Closed pmichel31415 closed 6 years ago
Thanks for finding these!
re 1: as of #345, this is documented under xnmt/settings.py
re 3: thanks for catching these; fixed as of #345
re 4: also not sure, but I did add show-inheritance
flags so we can at least follow the link to the base class
re 5: I hadn't thought about this, but turns out you can: simply overwrite EXP_DIR in the experiment.exp_global.placeholders dictionary. Or, adjust it in xnmt.settings to change locations for all experiments you run.
re 6: --dynet-seed changes seeds for both dynet and XNMT.
Thanks for the quick response @msperber I'll still leave this open for the translator structure documentation part
I think the documentation has been improved significantly, so maybe we can close this?
Hi, here's a list of questions I have. None of them warrant a separate issue, but clarifying those would help new users such as myself
please delete by hand if you want to overwrite it (or use --settings debug or otherwise set OVERWRITE_LOG=True)
(what is theOVERWRITE_LOG
variable and where can I change it?)MlpSoftmaxDecoder
was. Something that would help is to describe a common baseline model (with equations/diagrams and everything) and then show a config file where all the parameters are specified that reproduces itlayers
argument to theMlpSoftmaxDecoder
doesn't work for mebeam
argument toSimpleInference
raises an errorValueError: 'beam' is not a accepted argument of SimpleInference.__init__(). Valid are ['src_file', 'trg_file', 'ref_file', 'max_src_len', 'post_process', 'report_path', 'report_type', 'search_strategy', 'mode', 'max_len', 'batcher']
EXP_DIR
in the config file to point to a directory of my choice?I will update if I run into more issues.
PS: I'm using the latest master