def train():
history_score = []; max_score = -1
for i in range(num_round):
fetches = [model.optimizer, model.loss]
if batch_size > 0:
ls = []
bar = progressbar.ProgressBar()
print('[%d]\ttraining...' % i)
for j in bar(range(int(train_size / batch_size + 1))):
feat_ids, feat_vals, label = slice_libsvm(train_data, j * batch_size, batch_size)
#a = model.run_step(fetches, feat_ids, feat_vals, label)
#for i in range(len(feat_vals)): feat_vals[i][18] /= 1000;feat_vals[i][19] /= 10;feat_vals[i][20] /= 10 # ********************
_, l = model.run_step(fetches, feat_ids, feat_vals, label)
ls.append(l)
elif batch_size == -1:
feat_ids, feat_vals, label = slice_libsvm(train_data)
_, l = model.run_step(fetches, feat_ids, feat_vals, label)
ls = [l]
train_preds = []
print('[%d]\tevaluating...' % i)
bar = progressbar.ProgressBar()
for j in bar(range(int(train_size / 10000 + 1))):
feat_ids, feat_vals, label = slice_libsvm(train_data, j * 10000, 10000)
preds = model.run_step(model.pred_prob, feat_ids, feat_vals, label)
train_preds.extend(preds)
test_preds = []
bar = progressbar.ProgressBar()
for j in bar(range(int(test_size / 10000 + 1))):
feat_ids, feat_vals, label = slice_libsvm(test_data, j * 10000, 10000)
preds = model.run_step(model.pred_prob, feat_ids, feat_vals, label)
#auc = model.run_step(model.auc, feat_ids, feat_vals, label)
test_preds.extend(preds)
train_true = []; test_true = []
for e in train_data:
train_true.append(e[2])
for e in test_data:
test_true.append(e[2])
train_score = roc_auc_score(train_true, train_preds)
test_score = roc_auc_score(test_true, test_preds)
trprecision, trrecall, tracc = calScore(train_true, train_preds)
teprecision, terecall, teacc = calScore(test_true, test_preds)
print('[%d]\tloss: %f\ttrain-auc: %f\teval-auc: %f\t\tprecision: %f\trecall: %f\ttrain-acc: %f\ttest-acc: %f'
% (i, np.mean(ls), train_score, test_score, teprecision, terecall, tracc, teacc))
history_score.append(test_score)
if test_score > max_score:
model.save_model(FLAGS.model_dir)
max_score = test_score
if i > min_round and i > early_stop_round:
if np.argmax(history_score) == i - early_stop_round and history_score[-1] - history_score[-1 * early_stop_round] < 1e-5:
print('early stop\nbest iteration:\n[%d]\teval-auc: %f' % (np.argmax(history_score), np.max(history_score)))
model.save_model(FLAGS.model_dir)
break
def export_features():
train_data = read_libsvm(FLAGS.data_dir)
feat_ids, feat_vals, label = slice_libsvm(train_data)
fea = model.run_step(model.fea_out, feat_ids, feat_vals, label)
with open(FLAGS.feature_dir, 'w') as f:
for i in range(len(fea)):
f.write(str(label[i]) + ' ') # label
for j in range(len(fea[i])): # features
f.write(str(j + 1) + ':' + str(round(fea[i][j], 3)) + ' ')
f.write('\n')
print('export feature done, file : %s' % FLAGS.feature_dir)
exit()