Miscellaneous info on tensorflow

Back to home page

1 tensorflow dependencies

To do large complex numerical computation like muatrix multiplication, tf depends on libraries like numpy, which execute the same in other programing language (ref). But there overheads, and so to make it more efficient, tf migrates an entire operation to a different language to increase speed and reduce overheads.

2 index of the highest entry in a tensor

tf.argmax is used to find the index of the highest entry in a tensor along some axis (note here that, tensor can have more than 2 dimensions, unlike matrix).

example code: tf.argmax(y,1)

3 Hyperparatmeter defination

Hyperparatmeter are those parameters that you change durring run to get a good model. eg include learning rate, number of iterations etc

Estimators defination

Estimator as the tf website says is a high level representation of a complete model.The estimator API provides method to

• train the model
• evaluate the model
• generate predictions