add quantile forests

dvc.lock

@@ -68,9 +68,9 @@ stages:
       size: 1280486
     - path: src/probafcst//models/
       hash: md5
-      md5: 9e4fc803a61ab97704a486aef34eee91.dir
-      size: 60048
-      nfiles: 17
+      md5: 74c7ac520b842f6244b4bc9012232a8a.dir
+      size: 62864
+      nfiles: 19
     - path: src/probafcst//utils/tabularization.py
       hash: md5
       md5: 20b1a00caed3c407c77056b04bd5fa4a
@@ -170,11 +170,31 @@ stages:
               n_estimators: 200
               random_state: 0
               verbose: 0
+          qrf:
+            lags:
+            - 24
+            - 48
+            - 72
+            - 96
+            - 120
+            - 144
+            - 168
+            - 336
+            - 504
+            - 672
+            include_seasonal_dummies: true
+            cyclical_encodings: true
+            X_lag_cols: []
+            include_rolling_stats: false
+            kwargs:
+              n_estimators: 100
+              random_state: 0
+              n_jobs: -1
     outs:
     - path: models/energy_model.pkl
       hash: md5
-      md5: 818c68a30e12899f3f209a88aa8e96df
-      size: 2273354
+      md5: 7536442fe6561688c046c8a3bb75efba
+      size: 2276094
   train@bikes:
     cmd: python src/probafcst/pipeline/train.py --target bikes
     deps:
@@ -184,9 +204,9 @@ stages:
       size: 148879
     - path: src/probafcst//models/
       hash: md5
-      md5: 9e4fc803a61ab97704a486aef34eee91.dir
-      size: 60048
-      nfiles: 17
+      md5: 74c7ac520b842f6244b4bc9012232a8a.dir
+      size: 62864
+      nfiles: 19
     - path: src/probafcst//utils/tabularization.py
       hash: md5
       md5: 20b1a00caed3c407c77056b04bd5fa4a
@@ -283,11 +303,27 @@ stages:
               n_estimators: 100
               verbose: 0
               random_state: 0
+          qrf:
+            lags:
+            - 1
+            - 2
+            - 3
+            - 7
+            - 14
+            - 21
+            include_seasonal_dummies: true
+            cyclical_encodings: true
+            include_rolling_stats: false
+            X_lag_cols: []
+            kwargs:
+              n_estimators: 100
+              random_state: 0
+              n_jobs: -1
     outs:
     - path: models/bikes_model.pkl
       hash: md5
-      md5: 4537d54e03e658bc96b62baf3a52db78
-      size: 1219119
+      md5: a7529edfdb07b51c22a27e459fcdaa09
+      size: 1221859
   eval@energy:
     cmd: python src/probafcst/pipeline/evaluate.py --target energy
     deps:
@@ -297,8 +333,8 @@ stages:
       size: 1280486
     - path: models/energy_model.pkl
       hash: md5
-      md5: 818c68a30e12899f3f209a88aa8e96df
-      size: 2273354
+      md5: 7536442fe6561688c046c8a3bb75efba
+      size: 2276094
     - path: src/probafcst//backtest.py
       hash: md5
       md5: c352b34254b7299b5ec810bac65b0949
@@ -336,19 +372,19 @@ stages:
     outs:
     - path: output/energy_eval_results.csv
       hash: md5
-      md5: 918e9b4b808ef3a8411a160eb1708091
+      md5: ef24a00ba647a7a4d9149f3d012fee10
       size: 5279
     - path: output/energy_metrics.json
       hash: md5
-      md5: b83a61cf6499da1efbfef176b912dd59
+      md5: fde281598f36df5135c5dc745cb83f5c
       size: 583
     - path: output/energy_pinball_losses.svg
       hash: md5
-      md5: 8ba26fca45d6eaa56bfb689a0738bb9f
+      md5: 3293bd89ff7a37ae2756c6c331b1b4ce
       size: 26352
     - path: output/eval_plots/energy/
       hash: md5
-      md5: 3c8779e7c44205be26773e9a68dd75c1.dir
+      md5: 019ee565f374587f408ad9c4fa0ee875.dir
       size: 254098
       nfiles: 4
   eval@bikes:
@@ -360,8 +396,8 @@ stages:
       size: 148879
     - path: models/bikes_model.pkl
       hash: md5
-      md5: 4537d54e03e658bc96b62baf3a52db78
-      size: 1219119
+      md5: a7529edfdb07b51c22a27e459fcdaa09
+      size: 1221859
     - path: src/probafcst//backtest.py
       hash: md5
       md5: c352b34254b7299b5ec810bac65b0949
@@ -399,32 +435,32 @@ stages:
     outs:
     - path: output/bikes_eval_results.csv
       hash: md5
-      md5: 7c624729cc89dd89c2f0c59d15ac6bd1
-      size: 19770
+      md5: 0bbace17ec7598b9fd6484280e1a6741
+      size: 19755
     - path: output/bikes_metrics.json
       hash: md5
-      md5: 9887af223700c37da1aa8dda455f58d8
+      md5: bfa27abac68c38c55bb5db43a8e299b0
       size: 587
     - path: output/bikes_pinball_losses.svg
       hash: md5
-      md5: 9b9b6befc4fb9b62c395394a913b6f2d
+      md5: 64c1b6849335c08205809f9b1051df09
       size: 30193
     - path: output/eval_plots/bikes/
       hash: md5
-      md5: 8acfbacb2865c4b620ffbc22a4f0d299.dir
+      md5: 97409185f3a29ecf2bd19f39bb551c89.dir
       size: 136584
       nfiles: 4
   submit:
     cmd: python src/probafcst/pipeline/submit.py
     deps:
     - path: models/bikes_model.pkl
       hash: md5
-      md5: 4537d54e03e658bc96b62baf3a52db78
-      size: 1219119
+      md5: a7529edfdb07b51c22a27e459fcdaa09
+      size: 1221859
     - path: models/energy_model.pkl
       hash: md5
-      md5: 818c68a30e12899f3f209a88aa8e96df
-      size: 2273354
+      md5: 7536442fe6561688c046c8a3bb75efba
+      size: 2276094
     - path: src/probafcst//plotting.py
       hash: md5
       md5: 482a42cf8b0b9196d98b0d8e772d83d2
@@ -452,13 +488,13 @@ stages:
     outs:
     - path: output/bikes_forecast.svg
       hash: md5
-      md5: 054aa7d71bf8ddfac1a4fbe0dcb11dda
+      md5: 3ac12752d873a6a26d1f9e057cb1389c
       size: 31828
     - path: output/energy_forecast.svg
       hash: md5
-      md5: 204c87458b6bfb51a55d1964d0dfa724
+      md5: 3d27f873b21994ca1a6c636eb9fa27cb
       size: 68043
     - path: output/submission.csv
       hash: md5
-      md5: c2881c6556d14c23d5bf848ff6808ff2
+      md5: 5ae5f5042c96a647279d3234588cf15f
       size: 1605

notebooks/dvc.ipynb

@@ -1,5 +1,14 @@
 {
  "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# ruff: noqa"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": null,

params.yaml

@@ -123,6 +123,23 @@ train:
         n_estimators: 100
         verbose: 0
         random_state: 0
+    qrf:
+      lags:
+        - 1
+        - 2
+        - 3
+        - 7
+        - 14
+        - 21
+      include_seasonal_dummies: true
+      cyclical_encodings: true
+      include_rolling_stats: false
+      X_lag_cols: []
+      kwargs:
+        n_estimators: 100
+        random_state: 0
+        n_jobs: -1
+
 
 
   energy:
@@ -212,6 +229,26 @@ train:
         n_estimators: 200
         random_state: 0
         verbose: 0
+    qrf:
+      lags:
+        - 24
+        - 48
+        - 72
+        - 96
+        - 120
+        - 144
+        - 168
+        - 336 # 2 weeks
+        - 504 # 3 weeks
+        - 672
+      include_seasonal_dummies: true
+      cyclical_encodings: true
+      X_lag_cols: []
+      include_rolling_stats: false
+      kwargs:
+        n_estimators: 100
+        random_state: 0
+        n_jobs: -1
 
 eval:
   backend: loky # null or 'loky' for parallel evaluate

pyproject.toml

@@ -35,6 +35,7 @@ dependencies = [
     "lightgbm>=4.5.0",
     "scikit-learn>=1.5.2",
     "catboost>=1.2.7",
+    "quantile-forest>=1.3.11",
 ]
 
 

src/probafcst/models/_base.py

@@ -11,6 +11,7 @@
 from probafcst.models.darts import get_xgboost_model
 from probafcst.models.lgbm import LGBMQuantileForecaster
 from probafcst.models.linear_qr import LinearQuantileForecaster
+from probafcst.models.qrf import RandomForestQuantileForecaster
 from probafcst.models.xgboost import XGBQuantileForecaster
 
 
@@ -51,6 +52,12 @@ def get_model(
             model = LGBMQuantileForecaster(**model_params, quantiles=quantiles)
         case "catboost":
             model = CatBoostQuantileForecaster(**model_params, quantiles=quantiles)
+        case "qrf":
+            if n_jobs is not None:
+                model_params["kwargs"]["n_jobs"] = n_jobs
+            model = RandomForestQuantileForecaster(**model_params, quantiles=quantiles)
+        case _:
+            raise ValueError(f"Unsupported model: {params.selected}")
 
     return model
 

src/probafcst/models/qrf.py

@@ -0,0 +1,34 @@
+"""Quantile regression forecaster using a random forest model."""
+
+from quantile_forest import RandomForestQuantileRegressor
+
+from probafcst.models.regression import QuantileRegressionForecaster
+
+
+class RandomForestQuantileForecaster(QuantileRegressionForecaster):
+    """Quantile regression forecaster using a random forest model."""
+
+    def __init__(
+        self,
+        lags: list[int],
+        quantiles: list[float],
+        include_seasonal_dummies=True,
+        cyclical_encodings=True,
+        include_rolling_stats=False,
+        X_lag_cols: list[str] | None = None,
+        kwargs: dict | None = None,
+    ):
+        self.kwargs = kwargs or {}
+        model = RandomForestQuantileRegressor(
+            default_quantiles=quantiles,
+            **self.kwargs,
+        )
+        super().__init__(
+            model=model,
+            lags=lags,
+            quantiles=quantiles,
+            include_seasonal_dummies=include_seasonal_dummies,
+            include_rolling_stats=include_rolling_stats,
+            cyclical_encodings=cyclical_encodings,
+            X_lag_cols=X_lag_cols,
+        )