Why Tensoflow?

It is important to answer this simple question. Different deep learning libraries are available these days such as Theano, Caffe, Torch and etc. However the problem is the trade-off between flexibility and simplicity. Some of the libraries like Caffe has the simplicity characteristics but not flexibility. For instance defining new layers in Caffe can become unbelievably complicated. Some like Torch are fast and flexible but since its language is Lua and it is not widely used, there might be a lack of interest in going through its associated API programming. The TensorFlow, for the moment, looks to be the best to satisfy the aformentioned trade-off and also designed to be very fast because its not only developed for research purposes.

The advantage(and also this might be a disadvantage!!) of TensorFlow is that it is very popular and under fast development! This can make any tutorial to become obsolete very fast! However we try to update ours as fast as possible to not to be back in the game! So what is the advantage here? Since it is growing very fast, it is constantly under modification for its flaws and bugs, so if you see a bug today it might disappear tomorrow if the last changes be pulled in your installed version!

Another advantage of TensorFlow is its fast graph computations for new architectures which makes it as the ultimate weapon of the TensorFlow. The TensorFlow interface is very organized compared to Theano and Caffe.

Using the aforementioned reasoning, we are aimed to dig into TensorFlow for being a small and yet effective part of TensorFlow community.