# IndexError in train recipe **URL:** \ **Category:** Uncategorized\ **Tags:** usage, ner, spacy, custom, training\ **Created:** [April 23, 2022, 2:42pm UTC](https://support.prodi.gy/t/indexerror-in-train-recipe/5565 "2022-04-23T14:42:55Z")\ **Posts on this page:** 7\ **Page:** 1
**Author:** ![richielo](https://sea2.discourse-cdn.com/flex020/user_avatar/support.prodi.gy/richielo/32/2932_2.png) [@richielo](https://support.prodi.gy/u/richielo)\ **Post date:** [April 23, 2022, 2:42pm UTC](https://support.prodi.gy/t/indexerror-in-train-recipe/5565/1 "2022-04-23T14:42:55Z")
I am trying to use the train recipe after importing some preannotated jsonl data in this format: ```python {"text": " ( d\u00e1le jen objednatel ) ", "spans": [{"token_end": 4, "start": 15, "label": "LABEL", "end": 26, "token_start": 4}]} ``` I load it as a dataset: ```python prodigy db-in cs_data ./annotated.jsonl ``` Then when I use the train recipe, I run into an Index Error: ```python span = doc.char_span(start, end, label) File "spacy\tokens\doc.pyx", line 573, in spacy.tokens.doc.Doc.char_span File "spacy\tokens\span.pyx", line 98, in spacy.tokens.span.Span. __cinit__ IndexError: [E035] Error creating span with start 17 and end 1 for Doc of length 18. ``` My train command is: ```python prodigy train ./prodigy_models/ --ner cs_data --lang cs --verbose ``` Am I missing some fields in the annotated data file? Or is there something wrong with how I set up Spacy? I did check the data file and can't find any cases where end is smaller than start. Thanks! ---
**Author:** ![koaning](https://sea2.discourse-cdn.com/flex020/user_avatar/support.prodi.gy/koaning/32/230_2.png) [@koaning](https://support.prodi.gy/u/koaning)\ **Post date:** [April 25, 2022, 9:00am UTC](https://support.prodi.gy/t/indexerror-in-train-recipe/5565/2 "2022-04-25T09:00:15Z")
I just ran `spancat.manual` on this single jsonl example. ```python {"text": " ( d\u00e1le jen objednatel ) "} ``` ![image](https://us1.discourse-cdn.com/flex020/uploads/prodigy/original/2X/5/5ba1a8a630181ada1f7993b20297fdacc7ee9351.png) After labelling the name, I saved the label and quit prodigy. Next, I ran `prodigy db-out` to have a look at the data format that Prodigy provides. ```python { "text":" ( d\u00e1le jen objednatel ) ", "_input_hash":-1815072948, "_task_hash":1525844819, "tokens":[ {"text":"<","start":0,"end":1,"id":0,"ws":false}, {"text":"i","start":1,"end":2,"id":1,"ws":false}, {"text":">","start":2,"end":3,"id":2,"ws":true}, {"text":"(","start":4,"end":5,"id":3,"ws":true}, {"text":"d\u00e1le","start":6,"end":10,"id":4,"ws":true}, {"text":"jen","start":11,"end":14,"id":5,"ws":true}, {"text":"objednatel","start":15,"end":25,"id":6,"ws":true}, {"text":")","start":26,"end":27,"id":7,"ws":true}, {"text":"<","start":28,"end":29,"id":8,"ws":false}, {"text":"/i","start":29,"end":31,"id":9,"ws":false}, {"text":">","start":31,"end":32,"id":10,"ws":false} ], "_view_id":"spans_manual", "spans":[ {"start":6,"end":25,"token_start":4,"token_end":6,"label":"NAME"} ], "answer":"accept", "_timestamp":1650876701 } ``` Just to check, did I label the same thing as you? Could you share the tokens in your `jsonl` file as well? In general, I recommend parsing out the HTML before labelling it because the spaCy tokeniser might have a hard time dealing with HTML. I am able to refer to the dataset with this example that seems to run just fine. ```python > prodigy train --spancat spandemo ℹ Using CPU ========================= Generating Prodigy config ========================= ℹ Auto-generating config with spaCy Using 'spacy.ngram_range_suggester.v1' for 'spancat' with sizes 3 to 3 (inferred from data) ✔ Generated training config =========================== Initializing pipeline =========================== [2022-04-25 10:59:00,030] [INFO] Set up nlp object from config Components: spancat Merging training and evaluation data for 1 components - [spancat] Training: 1 | Evaluation: 0 (20% split) Training: 1 | Evaluation: 0 Labels: spancat (1) [2022-04-25 10:59:00,039] [INFO] Pipeline: ['spancat'] [2022-04-25 10:59:00,042] [INFO] Created vocabulary [2022-04-25 10:59:00,042] [INFO] Finished initializing nlp object [2022-04-25 10:59:00,064] [INFO] Initialized pipeline components: ['spancat'] ✔ Initialized pipeline ============================= Training pipeline ============================= Components: spancat Merging training and evaluation data for 1 components - [spancat] Training: 1 | Evaluation: 0 (20% split) Training: 1 | Evaluation: 0 Labels: spancat (1) ℹ Pipeline: ['spancat'] ℹ Initial learn rate: 0.001 E # LOSS SPANCAT SPANS_SC_F SPANS_SC_P SPANS_SC_R SCORE --- ------ ------------ ---------- ---------- ---------- ------ 0 0 1.80 0.00 0.00 0.00 0.00 200 200 1.17 0.00 0.00 0.00 0.00 400 400 0.00 0.00 0.00 0.00 0.00 ``` ---
**Author:** ![richielo](https://sea2.discourse-cdn.com/flex020/user_avatar/support.prodi.gy/richielo/32/2932_2.png) [@richielo](https://support.prodi.gy/u/richielo)\ **Post date:** [April 30, 2022, 3:52pm UTC](https://support.prodi.gy/t/indexerror-in-train-recipe/5565/3 "2022-04-30T15:52:13Z")
Thank you for getting back to me. The label I am using is 'objednatel' in this example. I can't really tell which sample is causing an error. In our use case, the html tags provide context which is why I am leaving them in. Am I supposed to provide tokens in the dataset? Perhaps the missing info is causing issues? I am only providing it in the format posted above. In this example, it should be: ['_', '(, 'd\u00e1le', 'jen', 'objednatel ', ')', '_'] ```python {"text":" ( d\u00e1le jen objednatel ) ","spans":[{"token_end":4,"start":15,"label":"DEFINING_KEYWORD","end":26,"token_start":4}],"_input_hash":-1815072948,"_task_hash":489858686,"answer":"accept"} ``` This is my output for db-out. Am I supposed to specify how is this tokenized? Sorry, I am new to this. I also noticed you are using --spancat, is that necessary? ---
**Author:** ![richielo](https://sea2.discourse-cdn.com/flex020/user_avatar/support.prodi.gy/richielo/32/2932_2.png) [@richielo](https://support.prodi.gy/u/richielo)\ **Post date:** [May 1, 2022, 1:05pm UTC](https://support.prodi.gy/t/indexerror-in-train-recipe/5565/4 "2022-05-01T13:05:52Z")
Just to follow up @koaning, I tried your train command too and get the same error ---
**Author:** ![koaning](https://sea2.discourse-cdn.com/flex020/user_avatar/support.prodi.gy/koaning/32/230_2.png) [@koaning](https://support.prodi.gy/u/koaning)\ **Post date:** [May 2, 2022, 8:02am UTC](https://support.prodi.gy/t/indexerror-in-train-recipe/5565/5 "2022-05-02T08:02:45Z")
Ah! My bad! I mistook your NER task for a spancat task. Sorry about that. I've gone through all the steps again. I just took your original example: ```python {"text": " ( d\u00e1le jen objednatel ) ", "spans": [{"token_end": 4, "start": 15, "label": "LABEL", "end": 26, "token_start": 4}]} ``` And saved it into a file named `ner-task.jsonl`. Next, I added this one line into a dataset. ```bash prodigy db-in cs_data ./ner-task.jsonl ``` And I continued by running the train command. ```python > prodigy train ./prodigy_models/ --ner cs_data --lang cs --verbose ℹ Using CPU ========================= Generating Prodigy config ========================= ℹ Auto-generating config with spaCy ✔ Generated training config =========================== Initializing pipeline =========================== [2022-05-02 09:49:14,984] [INFO] Set up nlp object from config Components: ner Merging training and evaluation data for 1 components - [ner] Training: 1 | Evaluation: 0 (20% split) Training: 1 | Evaluation: 0 Labels: ner (1) - [ner] LABEL [2022-05-02 09:49:14,994] [INFO] Pipeline: ['tok2vec', 'ner'] [2022-05-02 09:49:14,996] [INFO] Created vocabulary [2022-05-02 09:49:14,997] [INFO] Finished initializing nlp object Skipped 1 examples for 'ner' Skipped 1 examples for 'ner' Skipped 1 examples for 'ner' [2022-05-02 09:49:15,018] [DEBUG] [W033] Training a new parser or NER using a model with no lexeme normalization table. This may degrade the performance of the model to some degree. If this is intentional or the language you're using doesn't have a normalization table, please ignore this warning. If this is surprising, make sure you have the spacy-lookups-data package installed and load the table in your config. The languages with lexeme normalization tables are currently: cs, da, de, el, en, id, lb, mk, pt, ru, sr, ta, th Load the table in your config with: [initialize.lookups] @misc = "spacy.LookupsDataLoader.v1" lang = ${nlp.lang} tables = ["lexeme_norm"] Skipped 1 examples for 'ner' Skipped 1 examples for 'ner' [2022-05-02 09:49:15,048] [INFO] Initialized pipeline components: ['tok2vec', 'ner'] ✔ Initialized pipeline ============================= Training pipeline ============================= Components: ner Merging training and evaluation data for 1 components - [ner] Training: 1 | Evaluation: 0 (20% split) Training: 1 | Evaluation: 0 Labels: ner (1) - [ner] LABEL ℹ Pipeline: ['tok2vec', 'ner'] ℹ Initial learn rate: 0.001 E # LOSS TOK2VEC LOSS NER ENTS_F ENTS_P ENTS_R SCORE --- ------ ------------ -------- ------ ------ ------ ------ Skipped 1 examples for 'ner' 0 0 0.00 0.00 0.00 0.00 0.00 0.00 200 200 0.00 0.00 0.00 0.00 0.00 0.00 400 400 0.00 0.00 0.00 0.00 0.00 0.00 600 600 0.00 0.00 0.00 0.00 0.00 0.00 800 800 0.00 0.00 0.00 0.00 0.00 0.00 1000 1000 0.00 0.00 0.00 0.00 0.00 0.00 1200 1200 0.00 0.00 0.00 0.00 0.00 0.00 1400 1400 0.00 0.00 0.00 0.00 0.00 0.00 1600 1600 0.00 0.00 0.00 0.00 0.00 0.00 ✔ Saved pipeline to output directory ``` This ran without errors on my machine. That means it can't hurt to ask; what version of Prodigy/Python/spaCy are you running? What operating system? ---
**Author:** ![richielo](https://sea2.discourse-cdn.com/flex020/user_avatar/support.prodi.gy/richielo/32/2932_2.png) [@richielo](https://support.prodi.gy/u/richielo)\ **Post date:** [May 7, 2022, 8:44pm UTC](https://support.prodi.gy/t/indexerror-in-train-recipe/5565/6 "2022-05-07T20:44:27Z")
Thank you for trying it again. I double-checked the data and still don't see any cases where end \< start. It seems to have something to do with the tokenizer but am not sure what should be the next logical debugging step. I am running this on Windows 10 and Python 3.8.5. The relevant libraries' versions are: spacy==3.2.4 spacy-legacy==3.0.9 spacy-loggers==1.0.2 prodigy==1.11.7 ---
**Author:** ![koaning](https://sea2.discourse-cdn.com/flex020/user_avatar/support.prodi.gy/koaning/32/230_2.png) [@koaning](https://support.prodi.gy/u/koaning)\ **Post date:** [May 9, 2022, 5:34am UTC](https://support.prodi.gy/t/indexerror-in-train-recipe/5565/7 "2022-05-09T05:34:08Z")
Your versions seem up to date, so it's probably not that. Can you confirm that everything works for just the single example that I'm using? That single example I was using doesn't have a token that starts at 17, so I'm assuming the error appears at another example in your dataset. It feels like there's a single example somewhere, I suppose the easiest way to find the culprit might be by running something like; ```python from prodigy.components.db import connect db = connect() examples = db.get_dataset("DATASET_NAME") for i, example in enumerate(examples): if any((t['start'] == 17) & (t['end'] == 17) for t in example['tokens']): print("found strange example at row {i}") ```