import manager from knn_graph_clustering import Cluster c = manager.Client(False) b = c.load_basket_pickle('UrbanSound8K') b2 = c.load_basket_pickle('ESC-50.pkl') #cluster1 = Cluster(basket=b, k_nn=20) #cluster1.run(feature='fusion') #cluster1.plot() cluster1 = Cluster(basket=b, k_nn=20) scores_text = [] for k in [5, 6, 7, 8, 9, 10, 12, 14, 15, 20]: cluster1.run(feature='text', k_nn=k) scores_text.append(cluster1.scores) scores_acoustic = [] for k in [5, 6, 7, 8, 9, 10, 12, 14, 15, 20]: cluster1.run(feature='acoustic', k_nn=k) scores_acoustic.append(cluster1.scores) scores_fusion = [] for k in [5, 6, 7, 8, 9, 10, 12, 14, 15, 20]: cluster1.run(feature='fusion', k_nn=k) scores_fusion.append(cluster1.scores) print scores_text print scores_acoustic print scores_fusion
import manager from scipy.spatial.distance import pdist from sklearn.metrics.pairwise import euclidean_distances import webbrowser c = manager.Client() b = c.load_basket_pickle( 'UrbanSound8K') # Can load a basket from a search result instead k_nn = 10 # param for k-nn graph creation # __________________ FEATURE __________________ # # Extract features and create similarity matrix from: # Acoustic descriptors b.analysis_stats = [None] * len( b ) # this is because the basket is old and now analysis_stats contains None values initialy b.add_analysis_stats() b.remove_sounds_with_no_analysis() d = b.extract_descriptor_stats(scale=True) sound_similarity_matrix_d = euclidean_distances(d) sound_similarity_matrix_d = sound_similarity_matrix_d / sound_similarity_matrix_d.max( ) sound_similarity_matrix_d = 1 - sound_similarity_matrix_d # Tags t = b.preprocessing_tag() for idx, tt in enumerate(t): b.sounds[idx].tags = tt nlp = manager.Nlp(b) nlp.create_sound_tag_matrix()
def __init__(self): self.c = manager.Client()
def __init__(self, query, descriptor): self.c = manager.Client() self.query = query self.descriptor_name = descriptor # TODO : A way to load several descriptor