Merge pull request 'Use cluster strategies and support PCA' (#15) from clustering-strategy into main

Reviewed-on: #15

frontend/clusters.py

@@ -1,25 +1,40 @@
 from sklearn.cluster import DBSCAN, KMeans
 import numpy as np
+from dataclasses import dataclass
+from abc import ABC, abstractmethod
+from typing import Any, Optional
 
-class DBSCAN_cluster():
-    def __init__(self, eps, min_samples,data):
+@dataclass
+class ClusterResult:
+    labels: np.array
+    centers: Optional[np.array]
+    statistics: list[dict[str, Any]]
+
+
+class Cluster(ABC):
+    @abstractmethod
+    def run(self, data: np.array) -> ClusterResult:
+        pass
+
+
+class DBSCANCluster(Cluster):
+    def __init__(self, eps: float = 0.5, min_samples: int = 5):
         self.eps = eps
         self.min_samples = min_samples
-        self.data = data
-        self.labels = np.array([])
 
-    def run(self):
+    #@typing.override
+    def run(self, data: np.array) -> ClusterResult:
         dbscan = DBSCAN(eps=self.eps, min_samples=self.min_samples)
-        self.labels = dbscan.fit_predict(self.data)
-        return self.labels
+        labels = dbscan.fit_predict(data)
+        return ClusterResult(labels, None, self.get_statistics(data, labels))
 
-    def get_stats(self):
-        unique_labels = np.unique(self.labels)
+    def get_statistics(self, data: np.array, labels: np.array) -> list[dict[str, Any]]:
+        unique_labels = np.unique(labels)
         stats = []
         for label in unique_labels:
             if label == -1:
                 continue
-            cluster_points = self.data[self.labels == label]
+            cluster_points = data[labels == label]
             num_points = len(cluster_points)
             density = num_points / (np.max(cluster_points, axis=0) - np.min(cluster_points, axis=0)).prod()
             stats.append({
@@ -27,37 +42,42 @@ class DBSCAN_cluster():
                 "num_points": num_points,
                 "density": density
             })
-
         return stats
-    
 
-class KMeans_cluster():
-    def __init__(self, n_clusters, n_init, max_iter, data):
+    def __str__(self) -> str:
+        return "DBScan"
+
+
+class KMeansCluster(Cluster):
+    def __init__(self, n_clusters: int = 8, n_init: int = 1, max_iter: int = 300):
         self.n_clusters = n_clusters
         self.n_init = n_init
         self.max_iter = max_iter
-        self.data = data
-        self.labels = np.array([])
-        self.centers = []
 
-    def run(self):
+    #@typing.override
+    def run(self, data: np.array) -> ClusterResult:
         kmeans = KMeans(n_clusters=self.n_clusters, init="random", n_init=self.n_init, max_iter=self.max_iter, random_state=111)
-        self.labels = kmeans.fit_predict(self.data)
-        self.centers = kmeans.cluster_centers_
-        return self.labels
+        labels = kmeans.fit_predict(data)
+        centers = kmeans.cluster_centers_
+        return ClusterResult(labels, centers, self.get_statistics(data, labels, centers))
 
-
-    def get_stats(self):
-        unique_labels = np.unique(self.labels)
+    def get_statistics(self, data: np.array, labels: np.array, centers: np.array) -> list[dict[str, Any]]:
+        unique_labels = np.unique(labels)
         stats = []
 
         for label in unique_labels:
-            cluster_points = self.data[self.labels == label]
+            cluster_points = data[labels == label]
             num_points = len(cluster_points)
-            center = self.centers[label]
+            center = centers[label]
             stats.append({
-                'cluster': label,
-                'num_points': num_points,
-                'center': center
+                "cluster": label,
+                "num_points": num_points,
+                "center": center,
             })
         return stats
+
+    def __str__(self) -> str:
+        return "KMeans"
+
+
+CLUSTERING_STRATEGIES = [DBSCANCluster(), KMeansCluster()]

frontend/pages/clustering.py

@@ -0,0 +1,86 @@
+import streamlit as st
+import matplotlib.pyplot as plt
+from clusters import DBSCANCluster, KMeansCluster, CLUSTERING_STRATEGIES
+from sklearn.decomposition import PCA
+from sklearn.metrics import silhouette_score
+import numpy as np
+
+st.header("Clustering")
+
+if "data" in st.session_state:
+    data = st.session_state.data
+
+    general_row = st.columns([1, 1, 1])
+    clustering = general_row[0].selectbox("Clustering method", CLUSTERING_STRATEGIES)
+    data_name = general_row[1].multiselect("Columns", data.select_dtypes(include="number").columns)
+    n_components = general_row[2].number_input("Reduce dimensions to (PCA)", min_value=1, max_value=3, value=2)
+
+    with st.form("cluster_form"):
+        if isinstance(clustering, KMeansCluster):
+            row1 = st.columns([1, 1, 1])
+            clustering.n_clusters = row1[0].number_input("Number of clusters", min_value=1, max_value=data.shape[0], value=clustering.n_clusters)
+            clustering.n_init = row1[1].number_input("n_init", min_value=1, value=clustering.n_init)
+            clustering.max_iter = row1[2].number_input("max_iter", min_value=1, value=clustering.max_iter)
+        elif isinstance(clustering, DBSCANCluster):
+            row1 = st.columns([1, 1])
+            clustering.eps = row1[0].slider("eps", min_value=0.0001, max_value=1.0, step=0.05, value=clustering.eps)
+            clustering.min_samples = row1[1].number_input("min_samples", min_value=1, value=clustering.min_samples)
+
+        st.form_submit_button("Launch")
+
+    if len(data_name) > 0:
+        x = data[data_name].to_numpy()
+        n_components = min(n_components, len(data_name))
+        if len(data_name) > n_components:
+            pca = PCA(n_components)
+            x = pca.fit_transform(x)
+            if n_components == 2:
+                (fig, ax) = plt.subplots(figsize=(8, 8))
+                for i in range(0, pca.components_.shape[1]):
+                    ax.arrow(
+                        0,
+                        0,
+                        pca.components_[0, i],
+                        pca.components_[1, i],
+                        head_width=0.1,
+                        head_length=0.1
+                    )
+
+                    plt.text(
+                        pca.components_[0, i] + 0.05,
+                        pca.components_[1, i] + 0.05,
+                        data_name[i]
+                    )
+                circle = plt.Circle((0, 0), radius=1, edgecolor='b', facecolor='None')
+                ax.add_patch(circle)
+                plt.axis("equal")
+                ax.set_title("PCA result - Correlation circle")
+                st.pyplot(fig)
+
+        result = clustering.run(x)
+        st.write("## Cluster stats")
+        st.table(result.statistics)
+
+        st.write("## Graphical representation")
+        fig = plt.figure()
+        if n_components == 1:
+            plt.scatter(x, np.zeros_like(x))
+        elif n_components == 2:
+            ax = fig.add_subplot(projection='rectilinear')
+            plt.scatter(x[:, 0], x[:, 1], c=result.labels, s=50, cmap="viridis")
+            if result.centers is not None:
+                plt.scatter(result.centers[:, 0], result.centers[:, 1], c="black", s=200, marker="X")
+        else:
+            ax = fig.add_subplot(projection='3d')
+
+            ax.scatter(x[:, 0], x[:, 1],x[:, 2], c=result.labels, s=50, cmap="viridis")
+            if result.centers is not None:
+                ax.scatter(result.centers[:, 0], result.centers[:, 1], result.centers[:, 2], c="black", s=200, marker="X")
+        st.pyplot(fig)
+        if not (result.labels == 0).all():
+            st.write("Silhouette score:", silhouette_score(x, result.labels))
+    else:
+        st.error("Select at least one column")
+
+else:
+    st.error("file not loaded")

frontend/pages/clustering_dbscan.py

@@ -1,32 +0,0 @@
-import streamlit as st
-import matplotlib.pyplot as plt
-from clusters import DBSCAN_cluster
-
-st.header("Clustering: dbscan")
-
-if "data" in st.session_state:
-    data = st.session_state.data
-
-    with st.form("my_form"):
-        data_name = st.multiselect("Data Name", data.select_dtypes(include="number").columns, max_selections=3)
-        eps = st.slider("eps", min_value=0.0, max_value=1.0, value=0.5, step=0.01)
-        min_samples = st.number_input("min_samples", step=1, min_value=1, value=5)
-        st.form_submit_button("launch")
-
-    if len(data_name) >= 2 and len(data_name) <=3:
-        x = data[data_name].to_numpy()
-
-        dbscan = DBSCAN_cluster(eps,min_samples,x)
-        y_dbscan = dbscan.run()
-        st.table(dbscan.get_stats())
-
-        fig = plt.figure()
-        if len(data_name) == 2:
-            ax = fig.add_subplot(projection='rectilinear')
-            plt.scatter(x[:, 0], x[:, 1], c=y_dbscan, s=50, cmap="viridis")
-        else:
-            ax = fig.add_subplot(projection='3d')
-            ax.scatter(x[:, 0], x[:, 1],x[:, 2], c=y_dbscan, s=50, cmap="viridis")
-        st.pyplot(fig)
-else:
-    st.error("file not loaded")

frontend/pages/clustering_kmeans.py

@@ -1,44 +0,0 @@
-import streamlit as st
-import matplotlib.pyplot as plt
-from clusters import KMeans_cluster
-
-st.header("Clustering: kmeans")
-
-if "data" in st.session_state:
-    data = st.session_state.data
-
-    with st.form("my_form"):
-        row1 = st.columns([1,1,1])
-        n_clusters = row1[0].selectbox("Number of clusters", range(1,data.shape[0]))
-        data_name = row1[1].multiselect("Data Name",data.select_dtypes(include="number").columns, max_selections=3)
-        n_init = row1[2].number_input("n_init",step=1,min_value=1)
-
-        row2 = st.columns([1,1])
-        max_iter = row1[0].number_input("max_iter",step=1,min_value=1)
-
-
-        st.form_submit_button("launch")
-
-    if len(data_name) >= 2 and len(data_name) <=3:
-        x = data[data_name].to_numpy()
-
-        kmeans = KMeans_cluster(n_clusters, n_init, max_iter, x)
-        y_kmeans = kmeans.run()
-
-        st.table(kmeans.get_stats())
-
-        centers = kmeans.centers
-        fig = plt.figure()
-        if len(data_name) == 2:
-            ax = fig.add_subplot(projection='rectilinear')
-            plt.scatter(x[:, 0], x[:, 1], c=y_kmeans, s=50, cmap="viridis")
-            plt.scatter(centers[:, 0], centers[:, 1], c="black", s=200, marker="X")
-        else:
-            ax = fig.add_subplot(projection='3d')
-
-            ax.scatter(x[:, 0], x[:, 1],x[:, 2], c=y_kmeans, s=50, cmap="viridis")
-            ax.scatter(centers[:, 0], centers[:, 1], centers[:, 2], c="black", s=200, marker="X")
-        st.pyplot(fig)
-
-else:
-    st.error("file not loaded")