-
Notifications
You must be signed in to change notification settings - Fork 0
/
evaluating.py
162 lines (137 loc) · 7.15 KB
/
evaluating.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
### 4.EVALUATING THE CLUSTERING METHOD
import random
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import scipy
from tqdm import tqdm
import librosa
import librosa.display
import warnings
warnings.filterwarnings('ignore')
from kneed import DataGenerator, KneeLocator
import sklearn
from sklearn.cluster import KMeans
from sklearn.model_selection import GridSearchCV
from sklearn.neighbors import NearestNeighbors
from sklearn.metrics import silhouette_samples, silhouette_score
def normalizer(df):
feature_vector = df.to_numpy()
# Create a scaler object
scaler = sklearn.preprocessing.MinMaxScaler(feature_range=(0, 1))
# Apply scaler object to normalize feature vectors between 0 and 1
features = scaler.fit_transform(feature_vector)
return features
if __name__ == "__main__":
print('Evaluating the model...')
df = pd.read_csv('dataset.csv')
df = df.dropna() # Remove NaN values
# Reset the dataframe index
df = df.reset_index(drop=True)
# Set the selected feature vectors
df_variance = df[['SRm', 'SBm', 'MFCC4', 'SCm', 'ZCRv', 'MFCC6', 'SRv', 'MFCC7', 'MFCC5','SFm', 'duration', 'ZCRm', 'MFCC8']]
df_laplacian = df[['SBm', 'SRm', 'MFCC4', 'SCm', 'MFCC5', 'MFCC7', 'SFMm', 'MFCC3','MFCC6', 'MFCC2', 'ZCRm', 'ZCRv', 'ENm']]
# Normalize the feature vectors
variance = normalizer(df_variance)
laplacian = normalizer(df_laplacian)
# Apply non-metric dimensionality reduction algorithm with t-SNE, to project the data in two dimensions
tsne = sklearn.manifold.TSNE(n_components=2, perplexity=45, init='pca', n_iter=1500, n_iter_without_progress=200, learning_rate='auto')
X_tnse_var = tsne.fit_transform(variance)
X_tnse_lap = tsne.fit_transform(laplacian)
# Find the point of maximum curvature
n_neighbors = 13*2 # Number of dimensions * 2
neighbors = NearestNeighbors(n_neighbors=n_neighbors)
neighbors_var = neighbors.fit(X_tnse_var)
distances_var, indices_var = neighbors_var.kneighbors(X_tnse_var)
distances_var = np.sort(distances_var, axis=0)
distances_var = distances_var[:,n_neighbors-1]
i = np.arange(len(distances_var))
knee_var = KneeLocator(i, distances_var, S=1, curve='convex', direction='increasing', interp_method='polynomial')
eps_var = distances_var[knee_var.knee]
neighbors_lap = neighbors.fit(X_tnse_lap)
distances_lap, indices_lap = neighbors_lap.kneighbors(X_tnse_lap)
distances_lap = np.sort(distances_lap, axis=0)
distances_lap = distances_lap[:,n_neighbors-1]
dist = np.arange(len(distances_lap))
knee_lap = KneeLocator(dist, distances_lap, S=1, curve='convex', direction='increasing', interp_method='polynomial')
eps_lap = distances_lap[knee_lap.knee]
n_samples_var = []
sihlouette_var = []
n_samples_lap = []
sihlouette_lap = []
print('Finding optimal parameters for the model...')
for samples in tqdm([20,25,30,35,40,45,50,55,60,65,70,75]):
dbscan_var = sklearn.cluster.DBSCAN(eps=eps_var, min_samples=samples)
label_var = dbscan_var.fit_predict(X_tnse_var)
dbscan_lap = sklearn.cluster.DBSCAN(eps=eps_lap, min_samples=samples)
label_lap = dbscan_lap.fit_predict(X_tnse_lap)
noise_var_index = (dbscan_var.labels_ == -1)
silhouette_per_sample_var = silhouette_samples(X_tnse_var, dbscan_var.labels_, metric='euclidean')
silhouette_of_non_noise_var = silhouette_per_sample_var[~noise_var_index]
n_samples_var.append(samples)
sihlouette_var.append(silhouette_of_non_noise_var.mean())
noise_lap_index = (dbscan_lap.labels_ == -1)
silhouette_per_sample_lap = silhouette_samples(X_tnse_lap, dbscan_lap.labels_, metric='euclidean')
silhouette_of_non_noise_lap = silhouette_per_sample_lap[~noise_lap_index]
n_samples_lap.append(samples)
sihlouette_lap.append(silhouette_of_non_noise_lap.mean())
# Plot the results
fig, ax = plt.subplots(figsize=(10,10))
plt.plot(n_samples_var, sihlouette_var, linewidth=3, label='Variance')
plt.plot(n_samples_lap, sihlouette_lap, linewidth=3, label='Laplacian')
plt.xlabel('Minimum number of samples', fontsize=25)
plt.ylabel('Silhouette scores', fontsize=25)
plt.legend(fontsize=20)
plt.xticks(fontsize=20)
plt.yticks(fontsize=20)
plt.show()
plt.savefig('silhouette.jpg')
# Find the optimal value for min_samples
if max(sihlouette_var) > max(sihlouette_lap):
idx = np.argsort(sihlouette_var)[-1] # maximum silhouette score index
cluster = sklearn.cluster.DBSCAN(eps=eps_var, min_samples=n_samples_var[idx]).fit(X_tnse_var)
X_tnse = X_tnse_var
selection = 'Variance'
else:
idx = np.argsort(sihlouette_lap)[-1] # maximum silhouette score index
cluster = sklearn.cluster.DBSCAN(eps=eps_lap, min_samples=n_samples_lap[idx]).fit(X_tnse_lap)
X_tnse = X_tnse_lap
selection = 'Laplacian'
print(f"Number of bird syllables clusters found: {np.unique(cluster.labels_).size-1}")
print(f"Number of noisy samples {len(cluster.labels_[cluster.labels_==-1])} out of a total of {len(cluster.labels_)} samples")
# Compute scores (output is in range 0 to 1 where 1 is best)
noise_features_index = (cluster.labels_ == -1)
silhouette_per_sample = silhouette_samples(X_tnse, cluster.labels_, metric='euclidean')
silhouette_of_non_noise_samples = silhouette_per_sample[~noise_features_index]
print(f'\nSilhouette Coefficient score: {silhouette_of_non_noise_samples.mean()}')
for lab_val in np.unique(cluster.labels_):
if lab_val == -1: # displaying noisy samples in white
ax.scatter(X_tnse[cluster.labels_==lab_val, 0],
X_tnse[cluster.labels_==lab_val, 1], c='gray', alpha=0.1, label='Noise')
ax.legend()
else:
rgb = (random.random(), random.random(), random.random())
ax.scatter(X_tnse[cluster.labels_==lab_val, 0], X_tnse[cluster.labels_==lab_val, 1], c=[rgb], alpha=0.8)
ax.set_title(f'DBSCAN clustering using {selection} vector', fontsize=25)
ax.set_xlabel('t-SNE component 1', fontsize=20)
ax.set_ylabel('t-SNE component 2', fontsize=20)
plt.xticks(fontsize=20)
plt.yticks(fontsize=20)
plt.legend(fontsize=20)
plt.show()
plt.savefig('dbscan.jpg')
# Plot the bird syllables spectrograms
zero = np.asarray(np.where(cluster.labels_ == 0)).flatten()[10:17]
one = np.asarray(np.where(cluster.labels_ == 1)).flatten()[10:17]
indexes = np.append(zero, one)
two = np.asarray(np.where(cluster.labels_ == 2)).flatten()[10:17]
indexes = np.append(indexes, two)
three = np.asarray(np.where(cluster.labels_ == 3)).flatten()[10:17]
indexes = np.append(indexes, three)
fig, axes = plt.subplots(4,7, subplot_kw={'xticks':(), 'yticks':()}, figsize=(10,10))
for i, ax in zip(indexes, axes.ravel()):
y, sr = librosa.load(df['file-name'][i], sr=22050)
S = np.abs(librosa.stft(y[df.start[i]:df.end[i]], n_fft=128))
ax.imshow(librosa.amplitude_to_db(S, ref=np.max), cmap=plt.cm.binary, aspect='auto', norm=None, vmax=0, vmin=-80)
ax.invert_yaxis()
plt.savefig('repertoire.jpg')