shuffle = True
show_accuracy = True
MODEL_ROOT = '../models/elephas/'
PREDICTION_ROOT = '../predictions/'
MODEL = 'sixlayeradagrad_noLSTM_batch_{}'.format(batch_size)
print 'Loading data...'
(X_train,Y_train),(X_test,Y_test) = load_data()
TEST_ID = '../Data/pkl/img_q_id_test'
TEST_ID_PKL = pickle.load(open(TEST_ID+'.pkl','rb'))
ids = map(nameToId,[ TEST_ID_PKL[idx][1] for idx in range(len(TEST_ID_PKL)) ])
print 'Building model...'
model = Akar.keras_model(1)
#print 'Defining callbacks...'
#checkpoint = ModelCheckpoint('../models/elephas/checkpoint_'+MODEL+'.h5', monitor='val_loss', verbose=0, save_best_only=True, mode='auto')
#earlystopping = EarlyStopping(monitor='val_loss', patience=2, verbose=0)
print 'Start training...'
for epoch in [20,40,60,80,100,120,140,160,180,200]:
model.fit( X_train, Y_train, batch_size=batch_size, nb_epoch=nb_epoch, verbose=verbose, callbacks=[],validation_split=validation_split, shuffle=shuffle,show_accuracy=show_accuracy)
model.save_weights(MODEL_ROOT+MODEL+"_{}.h5".format(epoch))
probs = model.predict(X_test,batch_size=128)
answers = map(numToC,np.argmax(probs,axis=1).tolist())
prediction = zip(ids,answers)
for num in xrange(len(IM_ID)):
ID = IM_ID[num].split('_')[2].split('.')[0]
IM_ID_DICT[ID] = num
mem_shape = (40504, 1, 1000)
mem_image = np.memmap('../Data/val2014/vgg_feats.memmap',
dtype='float32',
mode='r',
shape=mem_shape)
#===== prepare pickList =====
pickList = range(0, len(ID_PKL))
numToC = {0: 'A', 1: 'B', 2: 'C', 3: 'D', 4: 'E'}
answers = []
print "start making model..."
#model = Keras_model_noLSTM.keras_model(1)
model = Keras_model_deep_noLSTM.keras_model(1)
model.load_weights(MODEL_NAME)
#===== Start testing =====
print "Start testing!"
for epoch in xrange(1): #NULL loop
tStart = time.time()
progress = ProgressBar().start()
for batch_num in xrange(0, int(len(pickList) / BATCH_SIZE)):
progress.update(int((batch_num + 1) / float(72801 / BATCH_SIZE) * 100))
start = batch_num * BATCH_SIZE
end = (batch_num + 1) * BATCH_SIZE
if end > len(pickList):
start = len(pickList) - BATCH_SIZE
end = len(pickList)
batched_image = []
batched_words = []
import sys
import Keras_model_deep_noLSTM as Akar
import vqa
(X_train,Y_train),(X_test,Y_test) = vqa.load_data()
PREDICTION_FILE_NAME = '../predictions/test_elephat_test'
MODEL_NAME = '../models/elephas/overfit_noLSTM_100.h5'
TEST_ID = '../Data/pkl/img_q_id_test'
TEST_ID_PKL = pickle.load(open(TEST_ID+'.pkl','rb'))
print "start making model..."
model = Akar.keras_model(1)
model.load_weights(MODEL_NAME)
print "Start testing..."
prediction = model._predict([X_test])
def nameToId(ans_string):
return '{0:{fill}{align}7}'.format(ans_string,fill='0',align='>')
def numToC(ans_int):
if ans_int == 0:
return 'A'
elif ans_int == 1:
return 'B'
elif ans_int == 2:
return 'C'
#===== prepare pickList =====
pickList = range(0,len(ID_PKL))
random.shuffle(pickList)
val_pickList = []
for idx in xrange(0,int(len(pickList)*VAL_RATIO)):
val_pickList.append( pickList.pop(idx) )
# maybe this will help? -- Ray.
# this really help, thanks! -- Angus.
#print '{0:{fill}{align}12}'.format(ID_PKL[0][0],fill='0',align='>')
# printing 300-dim word vector
#print word_vec[ QUESTION_PKL[0][0] ]
print "start making model..."
#model = Keras_model_attention.keras_model(1)
#model = Keras_model_noLSTM.keras_model(1)
model = Keras_model_deep_noLSTM.keras_model(1)
#===== Start training =====
csv_list = []
print "Start training!"
for epoch in xrange(MAX_EPOCH):
to_save = []
tStart = time.time()
print "Epoch:",epoch+1
progress = ProgressBar().start()
cost = 0
for batch_num in xrange(0,int(len(pickList)/BATCH_SIZE)):
progress.update(int((batch_num+1)/float(146962*(1-VAL_RATIO)/BATCH_SIZE)*100))
start = batch_num*BATCH_SIZE
end = (batch_num+1)*BATCH_SIZE
if end > len(pickList):