From 27e18751c4382268e634237e8d78e38a124acfe7 Mon Sep 17 00:00:00 2001
From: matthieu <matthieu.gouel@gmail.com>
Date: Thu, 8 Aug 2019 19:17:35 +0200
Subject: [PATCH] feat(classifiers): implement KNN algorithm

---
 README.md                     |  1 +
 alchina/classifiers/knn.py    | 53 +++++++++++++++++++++++++++
 tests/classifiers/test_knn.py | 67 +++++++++++++++++++++++++++++++++++
 3 files changed, 121 insertions(+)
 create mode 100644 alchina/classifiers/knn.py
 create mode 100644 tests/classifiers/test_knn.py

diff --git a/README.md b/README.md
index 39f95e9..98a1ad3 100644
--- a/README.md
+++ b/README.md
@@ -19,6 +19,7 @@ Alchina is a Machine Learning framework.
 
 - Linear classifier
 - Ridge classifier
+- K-Nearest Neighbors
 
 **Clusters**
 
diff --git a/alchina/classifiers/knn.py b/alchina/classifiers/knn.py
new file mode 100644
index 0000000..40b5f99
--- /dev/null
+++ b/alchina/classifiers/knn.py
@@ -0,0 +1,53 @@
+"""K-Nearest Neighbors"""
+
+import numpy as np
+
+from collections import Counter
+
+from alchina.exceptions import NotFitted
+from alchina.metrics import accuracy_score
+
+
+class KNNClassifier(object):
+    """K-Nearest Neighbors algorithm"""
+
+    def __init__(self, n_neighbors=3):
+        self.n_neighbors = n_neighbors
+
+        self.X_fit = None
+        self.y_fit = None
+
+    def euclidian(self, a, b):
+        """Compute the euclidian distance between two samples."""
+        return np.linalg.norm(a - b)
+
+    def fit(self, X, y):
+        """Train the model."""
+        self.X_fit = X
+        self.y_fit = y
+
+    def predict(self, X):
+        """Predict a target given features."""
+        if self.X_fit is None or self.y_fit is None:
+            raise NotFitted("the model must be fitted before usage")
+
+        labels = []
+        for x in X:
+            distances_labels = [
+                (self.euclidian(x, x_fit), y_fit)
+                for x_fit, y_fit in zip(self.X_fit, self.y_fit)
+            ]
+            neighbors = sorted(distances_labels, key=lambda d: d[0])[: self.n_neighbors]
+            neighbors_labels = [neighbor[1][0] for neighbor in neighbors]
+            labels.append(
+                sorted(
+                    neighbors_labels, key=Counter(neighbors_labels).get, reverse=True
+                )[0]
+            )
+        return np.array(labels).reshape(-1, 1)
+
+    def score(self, X, y):
+        """Score of the model."""
+        if self.X_fit is None or self.y_fit is None:
+            raise NotFitted("the model must be fitted before usage")
+        return accuracy_score(self.predict(X), y)
diff --git a/tests/classifiers/test_knn.py b/tests/classifiers/test_knn.py
new file mode 100644
index 0000000..45dffdf
--- /dev/null
+++ b/tests/classifiers/test_knn.py
@@ -0,0 +1,67 @@
+"""K-Nearest Neighbors tests."""
+
+import numpy as np
+import pytest
+
+from alchina.classifiers import KNNClassifier
+from alchina.exceptions import NotFitted
+
+
+# --- Linear classifier ---
+
+
+def test_knn_classifier():
+    """Test of `KNNClassifier` class."""
+    knn = KNNClassifier(1)
+
+    X = np.array([[0], [1]])
+    y = np.array([[0], [1]])
+
+    knn.fit(X, y)
+
+    assert knn.score(X, y) == 1
+
+
+def test_knn_classifier_predict():
+    """Test of `KNNClassifier` class with a prediction."""
+    knn = KNNClassifier(1)
+
+    X = np.array([[0], [1]])
+    y = np.array([[0], [1]])
+
+    knn.fit(X, y)
+
+    assert np.equal(knn.predict(np.array([0])), np.array([0]))
+
+
+def test_knn_classifier_multiclass():
+    """Test of `LinearClassifier` with multiclass."""
+    knn = KNNClassifier(1)
+
+    X = np.array([[0], [1], [2]])
+    y = np.array([[0], [1], [2]])
+
+    knn.fit(X, y)
+
+    assert knn.score(X, y) == 1
+
+
+def test_knn_classifier_predict_not_fitted():
+    """Test of `KNNClassifier` class with prediction without fit."""
+    knn = KNNClassifier(1)
+
+    X = np.array([[0], [1]])
+
+    with pytest.raises(NotFitted):
+        knn.predict(X)
+
+
+def test_knn_classifier_score_not_fitted():
+    """Test of `KNNClassifier` class with score calculation without fit."""
+    knn = KNNClassifier(1)
+
+    X = np.array([[0], [1]])
+    y = np.array([[0], [1]])
+
+    with pytest.raises(NotFitted):
+        knn.score(X, y) == 1