TensorFlow Simplified Interface

I think it's great that we're seeing a lot of experimentation in how to express models for machine learning. It's very clear we haven't found the best ways yet, and seeing what people choose to try to make easy to express (and harder) is good for progress.

(I'm still grouchy about the state of modularity within machine learning models. It's not easy enough to reuse / have libraries at the model level yet.)

I haven't used tflearn or prettytensor, but I have used skflow (and a bit of raw TensorFlow).

SKFlow is nice if you are already using scikit learn because you can drop it straight into your sklean Pipelines[1]. This is great in terms of making it usable alongside other systems.

For example, I currently have a project using an ensemble of regression methods (2 different RandomForest regressors, and 3 XGB methods, then multiple different seeds for each method). SKFlow lets me drop in a TensorFlow regressor as well.

(In actual fact I can't get TF to perform as well as a RF on my featureset, and XGB outperforms it by far. This is using a relatively simple NN though).

[1] http://scikit-learn.org/stable/modules/generated/sklearn.pip...

It requires a bit more thinking ahead-of-time thinking, but also gives far more flexibility.

That said, I have found Keras to be excellent for quick experiments. Even more, because it supports both Tensorflow and Theano as backends.