Home How to reduce the CNN-SVHN train input to know if the model is going well instead of using the full train dataset and wait hours to see the result?
Reply: 0

How to reduce the CNN-SVHN train input to know if the model is going well instead of using the full train dataset and wait hours to see the result?

user1271
1#
user1271 Published in May 27, 2018, 3:37 am

I am newbie in Tensorflow and I would like to train the SVHN model with just few inputs so the model will train-finish soon instead of waiting hours by using the full SVHN train dataset (train_32x32.mat). I would like to use few input images and scale up to see if the model is working properly. I was following the next script but it uses the full train/test dataset. All answers will be appreciate it, thanks.

class SVHN:

 def __init__(self, file_path, n_classes, use_extra=False, gray=False): 
     self.n_classes = n_classes  

     # # Load Train Set 
     train = sio.loadmat(file_path + "/train_32x32.mat") 
     self.train_labels = self.__one_hot_encode(train['y']) 
     self.train_examples = train['X'].shape[3] 
     self.train_data = self.__store_data(train['X'].astype("float32"), self.train_examples, gray) 


     # Load Test Set 
     test = sio.loadmat("../res/test_32x32.mat") 
     self.test_labels = self.__one_hot_encode(test['y']) 
     self.test_examples = test['X'].shape[3] 
     self.test_data = self.__store_data(test['X'].astype("float32"), self.test_examples, gray) 

To sum up, I would like to use few inputs (around 1000) for training/testing and then scale up.

Thanks

You need to login account before you can post.

About| Privacy statement| Terms of Service| Advertising| Contact us| Help| Sitemap|
Processed in 0.311258 second(s) , Gzip On .

© 2016 Powered by mzan.com design MATCHINFO