Add warm_start

docs/examples/plot_warm_start.py

@@ -0,0 +1,64 @@
+# -*- coding: utf-8 -*-
+"""
+===========================
+Fitting additional nodes
+===========================
+
+The local classifiers per node and per parent node support `warm_start=True`,
+which allows to add more estimators to an already fitted model.
+
+Nodes that were trained previously are skipped, and only new nodes are fitted.
+
+.. tabs::
+
+    .. code-tab:: python
+        :caption: LocalClassifierPerNode
+
+        rf = RandomForestClassifier()
+        classifier = LocalClassifierPerNode(local_classifier=rf, warm_start=True)
+        classifier.fit(X, y)
+
+
+    .. code-tab:: python
+        :caption: LocalClassifierPerParentNode
+
+        rf = RandomForestClassifier()
+        classifier = LocalClassifierPerParentNode(local_classifier=rf, warm_start=True)
+        classifier.fit(X, y)
+
+In the code below, there is a working example with the local classifier per parent node.
+However, the code can be easily updated by replacing lines 21-22 with the example shown in the tabs above.
+
+"""
+from sklearn.linear_model import LogisticRegression
+
+from hiclass import LocalClassifierPerParentNode
+
+# Define data
+X_1 = [[1], [2]]
+X_2 = [[3], [4], [5]]
+Y_1 = [
+    ["Animal", "Mammal", "Sheep"],
+    ["Animal", "Mammal", "Cow"],
+]
+Y_2 = [
+    ["Animal", "Reptile", "Snake"],
+    ["Animal", "Reptile", "Lizard"],
+    ["Animal", "Mammal", "Cow"],
+]
+X_test = [[5], [4], [3], [2], [1]]
+
+# Use logistic regression classifiers for every parent node
+# And warm_start=True to allow training with more data in the future.
+lr = LogisticRegression()
+classifier = LocalClassifierPerParentNode(local_classifier=lr, warm_start=True)
+
+# Train local classifier per parent node
+classifier.fit(X_1, Y_1)
+
+# Fit additional data
+classifier.fit(X_2, Y_2)
+
+# Predict
+predictions = classifier.predict(X_test)
+print(predictions)

hiclass/HierarchicalClassifier.py

@@ -65,6 +65,7 @@ def __init__(
         edge_list: str = None,
         replace_classifiers: bool = True,
         n_jobs: int = 1,
+        warm_start: bool = False,
         classifier_abbreviation: str = "",
     ):
         """
@@ -87,6 +88,8 @@ def __init__(
         n_jobs : int, default=1
             The number of jobs to run in parallel. Only :code:`fit` is parallelized.
             If :code:`Ray` is installed it is used, otherwise it defaults to :code:`Joblib`.
+        warm_start : bool, default=False
+            When set to `True`, reuse the solution of the previous call to fit and add more estimators for new nodes, otherwise, just fit a whole new DAG. See :ref:`Fitting additional nodes` for more information.
         classifier_abbreviation : str, default=""
             The abbreviation of the local hierarchical classifier to be displayed during logging.
         """
@@ -95,6 +98,7 @@ def __init__(
         self.edge_list = edge_list
         self.replace_classifiers = replace_classifiers
         self.n_jobs = n_jobs
+        self.warm_start = warm_start
         self.classifier_abbreviation = classifier_abbreviation
 
     def fit(self, X, y, sample_weight=None):

hiclass/LocalClassifierPerNode.py

@@ -42,6 +42,7 @@ def __init__(
         edge_list: str = None,
         replace_classifiers: bool = True,
         n_jobs: int = 1,
+        warm_start: bool = False,
     ):
         """
         Initialize a local classifier per node.
@@ -74,13 +75,16 @@ def __init__(
         n_jobs : int, default=1
             The number of jobs to run in parallel. Only :code:`fit` is parallelized.
             If :code:`Ray` is installed it is used, otherwise it defaults to :code:`Joblib`.
+        warm_start : bool, default=False
+            When set to `True`, reuse the solution of the previous call to fit and add more estimators for new nodes, otherwise, just fit a whole new DAG. See :ref:`Fitting additional nodes` for more information.
         """
         super().__init__(
             local_classifier=local_classifier,
             verbose=verbose,
             edge_list=edge_list,
             replace_classifiers=replace_classifiers,
             n_jobs=n_jobs,
+            warm_start=warm_start,
             classifier_abbreviation="LCPN",
         )
         self.binary_policy = binary_policy

hiclass/LocalClassifierPerParentNode.py

@@ -40,6 +40,7 @@ def __init__(
         edge_list: str = None,
         replace_classifiers: bool = True,
         n_jobs: int = 1,
+        warm_start: bool = False,
     ):
         """
         Initialize a local classifier per parent node.
@@ -61,13 +62,16 @@ def __init__(
         n_jobs : int, default=1
             The number of jobs to run in parallel. Only :code:`fit` is parallelized.
             If :code:`Ray` is installed it is used, otherwise it defaults to :code:`Joblib`.
+        warm_start : bool, default=False
+            When set to `True`, reuse the solution of the previous call to fit and add more estimators for new nodes, otherwise, just fit a whole new DAG. See :ref:`Fitting additional nodes` for more information.
         """
         super().__init__(
             local_classifier=local_classifier,
             verbose=verbose,
             edge_list=edge_list,
             replace_classifiers=replace_classifiers,
             n_jobs=n_jobs,
+            warm_start=warm_start,
             classifier_abbreviation="LCPPN",
         )