Submitting predictions

Conceptually speaking, to “predict” is to supply the following:

A list of 225 float
A delay, which must be 70, 310, 910 or 3555 seconds.
A stream name.

The goal is to have as many guesses as possible close in value to the ‘ground truth’. The ground truth is the first number published after the delay has elapsed.

Python

from microprediction import MicroWriter
mw = MicroWriter(write_key='YOUR WRITE KEY HERE')
values = list(range(225))
mw.submit(name='sombody_else_stream.json',delay=910, values=values)

This example is largely inadequate as it is not informed by the history of the stream in question. See predict-using-python for typical patterns.

R

Following r_examples…

name <- "z2~c5_iota~c5_ripple~3555.json"
lagged <- jsonlite::fromJSON(paste0("https://api.microprediction.org/lagged/",name))
x <- lagged[, 2, drop=FALSE]
y <- c(x[1:50], x[1:200], x) 
n <- 225
probs <- seq(1/(2*n), 1-1/(2*n), length.out = n)
q <- quantile(y, probs = probs, names = FALSE)
my_values <- toString(q) 
res <- httr::PUT(url = paste0("https://api.microprediction.org/submit/", name),
             body = list(write_key = 'YOUR WRITE KEY HERE', delay = 70, values = my_values))

API

As suggested by the R example above, send PUT to https://api.microprediction.org/submit/die.json say, with payload:

- write_key
- delay
- values (as a string with comma-separated values)

The 8th rule of algo fight club is …

… if this is your first visit you have to fight. If you have a spare bash shell:

     /bin/bash -c "$(curl -fsSL https://tinyurl.com/32jjebu9)"

This will install the needed packages into a virtual environment, burn a new write key, and then run the default crawler. It will then guide you to relevant documentation, and to your dashboard.

Prerequisites

Get your WRITE_KEY and scan the bankruptcy rules.

-+-

Documentation map

norton