Drew

May just need some fiddling with the rules. Search `` in `resources/languages/en/rules/time.clj`. For what it's worth, "the 23rd of next month" works out-of-the-box: ``` duckling.core=> (load! {:config {:en$time {:corpus ["numbers"...

Take a look at https://clojure.org/reference/java_interop#_calling_clojure_from_java http://clojure.github.io/clojure/javadoc/ It's straightforward to write a wrapper class ensure Duckling is loaded (perhaps synchronously during construction) and to invoke the Duckling `parse` method.

I've identified _why_ the end interval is being improperly shifted by an hour. In `duckling.time.obj/interval-start-end`, after having determined the smallest grain of the interval endpoints, the `end` function is used...

+1. As a framework for training and evaluation, the lack of an out-of-the-box saving mechanism is curious to me. I suspect the lack of a disk-based model checkpointer may be...

And to make myself slightly useful and not purely complain: @Alikerin , the DeepMind research GitHub has an example of writing and restoring a model checkpoint from disk. See: https://github.com/deepmind/deepmind-research/blob/master/adversarial_robustness/jax/experiment.py...

What is the rationale for not allowing users to rename a collection? I cannot think of a DBMS that would not allow users to rename tables / databases.

Naively, I would think (part of) the "simple" solution would be to decouple internal storage mechanisms from external, user-provided information, for instance by using immutable UUIDs for each collection. That...