Training NER fails

[ghost]

How to reproduce the behaviour

I am trying to train spacy on CoNLL 2003 NER data. I used the CLI convertor to convert from conll to spacy format.

python -m spacy convert eng.train . -c ner for all 3 files: train, dev and test and renamed the files to .json

To train, I am using this command:
python -m spacy train en scner/ engtrain.json engtesta.json -p ner -VV -D

Itn  NER Loss   NER P   NER R   NER F   Token %  CPU WPS
---  ---------  ------  ------  ------  -------  -------
✔ Saved model to output directory
scner3/model-final

I am getting this error:

Traceback (most recent call last):
  File "/home/su/.env/lib/python3.5/site-packages/spacy/cli/train.py", line 397, in train
    scorer = nlp_loaded.evaluate(dev_docs, verbose=verbose)
  File "/home/su/.env/lib/python3.5/site-packages/spacy/language.py", line 672, in evaluate
    docs, golds = zip(*docs_golds)
ValueError: not enough values to unpack (expected 2, got 0)

Running the debug-data command
python3 -m spacy debug-data en engtrain.json engtesta.json

gives me this:

=========================== Data format validation ===========================
✔ Corpus is loadable

=============================== Training stats ===============================
Training pipeline: tagger, parser, ner
Starting with blank model 'en'
946 training docs
0 evaluation docs
✔ No overlap between training and evaluation data
⚠ Low number of examples to train from a blank model (946)

============================== Vocab & Vectors ==============================
ℹ 203621 total words in the data (23623 unique)
ℹ No word vectors present in the model

========================== Named Entity Recognition ==========================
ℹ 4 new labels, 0 existing labels
0 missing values (tokens with '-' label)
✔ Good amount of examples for all labels
✔ Examples without occurrences available for all labels
✔ No entities consisting of or starting/ending with whitespace

=========================== Part-of-speech Tagging ===========================
ℹ 45 labels in data (57 labels in tag map)
✘ Label ')' not found in tag map for language 'en'
✘ Label '(' not found in tag map for language 'en'
✘ Label 'NN|SYM' not found in tag map for language 'en'
✘ Label '"' not found in tag map for language 'en'

============================= Dependency Parsing =============================
ℹ Found 203621 sentences with an average length of 1.0 words.
ℹ 1 label in train data
ℹ 1 label in projectivized train data

================================== Summary ==================================
✔ 5 checks passed
⚠ 2 warnings
✘ 4 errors

Your Environment

• Models: en_core_web_lg
• Python version: 3.5.2
• Platform: Linux-4.15.0-58-generic-x86_64-with-Ubuntu-16.04-xenial
• spaCy version: 2.2.2

I am trying to train the model only on NER, not on POS.

[adrianeboyd]

If the conversion with spacy convert works correctly you shouldn't have to rename the files to .json, so maybe something isn't configured correctly there. It should be the default, but you can also try explicitly adding -t json.

The NER summary from debug-data looks okay, but this is your immediate problem:

0 evaluation docs

There should be a clear error in debug-data for this case, since it will cause the problem with the evaluation above.

[lock]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.