Parsing with a trained parser: load the pre-trained model

[chenfeng15a]

Hi, Cartus:
When I run:

python3 train.py --load_model 1 --model result/pretrain14.model --dev_file data/test.transitions --gold_AMR_dev ../jamr/scripts/test/amr.txt

Bug occurs:

[dynet] random seed: 615808844
[dynet] allocating memory: 512MB
[dynet] memory allocation done.
0...
done reading training actions file
#sents: 2
#tokens: 46
#types: 36
#POStags: 13
#actions: 42
#preds: 35
0...
done reading training actions file
#sents: 2
#tokens: 29
#types: 36
#POStags: 19
#actions: 43
#action types: 8
#preds: 36
Start loading preds
#lemmas with associated PR acts = 25
creating dictionary of pretrain embeddings
finish creating
UNK index in tok_dict_pretrain: 0
UNK index in tok_dict_all: 0
0...
done loading raw sent
#sents: 2
0...
done loading raw sent
#sents: 2
Rand word embedding size: 36
Pretrained word embedding size: 55
loading pretrained word embeddings
finish loading
loading pretrained model
Traceback (most recent call last):
File "train.py", line 227, in
parser.load_model(args.model)
File "/home/chenfeng/parser/AMR-cartus/model/stack_lstm.py", line 109, in load_model
self.pc.populate(filename)
File "_dynet.pyx", line 1461, in _dynet.ParameterCollection.populate
File "_dynet.pyx", line 1516, in _dynet.ParameterCollection.populate_from_textfile
RuntimeError: Dimensions of parameter /_24 looked up from file ({8557,200}) do not match parameters to be populated ({44,200})

Asking for your help. Thanks a lot!

[Cartus]

I see. That should be a bug. I am afraid that I am not able to fix it now since I am working for the deadline. Hopefully, I can fix these bugs and release a new pre-trained model in late December.

Sorry for the inconvenience caused.

[chenfeng15a]

Ok, thanks a lot! Hope you finish your work perfectly as soon as possible~ 发自我的iPhone

…

------------------ Original ------------------ From: Cartus <notifications@github.com&gt; Date: Fri,Nov 29,2019 4:23 AM To: Cartus/AMR-Parser <AMR-Parser@noreply.github.com&gt; Cc: chenfeng15a <2441688980@qq.com&gt;, Author <author@noreply.github.com&gt; Subject: Re: [Cartus/AMR-Parser] Parsing with a trained parser: load the pre-trained model (#5) I see. That should be a bug. I am afraid that I am not able to fix it now since I am working for the deadline. Hopefully, I can fix these bugs and release a new pre-trained model in late December. Sorry for the inconvenience caused. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.