def test_forward_propagate(self):
learning_rate = 0.8
structure = {'num_inputs': 2, 'num_outputs': 1, 'num_hidden': 1}
candidate = NeuralNet(structure, learning_rate)
x = np.array([1, 0])
cand_out = candidate.forward_propagate(x)
expected_result = .500615025728
print(cand_out)
self.assertAlmostEqual(cand_out, expected_result, 4)
count = 0
min_values = np.load('min_vals.npy')
max_values = np.load('max_vals.npy')
for analyzed in testAnalysis:
testFeature = []
for key in keys:
if (key != 'mode'):
v = normalize(analyzed[key], min_values[key], max_values[key])
else:
v = analyzed[key]
testFeature.append(v)
testInput[count] = testFeature
count += 1
for i in xrange(len(testTracks)):
output = candidate.forward_propagate(testInput[i], 1)
temp = [j[0] for j in sorted(enumerate(output), key=lambda x: x[1])]
temp.reverse()
if output[temp[1]] > 0.5:
print temp[0], temp[1]
else:
print temp[0]
print testTracks[i]
else:
print("Can't get token for", username)
def predict(self, query):
# try:
username = '******'
#token = util.prompt_for_user_token(username)
# change this back
# token = "BQDgUh4f6qHjBHmuoRRSt07S7Z_98wtK8ImZkmBvcSWj5ItAdpus3xRu1udx7jPhuG2GemnYtUBnb2xJNoS7QEG2jrGknc815NjOI6rmfcDNxf-6LpiceI78MrnydEhWAc5vd8fx7bI"
token = "BQBUVzk5gnZP-AZSZw8okcUtxeYxys5nZQb5MlKFBYov4XXfwtnN-ukLBm4bb-8IAzLiHw9mTZ_VtKGTaMom0a5hATGiqUiYULguQez8Jjkwln94En_RJNOn-btiAoUijmoZ_Evgw_Q"
sp = spotipy.Spotify(token)
# check for errors here
first = sp.search(query)['tracks']['items'][0]
artist_name = first['artists'][0]['name']
artist_id = first['artists'][0]['id']
song_name = first['name']
searchRes = first['uri']
# call NN with input_features
structure = {'num_inputs': 10, 'num_hidden': 30, 'num_outputs': 8}
learning_rate = 0.3
candidate = NeuralNet(structure, learning_rate)
keys = [
'energy', 'liveness', 'tempo', 'speechiness', 'acousticness',
'instrumentalness', 'danceability', 'loudness', 'valence', 'mode'
]
min_vals = np.load('min_vals.npy').item()
max_vals = np.load('max_vals.npy').item()
testTracks = [searchRes]
testAnalysis = sp.audio_features(tracks=testTracks)
if testAnalysis == None:
return ["Oops, Spotify doesn't recognize that song"]
testInput = np.zeros([len(testTracks), 10])
rowI = 0
for analyzed in testAnalysis:
feature_list = []
for key in keys:
if analyzed[key] == None:
break
v = self.normalize(analyzed[key], min_vals[key], max_vals[key])
feature_list.append(v)
if len(feature_list) == 10:
testInput[rowI] = feature_list
rowI += 1
while rowI < (testInput.shape)[0]:
testInput = np.delete(testInput, rowI, 0)
if len(testInput) == 0:
return ["Oops, we don't recognize that song"]
result = None
for i in xrange(len(testTracks)):
output = candidate.forward_propagate(testInput[i], 1)
temp = [
j[0] for j in sorted(enumerate(output), key=lambda x: x[1])
]
temp.reverse()
if output[temp[1]] > 0.5:
if (temp[0] == 0):
if (output[0] > 0.99):
result = [
self.genre_map[temp[0]], self.genre_map[temp[1]]
]
else:
result = [
self.genre_map[temp[1]]
] #[self.genre_map[temp[0]], self.genre_map[temp[1]]]
else:
result = [self.genre_map[temp[0]], self.genre_map[temp[1]]]
else:
result = [self.genre_map[temp[0]]]
seed_genres = []
for genre in result:
if "metal" in genre:
seed_genres.append("heavy-metal")
seed_genres.append("rock")
elif "indie" in genre:
seed_genres.append("indie")
seed_genres.append("alternative")
else:
seed_genres.append(genre)
recommendations = sp.recommendations(seed_genres=seed_genres,
seed_artists=[artist_id])
suggestions = []
for track in (recommendations['tracks'])[:5]:
suggestion = {
'name': track['name'],
'artist': track['artists'][0]['name']
}
suggestions.append(suggestion)
return [result, (song_name, artist_name), suggestions]
