timemachines
Skaters
What’s a “skater”, abstractly?
A skater is a function suggesting a state machine for sequential assimilation and forecasting (hence the name timemachines).
\[f : (y_t, state; k) \mapsto ( [\hat{y}(t+1),\hat{y}(t+2),\dots,\hat{y}(t+k) ], [\sigma(t+1),\dots,\sigma(t+k)], posterior\ state))\]where $\sigma(t+l)$ estimates the standard error of the prediction $\hat{y}(t+l)$. (This looks better in the source code btw.)
If you prefer an legit (stateful) state machine, see FAQ question 1.
See interface for description of input and output arguments.
Finding a skater
Poke around in /skaters.
| Package | Location | Comment |
|---|---|---|
| StatsModels.TSA | skaters/tsa | |
| DARTS | skaters/drts | |
| SkTime | skaters/sk | |
| StatsForecast | Also in sktime, for now | |
| TBATS | skaters/bats | |
| PMD | skaters/sk | |
| Prophet | skaters/proph | A sine from god |
| NeuralProphet | skaters/nproph | |
| KATS | skaters/kts | |
| GreyKite | skaters/gk | |
| Orbit-ML | skaters/orbt | |
| PyCaret | skaters/pycrt | |
| River | skaters/rvr | Flowing |
| Divinity | skaters/divine | |
| PyDLM | skaters/pydlm | |
| PyFlux | skaters/flux | |
| GluonTS | skaters/glu | |
| Merlion | skaters/mrln |
And some home grown
| Package | Location | Remark |
|---|---|---|
| Simple | skaters/simple | |
| SMDK | skaters/smdk | |
| Elo | skaters/elo | Future-proof |
and maybe more (see here).
Importing a skater
Examples:
from timemachines.skaters.sk.skautoarima import sk_autoarima as f
from timemachines.skaters.simple.hypocraticensemble import quick_balanced_ema_ensemble as f
Or do it by name, which can be less efficient:
from timemachines.skaters.localskaters import local_skater_from_name
f = local_skater_from_name('rvr_p1_d0_q0')
The names of skaters appear on the Elo ratings tables.
Using a skater
“Skater” is a nmemonic for the non-data arguments (see interface) although you might need only the first two, s (state) and k (steps ahead). The script skating.py illustrates the usage pattern. Like so:
from timemachines.skaters.simple.thinking import thinking_slow_and_fast as f
import numpy as np
y = np.cumsum(np.random.randn(1000))
s = {}
x = list()
for yi in y:
xi, xi_std, s = f(y=yi, s=s, k=3)
x.append(xi)
There’s more in examples/basic_usage.
Skater arguments and return values
Again, see the skater interface documentation for more details about expectations placed on skater functions.
-+-
Documentation map
