Tim Vieira
Tim Vieira
What script did you use to generate the files matching `data/*.clean`? In particular, did you do any preprocessing to the data?
I'm having trouble loading the trained model that ships with the repo. ```bash $ python src/main.py --test data/23.auto.clean --vocab data/vocab.json --model data/model ... Loaded test trees from data/23.auto.clean Traceback (most...
I see that you are using an "unsafe" softmax distribution on [line 308 of parser.py](https://github.com/jhcross/span-parser/blob/master/src/parser.py#L308). This line is likely to overflow if `scores * alpha` is large. There is [a...
Anyone have a good pairing for ``` growth decay ``` as in "exponential growth" or "exponential decay"?
We should automatically update documentation and tutorials using a script. (Some documentation can be used as a doctest.)
@jeisner writes on Issue #1: > @timvieira: We should probably scrap dumping then entire charts all together. Eventually (when we have backward chaining) many items won't be eagerly derived. Yes,...
At the moment we use FIFO. For acyclic programs there is probably little advantage to straying away from topological order. Computing topological order in a Dyna program isn't totally obvious...
Backpointers
How should we support backpointers for extracting derivations?
User should (as it can't yet, but should) be able to interrupt execution of the solver, inspect intermediate results with debugger, retract troublesome rules, etc.. and resume execution with out...
Figure out how to get dropout to work. There's a difference between train and test in how dropout works. (We already need a train/test time flag for reinforce to go...