args.server_ip = '' args.server_port = '' args.output_mode = "classification" args.save_model_steps = 2000 args.resume_epochs = 0 args.resume_steps = 0 # Important configurations args.data_dir = 'dataset/preprocessed' args.train_file = None args.dev_file = 'friends_majority_result.json' args.result_file = 'friends_all_result.json' args.train_batch_size = 32 args.eval_batch_size = 32 args.do_train = False args.do_eval = False args.do_run = True args.num_train_epochs = 1.0 args.max_seq_length = 256 args.processor = Others_OneSentence_Processor args.output_dir = None args.resume_dir = os.path.join(model_dir, 'friends_others/epoch_6') args.learning_rate = 1e-5 args.seed = 69847 args.included_labels = 7 trainer = Trainer(args) trainer.execute()
print "Training started"
for n in range(1, trainSetSize+1):
filename = "train%d.tif" % n
print filename
trainer.train(trainAnno, container, filename)
trainRef = trainer.record(container)
###############################################
print "Dev set recognition started"
devRef = trainer.openAnnoFile("annotations/dev.txt")
rightNum = 0.0
percentage = 0.0
for n in range(1, devSetSize+1):
filename = "dev%d.tif" % n
print filename
predClass, probability = trainer.execute(trainRef, filename)
actualClass = int(devRef[filename])
if actualClass == int(predClass):
rightNum += 1
percentage = (rightNum / (devSetSize+1)) * 100
print "*** Recognition Result : ", str(percentage) + "% ***"
