CODEWITHSUNDEEP
pythontensorflow
Han M
Han M

Reputation: 399

When to use tensorflow estimators?

I have a general tensorflow question about when to use estimators. I feel sometimes estimators are not convenient to build something, since we need to meet some fixed requirements when building the graph. On the other hand, using lower level api can be tedious sometimes. Therefore, I want to ask when it is proper to use estimators and when it is not. Thanks!

Upvotes: 2

Views: 78

Answers (1)

Jonathan R
Jonathan R

Reputation: 3928

This is a very opinionated answer but I will still write it:

The Estimator-API was developed to simplify building and sharing models. You could compare it with Keras and in fact Estimators is built with tf.keras.layers so one could say it is a simplification of a simplification.

This is obviously good for beginners or people who come from other fields (as people working with ML often do) but can also delimit the things you can do.

As a general rule of thumb I would use Estimators if you want to work on or share a model with people that do not have a good CS background but want to get going either way.

Upvotes: 1

Related Questions