Join GitHub today
GitHub is home to over 31 million developers working together to host and review code, manage projects, and build software together.Sign up
Why did you call BasicLSTMCell a cell and not a layer? #14693
People are usually first introduced to MLPs or feed-forward (and fully connected) neural networks, before being introduced to RNNs and, in particular, LSTMs.
Why would you call
Moreover, the first parameter to
It almost seems that you created TF to make it as confusing as possible to make it seem hard. It also almost seems that the person who wrote the name of the class
A similar argument can be said for
Change classes such as
We work hard on building clean and consistent interfaces, but the truth is that the field moves quickly, and the description and terms used when an idea first breaks out aren't always the same as the best way to teach things turns out a few years later. And we also have a responsibility to our users to maintain some backwards compatibility with the older (and probably uglier) interfaces. I think in practice we can add new, cleaner interfaces, but it's a slower process to deprecate and remove the older, uglier ones.
I think @ebrevdo wrote the code in question (both the class and its init method). He might be able to comment on future, cleaner interfaces.
Nomenclature varies between different researchers. It's unfortunate that the naming doesn't match standard notation taught by NN courses, but this interface is locked for backwards compatibility until at least tf 2.0 due to our API guarantees. You may have an easier time with the naming in the Keras API. Will add this request to the internal list of possible breaking changes for future versions. Closing.
@ebrevdo Yes, nomenclature varies, but I am more concerned about consistency within the same class, module and possibly library.
Number of units??? Which units??? A cell has units???
At least, meanwhile, you should change the documentation to make it more understandable!!
Would you like to submit a PR to help improve the documentation? I'm happy to review.…
On Sun, Nov 19, 2017 at 6:03 AM, Nelson Brochado ***@***.***> wrote: @ebrevdo <https://github.com/ebrevdo> Yes, nomenclature varies, but I am more concerned about consistency within the same class, module and possibly library. The "cell" BasicLSTMCell contains a method called __init__ whose first parameter is num_units, which is vaguely documented <https://www.tensorflow.org/api_docs/python/tf/contrib/rnn/BasicLSTMCell> as The number of units in the LSTM cell. Number of units??? Which units??? A cell has units??? At least, meanwhile, you should change the documentation to make it more understandable!! — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#14693 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABtimxO70QZgMLDIemq8Iluxxjsz1heVks5s4DUogaJpZM4QjErd> .