def context_history(self):
net_input = ContextTracker.sentence_processor()
net_utilities = utils.create_emb_layer()
emotion_net = sen_model.Net(net_utilities, num_classes=7)
act_net = sen_model.Net(net_utilities, num_classes=4)
emotion = emotion_net(net_input)
act = act_net(net_input)
sal_finder = salience.SalienceFinder(e_tag=emotion, a_tag=act)
google_response_string = ""
max_key, max_val = sal_finder.process(google_response_string)
word_2_vec = ContextTracker.load_glove_model(GLOVE_PATH)
if not self.context_hist:
self.context_hist.append(max_key)
else:
if ContextTracker.cosine_distance(self.context_hist[-1], max_key, word_2_vec) > 0.5:
flag = False
for x in enumerate(self.context_hist[:-1]):
if ContextTracker.cosine_distance(self, x, max_key, word_2_vec) < 0.5:
self.context_hist.append(x)
flag = True
if not flag:
self.context_hist.append(max_key)
else:
self.context_hist.append(self.context_hist[-1])
if len(self.context_hist) > 50:
self.context_hist = self.context_hist[25:-1]
return self.context_hist, emotion, act, word_2_vec
"""
This function creates the embedding layer from the dataset weight_matrix
:return: emb_layer ---> the embedding layer
num_embeddings ---> the length of the dataset vocabulary
embedding_dim ---> the dimension of each embedding
"""
# weights_matrix = preprocess_dataset()
weights_matrix = np.loadtxt('testing_weights_matrix.txt', dtype=int)
num_embeddings = weights_matrix.shape[0]
embedding_dim = weights_matrix.shape[1]
weights_matrix = torch.from_numpy(weights_matrix).float()
emb_layer = nn.Embedding(num_embeddings, embedding_dim)
# torch.save(weights_matrix, "embed_weights.pth")
# weights_matrix = torch.load("embed_weights.pth") # <--- Only uncomment when required
emb_layer.weight.requires_grad = False # <--- Might need to be set to true as reqd
emb_layer.weight = nn.Parameter(weights_matrix)
return emb_layer, num_embeddings, embedding_dim
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu:0')
net = model.Net(create_emb_layer(), 7).to(device)
checkpoint = torch.load('./last_model_state.pth')
net.load_state_dict(checkpoint['model_state_dict'])
loaded_dataset = np.load("processed_test_dataset.npy")
loaded_labels = open(LABELS_PATH, "r").readlines()
net.eval()
with torch.no_grad():