-
Notifications
You must be signed in to change notification settings - Fork 22
Roadmap
This page tracks functionality that exists in the R smooth package but is not yet implemented in the Python port, plus methods that are partially implemented. It exists so the per-function reference pages can stay focused on what currently works rather than carrying TBA markers throughout.
Status legend:
| Marker | Meaning |
|---|---|
| Planned | On the implementation roadmap. |
| Partial | Exists in Python but missing features compared to R. |
| Not planned | No plan to port. |
For the runtime-numerical comparison between R and Python on already-implemented functionality, see R-Python-differences.
| Function | R name | Python | Status | Notes |
|---|---|---|---|---|
| Generalised Univariate Model |
gum(), auto.gum()
|
— | Planned | Whole function. See GUM. |
| State Space ARIMA |
ssarima(), auto.ssarima()
|
— | Planned | Whole function. See SSARIMA. The Python port focuses on MSARIMA first because it is faster and more general; pure SSARIMA may follow. |
| Scale model |
sm(), implant()
|
— | Planned | Heteroscedasticity modelling for ADAM. See Scale-Model. |
The standalone sim_* family is implemented in Python (sim_es, sim_ssarima, sim_ces, sim_sma, sim_oes, plus SimulateResult). The .simulate() method on fitted models is implemented for ADAM, OM, and OMG. The remaining gap is sim_gum — present in Python but only useful once GUM itself is ported.
| Function | R name | Python name | Status | Notes |
|---|---|---|---|---|
| ETS simulation | sim.es() |
sim_es() |
Implemented | — |
| ARIMA simulation | sim.ssarima() |
sim_ssarima() |
Implemented | Works without an SSARIMA class — direct data generator. |
| CES simulation | sim.ces() |
sim_ces() |
Implemented | — |
| SMA simulation | sim.sma() |
sim_sma() |
Implemented | — |
| Occurrence simulation | sim.oes() |
sim_oes() |
Implemented | — |
| GUM simulation | sim.gum() |
sim_gum() |
Implemented (standalone) | The data generator works; full GUM model not ported. |
.simulate() method on fitted model |
simulate(model) |
model.simulate(...) |
Implemented | Available on ADAM, OM, OMG. |
See Simulation-Functions for the R-side documentation and notes on how the Python sim_* functions mirror it.
These extractor methods exist on R model objects but are not yet on Python ADAM / ES / CES / etc.
| Method | R | Status | Notes |
|---|---|---|---|
pointLik() |
Yes | Planned | Point log-likelihoods per observation. |
pAIC() / pBIC()
|
Yes | Planned | Point information criteria. |
accuracy() |
Yes | Planned | Forecast accuracy measures. |
pls() |
Yes | Planned | Prediction-likelihood score. |
extractScale() |
Yes | Planned | Only meaningful once sm() is in. |
implant() |
Yes | Planned | Merges location and scale models — needs sm(). |
xtable() |
Yes | Not planned | LaTeX table output; R-only. |
These accept fewer options in Python than in R.
| Surface | R | Python | Status |
|---|---|---|---|
Reusing a previously fitted model as model=
|
Yes | Not supported | Not planned |
regressors="integrate" (GUM-only option) |
Yes | — | Implemented only once GUM is ported. |
orders for SSARIMA, GUM
|
Yes | — | Implemented only once those functions are ported. |
External regressors (X=, regressors=) are supported in Python on ADAM, ES, CES, MSARIMA, and OM / OMG. Earlier revisions of this page mistakenly listed them as Roadmap items.
A page in this wiki should describe what is implemented. If you find a TBA row in a function or parameter table, treat it as a documentation bug and move it here. The intent is that LLMs (and humans) reading a reference page never have to filter out empty cells to learn what a function actually does.
- Home
- R-Python-differences — numerical-parity status between the two implementations
- Glossary — terminology used across the wiki