# -*- coding: utf-8 -*-
"""
Created on Wed Dec 11 20:53:36 2019

@author: nicol
"""

import spotipy
import spotipy.util as util
from spotipy.oauth2 import SpotifyClientCredentials
from sklearn.manifold import TSNE
import matplotlib.pyplot as plt
from sklearn.cluster import DBSCAN
import pandas as pd
import numpy as np
from mpl_toolkits.mplot3d import Axes3D

# Fonction pour faire des "radar charts stylees"
def radar_chart(r,theta):
    angles = np.linspace(0,2*np.pi,len(theta),endpoint=False)
    r = np.concatenate((r,[r[0]]))
    angles = np.concatenate((angles,angles[[0]]))
    fig = plt.figure()
    ax = fig.add_subplot(111, polar=True)
    ax.plot(angles, r, 'o-', linewidth=2)
    ax.fill(angles, r, alpha=0.25)
    ax.set_thetagrids(angles * 180/np.pi, theta)
    ax.grid(True)

# Recup du data sur spotify, True si il faut le faire, False sinon
# Nous on a déjà le fichier des features donc pas besoin de recommencer
if False:
    cid =  'ea8db4c94aa744588a4dd0621d45e05e'
    secret = '26b39cad02aa416c86db8fc2307118b8'
    username = "dour"
    client_credentials_manager = SpotifyClientCredentials(client_id=cid, client_secret=secret)
    sp = spotipy.Spotify(client_credentials_manager=client_credentials_manager)# Get read access to your library
    auth = spotipy.oauth2.SpotifyClientCredentials(client_id = cid, client_secret=secret)
    token = auth.get_access_token()
    spot = spotipy.Spotify(auth=token)
    playlist = []
    audio_features = []
    artist = []
    for i in range(9):
        p = (spot.user_playlist_tracks(cid, 'spotify:playlist:2tKVxEhCfeW275FzQqOWg0', fields='items', limit=100,offset=i*100))['items']
        for j in range(len(p)):
            audio_features.append(spot.audio_features(p[j]['track']['id']))
            artist.append([p[j]['track']['name'],p[j]['track']['popularity']])

    columns = ['artist'].append(list(audio_features[0][0].keys()))
    for i in range(len(audio_features)):
        audio_features[i] = list(audio_features[i][0].values())

    audio = pd.DataFrame(audio_features,columns=['danceability', 'energy', 'key', 'loudness', 'mode', 'speechiness',
       'acousticness', 'instrumentalness', 'liveness', 'valence', 'tempo',
       'type', 'id', 'uri', 'track_href', 'analysis_url', 'duration_ms',
       'time_signature'])


# Au final sont intéressant : danceability, energy, loudness, mode, liveness, valence, tempo
features = ['danceability','energy','loudness','mode','liveness','valence','tempo']
audio_f = audio[features]
a = audio_f.values
# TSNE sur le data pour le rendre mieux en 3D
a_red = TSNE(n_components=2).fit_transform(a)
# Clustering en utilisant DBSCAN mais ici fait sur le data en dimensionnalité déjà réduite
# => mieux de le faire sur a et pas a_red, mais marche pas assez
clustering = DBSCAN(eps=3,min_samples=8).fit(a_red)

# Plot 3D du TSNE
core_samples_mask = np.zeros_like(clustering.labels_, dtype=bool)
core_samples_mask[clustering.core_sample_indices_] = True

labels = clustering.labels_
unique_labels = set(labels)


colors = [plt.cm.Spectral(each)
          for each in np.linspace(0, 1, len(unique_labels))]

fig = plt.figure()
ax = fig.add_subplot(111)

for k, col in zip(unique_labels, colors):
    if k == -1:
        # Black used for noise.
        col = [0, 0, 0, 1]

    class_member_mask = (labels == k)

    xy = a_red[class_member_mask & core_samples_mask]
    ax.plot(xy[:, 0], xy[:, 1], 'o', markerfacecolor=tuple(col),
             markeredgecolor='k', markersize=14)
fig.show()



audio_c = audio_f
audio_c['cluster'] = labels
# plot des distributions de chaque donnée audio pour chaque cluster
#for i in range(len(features)):
#    fig,a = plt.subplots(2,4)
#    a = a.ravel()
#    b = enumerate(a)
#    bins = np.linspace(min(audio_c[features[i]]), max(audio_c[features[i]]),10)
#    for idx,ax in b:
#        l = audio_c[features[i]].where(audio_c['cluster']==idx)
#        ax.hist(l[~np.isnan(l)],bins=bins)
#        ax.set_xlim(min(audio_c[features[i]]), max(audio_c[features[i]]))
#
## radar chart pour chaque cluster de toutes les données pour voir les
## différences entre chaque cluster
#for lab in unique_labels:
#    l = audio_c.where(audio_c['cluster'] == lab)
#    l = l.dropna()
#    l = l.drop('cluster',axis=1)
#    r=np.array(np.mean(l.values,axis=0))
#    r = (r-np.min(audio_c.values[:,:7],axis=0))/(np.max(audio_c.values[:,:7],axis=0)-np.min(audio_c.values[:,:7],axis=0))
#    theta=l.columns
#    radar_chart(r,theta)