def acc_hook(scores, y_data, other_id):
true_and_prediction = []
corr = 0.
total = 0.
for score, true_label in zip(scores, y_data):
predicted = []
true_tag = []
for i in range(len(score)):
t1 = true_label[i]
if t1 != 0:
true_tag.append(i)
t2 = np.argmax(score[i])
if t2 != 0:
predicted.append(i)
if t1 != 0:
total += 1
t2 = np.argmax(score[i][1:]) + 1
if t1 == t2:
corr += 1.0
if predicted == []: predicted = [other_id]
if true_tag == []: true_tag = [other_id]
true_and_prediction.append((true_tag, predicted))
stct = strict(true_and_prediction)
macro = loose_macro(true_and_prediction)
micro = loose_micro(true_and_prediction)
report = []
if total != 0:
report.append("Accuracy:%f %d" % (corr / total, total))
report.append(" strict (p,r,f1):%f %f %f:" % stct)
report.append("loose macro (p,r,f1):%f %f %f" % macro)
report.append("loose micro (p,r,f1):%f %f %f" % micro)
report.append("%f\t%f\t%f" % (stct[-1], macro[-1], micro[-1]))
return stct[2], report, [t[1] for t in true_and_prediction]
def acc_hook(scores,
y_data,
gaussian=False,
delta=0,
topK=1,
use_path=False,
label_hierarchy=None):
true_and_prediction = get_true_and_prediction(scores, y_data, gaussian,
delta, topK, use_path,
label_hierarchy)
print(" strict (p,r,f1):", strict(true_and_prediction))
print("loose macro (p,r,f1):", loose_macro(true_and_prediction))
print("loose micro (p,r,f1):", loose_micro(true_and_prediction))
def acc_hook(scores, y_data):
true_and_prediction = []
for score, true_label in zip(scores, y_data):
predicted_tag = []
true_tag = []
for label_id, label_score in enumerate(list(true_label)):
if label_score > 0:
true_tag.append(label_id)
lid, ls = max(enumerate(list(score)), key=lambda x: x[1])
predicted_tag.append(lid)
for label_id, label_score in enumerate(list(score)):
if label_score > 0.5:
if label_id != lid:
predicted_tag.append(label_id)
true_and_prediction.append((true_tag, predicted_tag))
print(" strict (p,r,f1):", strict(true_and_prediction))
print("loose macro (p,r,f1):", loose_macro(true_and_prediction))
print("loose micro (p,r,f1):", loose_micro(true_and_prediction))
def acc_hook(scores,Y,th=3000):
true_and_prediction = []
num_of_samples = Y.shape[0]
for score,true_label in zip(scores,Y):
predicted_tag = []
true_tag = []
for label_id,label_score in enumerate(list(true_label)):
if label_score > 0:
true_tag.append(label_id)
lid,ls = max(enumerate(list(score)),key=lambda x: x[1])
predicted_tag.append(lid)
for label_id,label_score in enumerate(list(score)):
if label_score > th:
if label_id != lid:
predicted_tag.append(label_id)
true_and_prediction.append((true_tag, predicted_tag))
print(" strict (p,r,f1):",strict(true_and_prediction))
print("loose macro (p,r,f1):",loose_macro(true_and_prediction))
print("loose micro (p,r,f1):",loose_micro(true_and_prediction))
def acc_hook(batcher,output_model,x_context,x_target,feature,y,keep_prob_context,keep_prob_target,context_length,sess):
true_and_prediction = []
[x_context_data, x_target_mean_data, y_data,feature_data] = batcher.next()
num_of_samples = y_data.shape[0]
feed = {y:y_data,keep_prob_context:[1.],keep_prob_target:[1.],feature:feature_data}
for i in range(context_length*2+1):
feed[x_context[i]] = x_context_data[:,i,:]
feed[x_target] = x_target_mean_data
scores = sess.run(output_model, feed_dict = feed)
"""
for score,true_label in zip(scores,y_data):
predicted_tag = []
true_tag = []
for label_id,label_score in enumerate(list(true_label)):
if label_score > 0:
true_tag.append(label_id)
for label_id,label_score in enumerate(list(score)):
if label_score > 0.5:
predicted_tag.append(label_id)
true_and_prediction.append((true_tag, predicted_tag))
"""
for score,true_label in zip(scores,y_data):
predicted_tag = []
true_tag = []
for label_id,label_score in enumerate(list(true_label)):
if label_score > 0:
true_tag.append(label_id)
lid,ls = max(enumerate(list(score)),key=lambda x: x[1])
predicted_tag.append(lid)
for label_id,label_score in enumerate(list(score)):
if label_score > 0.5:
if label_id != lid:
predicted_tag.append(label_id)
true_and_prediction.append((true_tag, predicted_tag))
print(" strict (p,r,f1):",strict(true_and_prediction))
print("loose macro (p,r,f1):",loose_macro(true_and_prediction))
print("loose micro (p,r,f1):",loose_micro(true_and_prediction))
is_leaf=is_only_leaf(gt_labels[0])
if is_leaf != 0:
continue
predict_quant_embd,id_list = model.forward(context.to(device), lens.to(device),ids,left_right)
predict_quant_embd=predict_quant_embd.data.cpu().numpy()
mask_matrix=np.multiply(quantam_concept_embds, predict_quant_embd)
normalizing_constant=np.sum(np.abs(predict_quant_embd)**2,axis=-1)
score=(np.sum(np.abs(mask_matrix) ** 2, axis=-1))/normalizing_constant
predicted_labels_id,best_level_1_node,best_level_1_score,best_level_2_node,best_level_2_score=predict_labels(score,threshold_level_wise,constant_to_divide)
predicted_labels=[total_node_list[i] for i in predicted_labels_id]
temp_dict = {}
temp_dict['gt_label'] = gt_labels[0]
temp_dict['predicted_label'] = predicted_labels
temp_dict['best_level_1_node']=best_level_1_node
temp_dict['best_level_2_node'] = best_level_2_node
temp_dict['best_level_1_score']=best_level_1_score
temp_dict['best_level_2_score']=best_level_2_score
total_gt_label_len.append(len(gt_labels[0]))
total_predicted_label_len.append(len(predicted_labels))
p,r,f=evaluate.loose_macro([(gt_labels_id, predicted_labels_id)])
p1,r1,f1=evaluate.strict([(gt_labels_id, predicted_labels_id)])
true_and_prediction.append((gt_labels_id, predicted_labels_id))
print("strict (p,r,f1):",evaluate.strict(true_and_prediction))
print("loose micro (p,r,f1):",evaluate.loose_micro(true_and_prediction))
print("loose macro (p,r,f1):",evaluate.loose_macro(true_and_prediction))
def acc_hook(scores, y_data):
true_and_prediction = get_true_and_prediction(scores, y_data)
print(" strict (p,r,f1):",strict(true_and_prediction))
print("loose macro (p,r,f1):",loose_macro(true_and_prediction))
print("loose micro (p,r,f1):",loose_micro(true_and_prediction))