\) or $$R^3$. The mapping from one dimension to two or three is implied. The former is used by streams prefixed by z2~, the latter by streams prefixed by z3~.


Very similar to the use of set()

from microprediction import MicroWriter
mw = MicroWriter(write_key='YOUR WRITE KEY HERE')
names = ['my_stream_a.json','my_stream_b.json','my_stream_c.json', 'my_stream_d.json']
values = [5,6,4,9]
res = mw.cset(names=names,values=values)

Python examples

See /submission_examples_copulas.

The meaning of values in z2~ streams

Probably you’ve already grok’d zscores, and the bivariate and trivariate counterparts are not dissimilar:

\[z(x_1,x_2) = \Phi^{-1} \left( H\left( F^1_{70}(x_1), F^2_{70}(x_2) \right) \right)\]


The choice of space-filling curve H

As for H, a Morton Z-curves are employed. Actually the most up to date reference for these embeddings is the code, so see zcurve_conventions. At time of writing these docs, the heart of this calculation is:


where ‘prctls’ comprisese the pair \(F^1_{70}(x_1), F^2_{70}(x_2)\) in the bivariate case, and similarly for trivariate.


Documentation map

Edit this page.