kmeans clustering

local/classification_kmeans.py

+# -*- coding: utf-8 -*-
+
+"""
+{Description}
+{License_info}
+"""
+
+import os, sys
+import numpy as np
+import json
+import matplotlib
+import matplotlib.pyplot as plt
+from sklearn.metrics.pairwise import cosine_similarity
+from sklearn.metrics.pairwise import euclidean_distances
+from sklearn.metrics import pairwise_distances
+import faiss
+from sklearn.metrics import ConfusionMatrixDisplay
+from sklearn.metrics import accuracy_score
+from sklearn.metrics import f1_score
+from sklearn.metrics import recall_score
+from sklearn.metrics import precision_score
+from sklearn.metrics import precision_recall_fscore_support
+from itertools import permutations
+
+plt.rcParams.update({'font.size': 15})
+
+def find_best_map(original_map, y_true, spks, distances):
+    print("-- (INFO) best map between y_pred and y_true (based on accuracy)")
+    keys, values = zip(*original_map.items())
+    value_perms = permutations(values)
+    best_map = None
+    best_accuracy = -np.inf
+    for value_perm in value_perms:
+        temp_map = dict(zip(keys, value_perm))
+        y_pred = []
+        col = []
+        for idx, (spk, distance) in enumerate(zip(spks, distances)):
+            clus_idx = np.argmin(distance)
+            y_pred.append(temp_map[clus_idx])
+        accuracy = accuracy_score(y_true, y_pred)  
+        if accuracy > best_accuracy:
+            best_map = temp_map
+            best_accuracy = accuracy
+    print(f'Best map: {best_map}')
+    print(f'Best Accuracy: {best_accuracy}')
+    return best_map
+
+def plot_CM(y_true, y_pred, class_names):
+    np.set_printoptions(precision=2)
+    # Plot non-normalized confusion matrix
+    titles_options = [
+    ("confusion matrix without normalization", None),
+    ("normalized confusion matrix", "true"),
+    ]
+    for title, normalize in titles_options:
+        disp = ConfusionMatrixDisplay.from_predictions(
+            y_true,
+            y_pred,
+            display_labels=class_names,
+            labels=class_names,
+            cmap=plt.cm.Blues,
+            normalize=normalize,
+            colorbar=False,
+            values_format='.2f'
+        )
+        plt.xticks(rotation=90)
+        #disp.ax_.set_title(title)
+        print(f"-- (INFO) {title}")
+        print(disp.confusion_matrix)
+    return disp
+
+
+def main(X_path, y_path, pcaA_path, pcab_path, out_path, json_path, NC):
+    # load json ... 
+    # dict with {"corpusA_speakingstyle": [spk1 spk2,...], 
+    #            "corpusB_speakingstyle": [spk1 spk2,...], ...}
+    with open(json_path, 'r') as f:
+        corpora = json.load(f)
+
+    X = np.load(X_path)
+    spks = np.load(y_path, allow_pickle=True)
+
+    A = np.load(pcaA_path)
+    b = np.load(pcab_path)
+    print('\nspkIDs:\n{}'.format(spks))
+    print(f'\ninput matrix X (shape: {np.shape(X)}):\n{X}')
+    print(f'\nprojection matrix A (shape: {np.shape(A)}):\n{A}')
+    
+    # PROJECTION COMBINATIONS
+    proj_del = {2: ['PCA1', 'PCA2'],
+                1: ['PCA1', 'PCA3'],
+                0: ['PCA2', 'PCA3'],
+                'none': ['PCA1', 'PCA2', 'PCA3']}
+
+    y_true = []
+    for key, spkIDs in corpora.items():
+        count = len(spkIDs)
+        y_true.append([key]*count)
+    y_true = [item for sublist in y_true for item in sublist]
+    
+    y_pred_map = {}
+    for idx, key in enumerate(np.unique(y_true)):
+        y_pred_map[idx] = key
+
+    for pdel in proj_del.keys():
+        print("\n----\nProject with PCA dimensions {}\n----".format(proj_del[pdel]))
+        X_proj3D = np.dot(X, A) + b # [styles x 320^2] * [320^2 x 2] or [styles x corpora] + [corpora x 2]
+        if pdel == 'none':
+            X_proj = X_proj3D
+        else:
+            X_proj = np.delete(X_proj3D, pdel, 1)
+
+        print("... kmeans ...")
+        d = X_proj3D.shape[1]
+        kmeans = faiss.Kmeans(
+            d,
+            int(NC),
+            niter=10,
+            verbose=True,
+            gpu=False,
+        )
+        kmeans.train(X_proj3D)
+
+        if pdel == 'none':
+            centroids = kmeans.centroids
+        else:
+            centroids = np.delete(kmeans.centroids, pdel, 1)
+        print('-- (INFO) centroids (3D)\n{}'.format(kmeans.centroids))
+        print('-- (INFO) centroids (without dimension {})\n{}'.format(pdel, centroids))
+        print('-- (INFO) centroids shape: {}'.format(centroids.shape))
+
+        col, y_pred = [], []
+        col_map = {0: 'blue', 1: 'red', 2: 'green', 3: 'yellow', 4: 'purple', 5: 'olive', 6: 'pink'}
+        distances = pairwise_distances(X_proj, centroids, metric='euclidean')
+        y_pred_map = find_best_map(y_pred_map, y_true, spks, distances)
+        print("-- (INFO) resulting distances to k-means centroids")
+        for idx, (spk, distance) in enumerate(zip(spks, distances)):
+            print(f"{spk}: {distance}")
+            clus_idx = np.argmin(distance)
+            y_pred.append(y_pred_map[clus_idx])
+            col.append(col_map[clus_idx])
+
+        fig, ax = plt.subplots(figsize=(30,30))
+        print('-- (INFO) y_true:\n{}\n-- (INFO) y_pred:\n{}'.format(y_true,y_pred))
+        disp = plot_CM(y_true, y_pred, list(y_pred_map.values()))
+        ax = disp.ax_
+        for text in ax.texts:
+            value = float(text.get_text())
+            if value == 0.00:
+                text.set_text('0')
+            else:
+                text.set_text('{:.2f}'.format(value))
+        plt.tight_layout()
+        PCAlabel = '_'.join([s for s in proj_del[pdel]])
+        title='CM_{}_{}_nclus{}.png'.format(X_path.split('/')[-1].replace('.npy',''), PCAlabel, NC)
+        plt.savefig(os.path.join(out_path,title))
+
+        print("-- (INFO) summary scores")
+        print('Accuracy: {}'.format(accuracy_score(y_true, y_pred)))
+        print('Precision-score: {}'.format(precision_score(y_true, y_pred, average='weighted')))
+        print('Recall-score: {}'.format(recall_score(y_true, y_pred, average='weighted')))
+        print('F1-score: {}'.format(f1_score(y_true, y_pred, average='weighted')))
+
+if __name__ == "__main__":
+    ############################# FEATURES PATH ###########################
+    try:
+        X_path = sys.argv[1]
+        print("input matrix path is: " + X_path)
+    except:
+        print("ERROR: X_path not specified")
+    ############################# LABELS PATH ###########################
+    try:
+        y_path = sys.argv[2]
+        print("label vector path is: " + y_path)
+    except:
+        print("ERROR: y_path not specified")
+    ############################# PCA A PATH ###########################
+    try:
+        pcaA_path = sys.argv[3]
+        print("pca matrix A path is: " + pcaA_path)
+    except:
+        print("ERROR: pcaA_path not specified")
+    ############################# PCA b PATH ###########################
+    try:
+        pcab_path = sys.argv[4]
+        print("pca vector b path is: " + pcab_path)
+    except:
+        print("ERROR: pcab_path not specified")
+    ############################# OUT PATH ###########################
+    try:
+        out_path = sys.argv[5]
+        print("output path is: " + out_path)
+    except:
+        print("ERROR: out_path not specified")
+    ############################# JSON PATH ###########################
+    try:
+        json_path = sys.argv[6]
+        print("json path is: " + json_path)
+    except:
+        print("ERROR: json_path not specified")
+    ############################# NUMBER CLUSTERS ###########################
+    try:
+        NC = sys.argv[7]
+        print("NC (Number Clusters) is: " + NC)
+    except:
+        print("ERROR: NC (Number Clusters) not specified")
+
+    main(X_path, y_path, pcaA_path, pcab_path, out_path, json_path, NC)
\ No newline at end of file

run.sh

@@ -55,6 +55,7 @@ if [ $stage == 1 ]  || [ $stage == 0 ]; then
                                  --DATA_dir ${runDATA}
 fi
 
+## COUNT CODEBOOK USAGE
 if [ $stage == 2 ] || [ $stage == 0 ]; then
     printf "\n... Count frequencies of codebooks ...\n"
     python3 local/codebook_freqs.py $exp_dir \
@@ -67,6 +68,7 @@ if [ $stage == 2 ] || [ $stage == 0 ]; then
                                              $exp_dir/data/${runDATA}.json
 fi
 
+## CALCULATE SIMILARITY MATRIX
 if [ $stage == 3 ] || [ $stage == 0 ]; then
     printf "\n... Similarity Matrix ...\n"
     python3 local/similarity_matrix.py \
@@ -76,6 +78,7 @@ if [ $stage == 3 ] || [ $stage == 0 ]; then
             $exp_dir/data/${runDATA}.json
 fi
 
+## PCA SPACE AND PLOTS
 if [ $stage == 4 ] || [ $stage == 0 ]; then
     printf "\n... PCA of similarity matrix ...\n"
     python3 $FAIRSEQ/examples/wav2vec/unsupervised/scripts/pca.py \
@@ -91,3 +94,15 @@ if [ $stage == 4 ] || [ $stage == 0 ]; then
         $exp_dir/plots \
         $exp_dir/data/${runDATA}.json
 fi
+
+if [ $stage == 5 ] || [ $stage == 0 ]; then
+    printf "\n... Classification with k-means in PCA space ...\n"
+    python3 local/classification_kmeans.py \
+        $exp_dir/numpy/similarity_matrix.npy \
+        $exp_dir/numpy/splits_labels.npy \
+        $exp_dir/numpy/pca/3_pca_A.npy \
+        $exp_dir/numpy/pca/3_pca_b.npy \
+        $exp_dir/plots \
+        $exp_dir/data/${runDATA}.json \
+        2
+fi
\ No newline at end of file