notebook

Author: xadupre

_doc/api/datasets/data_ts.rst

@@ -0,0 +1,7 @@
+
+teachpyx.datasets.data_ts
+=========================
+
+.. automodule:: teachpyx.datasets.data_ts
+    :members:
+    :no-undoc-members:

_doc/api/datasets/index.rst

@@ -17,6 +17,7 @@ teachpyx.datasets
 
 
     data_helper
+    data_ts
     documentation
     enedis
     gpd_helper

_doc/articles/2025/2025-03-01-route2025.rst

@@ -156,12 +156,46 @@ accessible depuis le package `datasets <https://huggingface.co/docs/datasets/en/
 Séance 5 (6/3)
 ==============
 
+**Régression, Classification linéaires**
+
+:epkg:`statsmodels` pour obtenir le résultat de tests de nullité des coefficients
+
+* :ref:`Régression logistique en 2D <nbl-practice-ml-winesc_color_line>`
+* :ref:`Plusieurs modèles, données disjointes <nbl-practice-ml-winesc_color_linear>`
+
+**Interprétabilité**
+
+* `Partial Dependence <https://scikit-learn.org/stable/modules/partial_dependence.html>`_
+* `Permutation Importance <https://scikit-learn.org/stable/modules/permutation_importance.html>`_
+* `LIME <https://arxiv.org/abs/1602.04938>`_
+* `Shapley value <https://en.wikipedia.org/wiki/Shapley_value>`_,
+  `SHAP <https://shap.readthedocs.io/en/latest/index.html>`_
+* `Counterfactual Reasoning and Learning Systems <https://arxiv.org/abs/1209.2355>`_
+
 **séries temporelles**
 
-* :epkg:`statsmodels`, :epkg:`sktime`, :epkg:`skforecast`, :epkg:`prophet`, :epkg:`pyflux`
+Le modèle de référence est :epkg:`statsmodels`
+
+* :ref:`Single Spectrum Analysis (SSA) <nbl-practice-ml-timeseries_ssa>`
+* :ref:`Décomposition d'une série temporelle <nbl-practice-ml-timeseries_seasonal>`
+
+:epkg:`sktime` propose une API plus proche de :epkg:`scikit-learn`
+et d'autres modèles comme le clusting ou la segmentation de séries temporelles.
+
+:epkg:`prophet` fait aussi de la prédiction et contient aussi des algorithmes
+de détection de changement de régime, il contient une bonne base de jours
+fériés.
+
+:epkg:`pyflux` permet d'estimer des modules `GARCH
+<https://en.wikipedia.org/wiki/Autoregressive_conditional_heteroskedasticity>`_.
+
+**Analyse de survie**
+
 * :epkg:`scikit-survival`, :epkg:`lifelines`, analyses de survie,
   `Analyse de survie <https://sdpython.github.io/doc/mlstatpy/dev/c_ml/survival_analysis.html>`_,
 
+**Deep Learning**
+
 * `DeepAR <https://arxiv.org/abs/1704.04110>`_
 * temps réel
 

_doc/notebook_gallery.rst

@@ -179,7 +179,6 @@ Machine Learning
     practice/ml/winesc_multi_stacking
     practice/ml/artificiel_multiclass
     practice/ml/ml_features_model
-    practice/ml/timeseries_ssa
     practice/ml/ml_a_tree_overfitting
     practice/ml/gradient_boosting
     practice/ml/ridge_lasso
@@ -189,3 +188,8 @@ Machine Learning
     practice/ml/pretraitement_son
     practice/ml/pretraitement_texte
 
+.. nblinkgallery::
+    :caption: timeseries
+
+    practice/ml/timeseries_ssa
+    practice/ml/timeseries_seasonal

_doc/practice/index_ml.rst

@@ -55,6 +55,7 @@ Machine Learning
     :caption: Séries Temporelles
 
     ml/timeseries_ssa
+    ml/timeseries_seasonal
 
 .. toctree::
     :maxdepth: 1

_doc/practice/ml/timeseries_seasonal.ipynb

@@ -0,0 +1,14 @@
+import unittest
+from teachpyx.ext_test_case import ExtTestCase
+from teachpyx.datasets.data_ts import generate_sells
+
+
+class TestDataTs(ExtTestCase):
+
+    def test_generate_sells(self):
+        df = generate_sells()
+        self.assertEqual(len(df), 731)
+
+
+if __name__ == "__main__":
+    unittest.main()

_unittests/ut_datasets/test_data_ts.py

@@ -0,0 +1,41 @@
+from typing import List, Optional
+from datetime import datetime, timedelta
+import numpy
+
+
+def generate_sells(
+    duration: int = 730,
+    end: Optional[datetime] = None,
+    week_coef: Optional[List[float]] = None,
+    month_coef: Optional[List[float]] = None,
+    trend: float = 1.1,
+):
+    """
+    Generates dummy data and trends and seasonality.
+    """
+    if week_coef is None:
+        week_coef = numpy.array([0.1, 0.12, 0.12, 0.15, 0.20, 0.0, 0.0])
+        week_coef[5] = 1.0 - week_coef.sum()
+    if month_coef is None:
+        month_coef = [0.8, 1, 1, 1, 1, 1, 0.8, 0.6, 1, 1, 1, 1.5]
+        month_coef = numpy.array(month_coef)
+        month_coef /= month_coef.sum()
+
+    if end is None:
+        end = datetime.now()
+    begin = end - timedelta(duration)
+    day = timedelta(1)
+
+    rows = []
+    rnd = (numpy.random.randn(duration + 1) * 0.1) + 1
+    exp = (1 + numpy.exp(-numpy.arange(duration + 1) / duration * trend)) ** (-1)
+    pos = 0
+    while begin <= end:
+        month = begin.month
+        weekd = begin.weekday()
+        value = rnd[pos] * week_coef[weekd] * month_coef[month - 1] * exp[pos]
+        pos += 1
+        obs = dict(date=begin, value=value)
+        rows.append(obs)
+        begin += day
+    return rows