KeyError: 'closure'

with
--- a/tutorial_rnnlm.py
+++ b/tutorial_rnnlm.py
@@ -6,8 +6,8 @@ import dynet as dy
import numpy as np

# format of files: each line is "word1/tag2 word2/tag2 ..."
-train_file="CHAR_TRAIN"
-test_file="CHAR_DEV"
+train_file=r"data/small-train.txt"
+test_file=r"data/small-test.txt"

I run:

rzai@rzai00:~/prj/dynet_tutorial_examples$ python tutorial_rnnlm.py
[dynet] random seed: 3680189036
[dynet] allocating memory: 512MB
[dynet] memory allocation done.
[epoch=0 eta=0.001 clips=499 updates=499] 3.59679303528
[epoch=0 eta=0.001 clips=500 updates=500] 2.90541985253
[epoch=0 eta=0.001 clips=500 updates=500] 2.82589651857
[epoch=0 eta=0.001 clips=500 updates=500] 2.78318269468
[epoch=0 eta=0.001 clips=500 updates=500] 2.7271153855
[epoch=0 eta=0.001 clips=500 updates=500] 2.71135471952
[epoch=0 eta=0.001 clips=500 updates=500] 2.67784155937
[epoch=0 eta=0.001 clips=500 updates=500] 2.67465448101
[epoch=0 eta=0.001 clips=500 updates=500] 2.6514647308
[epoch=0 eta=0.001 clips=500 updates=500] 2.63369510998
[epoch=0 eta=0.001 clips=500 updates=500] 2.61760151217
[epoch=0 eta=0.001 clips=500 updates=500] 2.60906285933
[epoch=0 eta=0.001 clips=500 updates=500] 2.58898071744
[epoch=0 eta=0.001 clips=500 updates=500] 2.5481264534
[epoch=0 eta=0.001 clips=500 updates=500] 2.55303348667
[epoch=0 eta=0.001 clips=500 updates=500] 2.54707865508
[epoch=0 eta=0.001 clips=500 updates=500] 2.53384964483
[epoch=0 eta=0.001 clips=500 updates=500] 2.5381710203
[epoch=0 eta=0.001 clips=500 updates=500] 2.63992879771
Traceback (most recent call last):
File "tutorial_rnnlm.py", line 107, in
loss_exp = calc_lm_loss(sent)
File "tutorial_rnnlm.py", line 78, in calc_lm_loss
wids = [vw.w2i[w] for w in sent]
KeyError: 'closure'
rzai@rzai00:~/prj/dynet_tutorial_examples$

Start with tutorial_rnnlm.py around calc_lm_loss at line 78, then inspect how its vocabulary and input sentences are built from data/small-train.txt and data/small-test.txt. Run python tutorial_rnnlm.py with those files and trace why 'closure' is absent from vw.w2i. Done means the tutorial processes the supplied data without this KeyError.

Escrito por el modelo de indexación a partir del texto del issue.