Commit e403ed5fb6202dae56d47815d5961cced00f1c85
1 parent
11ee97e2cc
Exists in
master
Add a script that allow user to evaluate a representation using classification labels.
Showing 1 changed file with 126 additions and 0 deletions Side-by-side Diff
scripts/evaluations/clustering.py
| 1 | +''' | |
| 2 | +This script allows the user to evaluate a classification system on new labels using clustering methods. | |
| 3 | +The algorithms are applied on the given latent space (embedding). | |
| 4 | +''' | |
| 5 | +import argparse | |
| 6 | +import numpy as np | |
| 7 | +import pandas as pd | |
| 8 | +import os | |
| 9 | +import time | |
| 10 | +from sklearn.preprocessing import LabelEncoder | |
| 11 | +from sklearn.metrics.pairwise import pairwise_distances | |
| 12 | +from sklearn.metrics import f1_score | |
| 13 | +from sklearn.cluster import KMeans | |
| 14 | +from sklearn.manifold import TSNE | |
| 15 | +import matplotlib.pyplot as plt | |
| 16 | + | |
| 17 | +from volia.data_io import read_features,read_lst | |
| 18 | + | |
| 19 | +if __name__ == "__main__": | |
| 20 | + # Argparse | |
| 21 | + parser = argparse.ArgumentParser("Compute clustering on a latent space") | |
| 22 | + parser.add_argument("features") | |
| 23 | + parser.add_argument("utt2", | |
| 24 | + type=str, | |
| 25 | + help="file with [utt] [value]") | |
| 26 | + parser.add_argument("--prefix", | |
| 27 | + type=str, | |
| 28 | + help="prefix of saved files") | |
| 29 | + parser.add_argument("--outdir", | |
| 30 | + default=None, | |
| 31 | + type=str, | |
| 32 | + help="Output directory") | |
| 33 | + | |
| 34 | + args = parser.parse_args() | |
| 35 | + | |
| 36 | + assert args.outdir | |
| 37 | + | |
| 38 | + start = time.time() | |
| 39 | + | |
| 40 | + # Load features and utt2 | |
| 41 | + features = read_features(args.features) | |
| 42 | + utt2 = read_lst(args.utt2) | |
| 43 | + | |
| 44 | + ids = list(features.keys()) | |
| 45 | + feats = np.vstack([ features[id_] for id_ in ids ]) | |
| 46 | + classes = [ utt2[id_] for id_ in ids ] | |
| 47 | + | |
| 48 | + # Encode labels | |
| 49 | + le = LabelEncoder() | |
| 50 | + labels = le.fit_transform(classes) | |
| 51 | + num_classes = len(le.classes_) | |
| 52 | + | |
| 53 | + # Compute KMEANS clustering on data | |
| 54 | + estimator = KMeans( | |
| 55 | + n_clusters=num_classes, | |
| 56 | + n_init=100, | |
| 57 | + tol=10-6, | |
| 58 | + algorithm="elkan" | |
| 59 | + ) | |
| 60 | + estimator.fit(feats) | |
| 61 | + print(f"Kmeans: processed {estimator.n_iter_} iterations - intertia={estimator.inertia_}") | |
| 62 | + | |
| 63 | + # contains distance to each cluster for each sample | |
| 64 | + dist_space = estimator.transform(feats) | |
| 65 | + predictions = np.argmin(dist_space, axis=1) | |
| 66 | + | |
| 67 | + # gives each cluster a name (considering most represented character) | |
| 68 | + dataframe = pd.DataFrame({ | |
| 69 | + "label": pd.Series(list(map(lambda x: le.classes_[x], labels))), | |
| 70 | + "prediction": pd.Series(predictions) | |
| 71 | + }) | |
| 72 | + | |
| 73 | + def find_cluster_name_fn(c): | |
| 74 | + mask = dataframe["prediction"] == c | |
| 75 | + return dataframe[mask]["label"].value_counts(sort=False).idxmax() | |
| 76 | + | |
| 77 | + cluster_names = list(map(find_cluster_name_fn, range(num_classes))) | |
| 78 | + predicted_labels = le.transform( | |
| 79 | + [cluster_names[pred] for pred in predictions]) | |
| 80 | + | |
| 81 | + # F-measure | |
| 82 | + fscores = f1_score(labels, predicted_labels, average=None) | |
| 83 | + fscores_str = "\n".join(map(lambda i: "{0:25s}: {1:.4f}".format(le.classes_[i], fscores[i]), range(len(fscores)))) | |
| 84 | + print(f"F1-scores for each classes:\n{fscores_str}") | |
| 85 | + print(f"Global score : {np.mean(fscores)}") | |
| 86 | + with open(os.path.join(args.outdir, args.prefix + "eval_clustering.log"), "w") as fd: | |
| 87 | + print(f"F1-scores for each classes:\n{fscores_str}", file=fd) | |
| 88 | + print(f"Global score : {np.mean(fscores)}", file=fd) | |
| 89 | + | |
| 90 | + # Process t-SNE and plot | |
| 91 | + tsne_estimator = TSNE() | |
| 92 | + embeddings = tsne_estimator.fit_transform(feats) | |
| 93 | + print("t-SNE: processed {0} iterations - KL_divergence={1:.4f}".format( | |
| 94 | + tsne_estimator.n_iter_, tsne_estimator.kl_divergence_)) | |
| 95 | + | |
| 96 | + fig, [axe1, axe2] = plt.subplots(1, 2, figsize=(10, 5)) | |
| 97 | + for c, name in enumerate(le.classes_): | |
| 98 | + c_mask = np.where(labels == c) | |
| 99 | + axe1.scatter(embeddings[c_mask][:, 0], embeddings[c_mask][:, 1], label=name, alpha=0.2, edgecolors=None) | |
| 100 | + | |
| 101 | + try: | |
| 102 | + id_cluster = cluster_names.index(name) | |
| 103 | + except ValueError: | |
| 104 | + print("WARNING: no cluster found for {}".format(name)) | |
| 105 | + continue | |
| 106 | + c_mask = np.where(predictions == id_cluster) | |
| 107 | + axe2.scatter(embeddings[c_mask][:, 0], embeddings[c_mask][:, 1], label=name, alpha=0.2, edgecolors=None) | |
| 108 | + | |
| 109 | + axe1.legend(loc="lower center", bbox_to_anchor=(0.5, -0.35)) | |
| 110 | + axe1.set_title("true labels") | |
| 111 | + axe2.legend(loc="lower center", bbox_to_anchor=(0.5, -0.35)) | |
| 112 | + axe2.set_title("predicted cluster label") | |
| 113 | + | |
| 114 | + plt.suptitle("Kmeans Clustering") | |
| 115 | + | |
| 116 | + loc = os.path.join( | |
| 117 | + args.outdir, | |
| 118 | + args.prefix + "kmeans.pdf" | |
| 119 | + ) | |
| 120 | + plt.savefig(loc, bbox_inches="tight") | |
| 121 | + plt.close() | |
| 122 | + | |
| 123 | + print("INFO: figure saved at {}".format(loc)) | |
| 124 | + | |
| 125 | + end = time.time() | |
| 126 | + print("program ended in {0:.2f} seconds".format(end-start)) |