add implementation

multiclass_brier_score.py

@@ -11,50 +11,58 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-"""TODO: Add a description here."""
+
+"""brier_score metric for multiclass problem."""
+
+import numpy as np
 
 import evaluate
 import datasets
 
 
-# TODO: Add BibTeX citation
-_CITATION = """\
-@InProceedings{huggingface:module,
-title = {A great new module},
-authors={huggingface, Inc.},
-year={2020}
+_CITATION = """
+@article{brier1950verification,
+  title={Verification of forecasts expressed in terms of probability},
+  author={Brier, Glenn W},
+  journal={Monthly weather review},
+  volume={78},
+  number={1},
+  pages={1--3},
+  year={1950}
 }
 """
 
-# TODO: Add description of the module here
-_DESCRIPTION = """\
-This new module is designed to solve this great ML task and is crafted with a lot of care.
+_DESCRIPTION = """
+Measure to compare true observed labels with predicted probabilities in multiclass classification tasks.
 """
 
 
-# TODO: Add description of the arguments of the module here
 _KWARGS_DESCRIPTION = """
-Calculates how good are predictions given some references, using certain scores
+Multiclass Brier Score: Measure to compare true observed labels with predicted probabilities in multiclass classification tasks.
 Args:
-    predictions: list of predictions to score. Each predictions
-        should be a string with tokens separated by spaces.
-    references: list of reference for each prediction. Each
-        reference should be a string with tokens separated by spaces.
+    pred_probs: array-like of shape (n_sample, m_classes).
+    references: array-like array of shape (n_sample,).
 Returns:
-    accuracy: description of the first score,
-    another_score: description of the second score,
+    brier_score: float, average brier score over all samples.
 Examples:
     Examples should be written in doctest format, and should illustrate how
     to use the function.
 
-    >>> my_new_module = evaluate.load("my_new_module")
-    >>> results = my_new_module.compute(references=[0, 1], predictions=[0, 1])
-    >>> print(results)
-    {'accuracy': 1.0}
-"""
+    >>> brier_metric = multiclass_brier_score()
+    >>> brier_score = brier_metric.compute(pred_probs=[[0.0, 1.0, 0.0]], references=[1])
+    >>> print(brier_score)
+    {'brier_score': 0.0}
 
-# TODO: Define external resources urls if needed
-BAD_WORDS_URL = "http://url/to/external/resource/bad_words.txt"
+    >>> brier_metric = multiclass_brier_score()
+    >>> brier_score = brier_metric.compute(pred_probs=[[0.1, 0.1, 0.8]], references=[2])
+    >>> print(round(brier_score['brier_score'], 2))
+    0.06
+
+    >>> brier_metric = multiclass_brier_score()
+    >>> brier_score = brier_metric.compute(pred_probs=[[0.1, 0.1, 0.8], [0.0, 1.0, 0.0]], references=[2, 1])
+    >>> print(round(brier_score['brier_score'], 2))
+    0.03
+"""
 
 
 @evaluate.utils.file_utils.add_start_docstrings(_DESCRIPTION, _KWARGS_DESCRIPTION)
@@ -71,25 +79,35 @@ class multiclass_brier_score(evaluate.Metric):
             inputs_description=_KWARGS_DESCRIPTION,
             # This defines the format of each prediction and reference
             features=datasets.Features({
-                'predictions': datasets.Value('int64'),
-                'references': datasets.Value('int64'),
+                'pred_probs': datasets.Sequence(datasets.Value("float")),
+                'references': datasets.Value('int32'),
             }),
-            # Homepage of the module for documentation
-            homepage="http://module.homepage",
             # Additional links to the codebase or references
-            codebase_urls=["http://github.com/path/to/codebase/of/new_module"],
-            reference_urls=["http://path.to.reference.url/new_module"]
+            #codebase_urls=["http://github.com/path/to/codebase/of/new_module"],
+            reference_urls=["https://search.r-project.org/CRAN/refmans/mlr3measures/html/mbrier.html"]
         )
 
-    def _download_and_prepare(self, dl_manager):
-        """Optional: download external resources useful to compute the scores"""
-        # TODO: Download external resources if needed
-        pass
 
-    def _compute(self, predictions, references):
-        """Returns the scores"""
-        # TODO: Compute the different scores of the module
-        accuracy = sum(i == j for i, j in zip(predictions, references)) / len(predictions)
+    def _compute(self, pred_probs: np.ndarray, references: np.ndarray):
+        """
+        brier_score = 1/n * sum_{i=1}^n sum_{j=1}^m (y_{ij} - p{ij})^2
+        Args:
+            pred_probs: numpy array of shape (n, m) where n is the number of samples and m is the number of classes
+            references: numpy array of shape (n,) where n is the number of samples
+        """
+        assert len(pred_probs) == len(references), "The length of the predictions and references should be the same"
+        pred_probs = np.array(pred_probs)
+        n = len(references)
+        m = pred_probs.shape[1]
+        # generate one-hot encoding for the references
+        references_onehot = np.zeros((n, m))
+        references_onehot[np.arange(n), references] = 1   # shape: (n, m)
+        brier_score = np.sum((references_onehot - pred_probs)**2) / float(n)
         return {
-            "accuracy": accuracy,
-        }
+            "brier_score": brier_score,
+        }
+
+
+if __name__ == "__main__":
+    import doctest
+    doctest.testmod()