def main():
os.chdir('/home/seonhoon/Desktop/workspace/ImageQA/data/')
n_vocab = 12047
y_vocab = 430
dim_word = 1024
dim = 1024
dim_ctx = 512
maxlen = 60
train = pd.read_pickle('train.pkl')
# train, valid = train_test_split(train, test_size=0.2)
valid = pd.read_pickle('test.pkl')
train_x = [q for q in train['q']]
train_y = [a[0] for a in train['a']]
train_y = np.array(train_y)[:, None]
train_y = np_utils.to_categorical(train_y, y_vocab).astype('int32')
train_x, train_x_mask = prepare_data(train_x, maxlen)
img_folder = '/home/seonhoon/Desktop/workspace/ImageQA/data/images/'
train_imgs = train['img_id'].apply(
lambda x: img_folder + 'train/' + x + '.jpg')
train_imgs = train_imgs.tolist()
valid_x = [q for q in valid['q']]
valid_y = [a[0] for a in valid['a']]
valid_x, valid_x_mask = prepare_data(valid_x, maxlen)
valid_imgs = valid['img_id'].apply(
lambda x: img_folder + 'test/' + x + '.jpg')
valid_imgs = valid_imgs.tolist()
print 'train x :', train_x.shape
print 'train x_mask:', train_x_mask.shape
print 'train y : ', train_y.shape
print 'train imgs :', len(train_imgs)
print 'valid x :', valid_x.shape
print 'valid x_mask:', valid_x_mask.shape
#print 'valid y : ', valid_y.shape
print 'valid imgs :', len(valid_imgs)
model = BIRNN_GRU(n_vocab, y_vocab, dim_word, dim, dim_ctx)
model.train(train_x,
train_x_mask,
train_imgs,
train_y,
valid_x,
valid_x_mask,
valid_imgs,
valid_y,
valid=2,
lr=0.008,
dropout=0.4,
batch_size=512,
epoch=60,
save=4)
def main():
y_class = 10
dim_freq = 5512
dim = 1024
step=112
dim_ctx = 4096
os.chdir('/home/seonhoon/Desktop/workspace/music/data6/')
train=pd.read_pickle('train_fft.pkl')
valid=pd.read_pickle('valid_fft.pkl')
train_x=np.array([ q for q in train['x'] ]).reshape(-1,step,dim_freq).astype('float32')
train_y=[ a for a in train['y'] ]
train_y=np.array(train_y)[:,None]
train_y = np_utils.to_categorical(train_y, y_class).astype('int32')
train_img = np.array([ f for f in train['cnn_feature'] ]).astype('float32')
valid_x=np.array([ q for q in valid['x'] ]).reshape(-1,step,dim_freq).astype('float32')
valid_y=[ a for a in valid['y'] ]
valid_y_true = valid_y
valid_y=np.array(valid_y)[:,None]
valid_y_original=valid_y
valid_y = np_utils.to_categorical(valid_y, y_class).astype('int32')
valid_img = np.array([ f for f in valid['cnn_feature'] ]).astype('float32')
print 'train x :', train_x.shape
print 'train y : ', train_y.shape
print 'train imgs :', train_img.shape
print 'valid x :', valid_x.shape
print 'valid y : ', valid_y.shape
print 'valid imgs :', valid_img.shape
model = BIRNN_GRU(y_class, dim_freq, dim, dim_ctx)
model.train(train_x, train_img, train_y,
valid_x, valid_img, valid_y_original, valid=1,
lr=0.0001, dropout=0.8, batch_size=512, epoch=100, save=10)
prediction, probs = model.prediction(valid_x, valid_img, valid_y, lr=0.0001, batch_size=512)
print valid_y_true
print prediction
correct = 0
for i in xrange(len(valid_y_true)):
if valid_y_true[i]==prediction[i]:
correct += 1
print correct
true = pd.DataFrame(valid_y_true)
pred = pd.DataFrame(prediction)
true.to_pickle('/home/seonhoon/Desktop/workspace/music/data6/true.pkl')
pred.to_pickle('/home/seonhoon/Desktop/workspace/music/data6/pred.pkl')
draw_cm(valid_y_true, prediction)
def main():
os.chdir('/home/seonhoon/Desktop/workspace/ImageQA/data/')
n_vocab = 12047
y_vocab = 430
dim_word = 1024
dim = 1024
dim_ctx = 512
maxlen = 60
train=pd.read_pickle('train.pkl')
# train, valid = train_test_split(train, test_size=0.2)
valid=pd.read_pickle('test.pkl')
train_x=[ q for q in train['q'] ]
train_y=[ a[0] for a in train['a'] ]
train_y=np.array(train_y)[:,None]
train_y = np_utils.to_categorical(train_y, y_vocab).astype('int32')
train_x , train_x_mask = prepare_data(train_x, maxlen)
img_folder='/home/seonhoon/Desktop/workspace/ImageQA/data/images/'
train_imgs = train['img_id'].apply(lambda x : img_folder+'train/'+x+'.jpg')
train_imgs = train_imgs.tolist()
valid_x=[ q for q in valid['q'] ]
valid_y=[ a[0] for a in valid['a'] ]
valid_x , valid_x_mask = prepare_data(valid_x, maxlen)
valid_imgs = valid['img_id'].apply(lambda x : img_folder+'test/'+x+'.jpg')
valid_imgs = valid_imgs.tolist()
print 'train x :', train_x.shape
print 'train x_mask:', train_x_mask.shape
print 'train y : ', train_y.shape
print 'train imgs :', len(train_imgs)
print 'valid x :', valid_x.shape
print 'valid x_mask:', valid_x_mask.shape
#print 'valid y : ', valid_y.shape
print 'valid imgs :', len(valid_imgs)
model = BIRNN_GRU(n_vocab, y_vocab, dim_word, dim, dim_ctx)
model.train(train_x, train_x_mask, train_imgs, train_y,
valid_x, valid_x_mask, valid_imgs, valid_y, valid=2,
lr=0.008, dropout=0.4, batch_size=512, epoch=60, save=4)