For simple testing, set up a test in simulate_and_test.py. Here you fake the values you want to measure (y_true) and add noise to them (y_observed = y_true + noise). You then run y_observed through the Kalman filter. This is neat because you can plot true/noise/filtered to see the effect of the filter.
The standard deviation doesn't really matter, the filter will converge pretty quickly.
Here is a plot of a function (red) with added noise (yellow) that has been filtered (blue):
Here are some different filter settings for different timespans of one minute bitcoin data. The filter step length is adjusted for the time range: you typically want more filtering for larger timeranges.
