def run(self):
emotion_scores = []
emotion_rec = []
relevances = []
# Loading test data and representing them in the feature space.
write("[+] Loading items:".ljust(54,'.')) if self.debug else ''
if self.items is None:
self.items = json.loads(open(self.input_path).read())
test_data, starts, ends = self.get_text()
X_test = []
count = 0
for i, sentences in enumerate(test_data):
emotion_rec.append([])
relevances.append([])
emotion_scores.append([0. for k in range(12)])
cur_x = lil_matrix((len(sentences), len(self.dictionary.keys())))
for j, sentence in enumerate(sentences):
relevances[i].append([0. for k in range(12)])
sword_vector = self.vectorize_document(sentence)
sfeature_vector = self.feature_extraction(sword_vector)
for key, val in sfeature_vector:
cur_x[j,key] = val
count += 1
X_test.append(cur_x)
write("[OK]\n") if self.debug else ''
# Training and predicting each emotion class on the given data.
write("[+] Modeling emotions:") if self.debug else ''
for num_class in range(12):
self.num_class = num_class + 1
cur_X = self.X
cur_Y = self.Y[:,(self.num_class - 1):self.num_class]
best = None
dataset_name = "ted_talks"
method = "ap_weights"
rating_class = self.rating_classes[self.num_class-1]
opt = []
for b in ['_m1','','_m3']:
tmp = open('parameters/%d/%s%s.txt' % (self.num_class,method,b))
best = float(tmp.readlines()[0].split(": ")[1].split("}")[0])
opt.append(best)
try:
write(("\n" + " "*8 + "-> %s" % rating_class).ljust(55,'.')) if self.debug else ''
f = gzip.open("models/%s_model.pk" % rating_class, 'rb')
self.model = pickle.load(f)
write("[OK]") if self.debug else ''
except:
write("\n") if self.debug else ''
self.model = APWeights(20, l1=opt[0], l2=opt[1], l3=opt[2], reg=self)
self.model.fit(cur_X, cur_Y)
f = gzip.open("models/%s_model.pk" % rating_class,"wb")
pickle.dump(self.model, f)
pred = self.model.predict(X_test)
for j, val in enumerate(pred):
emotion_scores[j][num_class] = val.view(np.ndarray)[0]
for j, weights in enumerate(self.model.P_test):
for i, w in enumerate(weights):
relevances[j][i][num_class] = w
# Compute emotion-based recommendations
write("\n[+] Generating recommendations".ljust(55,'.')) if self.debug else ''
sim = np.zeros((len(X_test), len(X_test)))
for i, v1 in enumerate(emotion_scores):
for j, v2 in enumerate(emotion_scores):
sim[i][j] = self.cosine_measure(v1, v2)
write("[OK]") if self.debug else ''
real_idxs = [nid for nid in range(len(X_test))]
# Write results into a file
write("\n[+] Saving to output file".ljust(55,'.')) if self.debug else ''
output = self.items
for j, h in enumerate(self.items):
segments = test_data[j]
h['segments'] = []
for i,seg in enumerate(segments):
seg_h = {'text':seg,
'start':starts[j][i],
'end':ends[j][i],
'relevance_scores':relevances[j][i]}
h['segments'].append(seg_h)
top, top_sim = self.n_most_similar(j, sim, real_idxs)
h['emotion_classes'] = self.rating_classes
h['emotion_scores'] = emotion_scores[j]
h['emotion_rec'] = [self.items[idx]['id'] for idx in top]
h['emotion_rec_scores'] = top_sim
json.dump(output, open(self.output_path,"wb"))
write("[OK]\n") if self.debug else ''
print "[x] Finished."
