Hi! Thanks for building this great tool, had some issues with the setup, but here's what I have in terms of a rqs.txt (incase this is helpful).

blis==0.7.4 catalogue==1.0.0 certifi==2021.5.30 chardet==4.0.0 click==7.1.2 cymem==2.0.5 datasets==1.6.0 dill==0.3.3 elasticsearch==7.10.0 filelock==3.0.12 Flask==1.1.0 fsspec==2021.6.0 gensim==3.8.0 huggingface-hub==0.0.10 idna==2.10 importlib-metadata==4.5.0 itsdangerous==2.0.1 Jinja2==3.0.1 joblib==1.0.1 MarkupSafe==2.0.1 multiprocess==0.70.11.1 murmurhash==1.0.5 numpy==1.20.0 packaging==20.9 pandas==1.2.4 pathy==0.5.2 plac==1.1.3 preshed==3.0.5 pyarrow==4.0.1 pydantic==1.7.4 pyparsing==2.4.7 python-dateutil==2.8.1 pytz==2021.1 regex==2021.4.4 requests==2.25.1 sacremoses==0.0.45 scikit-learn==0.24.0 scipy==1.6.3 six==1.16.0 smart-open==3.0.0 spacy==2.3.5 spacy-legacy==3.0.5 srsly==1.0.5 thinc==7.4.5 threadpoolctl==2.1.0 tokenizers==0.10.3 torch==1.8.1 tqdm==4.49.0 transformers==4.5.1 typer==0.3.2 typing-extensions==3.10.0.0 urllib3==1.26.5 wasabi==0.8.2 Werkzeug==2.0.1 xxhash==2.0.2 zipp==3.4.1

Basically I went through your setup.py manually, removing the the ~=, and making it ==. also for sklearn, I installed scikit-learn==0.24.0, not sure what sklearn~=0.0 does. Also the version of spacy is different, was seeing some errors in loading the spacy model, and followed this ticket to resolve them; I'm using spacy==2.3.5. More generally I'm using python3.7 as that's what your medium post was using and am on a mac, OS version 11.4 (Big Sur).

Also the config for your medmen trained CDB class uses the en_core_sci_lg model, not the en_core_sci_md model.

Anyways, the question I have is, in order for the CDB class to have the cui2icd10 key in addl_info filled out, do I need the UMLS license? Seems like that mapping is blank in the provided medmen trained model, and wanted to see if the model trained on NLM would have the icd codes filled out. I tried applying for the license, but am getting 500s on the sign up page right now, will check again later.

It would be nice if all configurable parameters, including MetaCAT's run and model options, would be gathered in a single Config() object, and read them from a YAML formatted file using PyYAML.

Hi @w-is-h . Somewhere between April 10 (7ce8f42) and current master (700c275), there was a change that results in many lines being skipped. Is this intended or a bug? The reason the lines are skipped is not clear to me. I think the code states it should print the exception, but instead prints the CUI.

Hi @w-is-h at our hospital we're using our own "MedCAT model creator", which is basically a pipeline containing the steps that MedCAT documented in Jupyter notebooks. Our code:

• Loads input concepts from CSV and input documents from txt
• Creates vocab, create CDB, do unsupervised training, optionally do supervised training
• Write files to a configured location
• Also contains an integration test with a sample of wikipedia data and UMLS concepts to verify some expected entities are found. That looks a bit like what you are doing in MedCAT/tests/medmentions/. We could adjust our test to use the MedMentions data already included in MedCAT.

It might be nice to put this functionality into MedCAT itself. Are you open for a PR for this? We can also discuss it in more detail in a call if you want.

When creating the CAT object we are passing it cdb.config. This is already contained in the cdb object and so this should probably be an optional parameter.

The function utils.loggers.add_handlers` is used in several modules, which results in independent logging handlers when MedCAT is imported as a library.

The creation of handlers in a library is considered an anti-pattern by the logging maintainers: https://docs.python.org/3/howto/logging-cookbook.html#patterns-to-avoid

It creates handlers that are difficult for the end application to suppress or modify.

If you're amenable to the change, I'll gladly submit a PR.

The 3.0 version of the code base seems to not be compatible scispacy versions 0.4.0 (uses spacy 3.0) or 0.50 (uses spacy 3.2) since MedCAT depends on spacy 3.1. Should we use scispacy 0.5.0 and use spacy 3.2?

If I install MedCAT 1.0.40, I get the error below when calling the MedCAT Service. This error is fixed by installing medcat-1.0.39.

Error:

ValueError: [E966] `nlp.add_pipe` now takes the string name of the registered component factory, not a callable component. Expected string, but got functools.partial(<function tag_skip_and_punct at 0x7ff0b0e12cb0>, config=<medcat.config.Config object at 0x7ff16c125350>) (name: 'tag_skip_and_punct').

- If you created your component with `nlp.create_pipe('name')`: remove nlp.create_pipe and call `nlp.add_pipe('name')` instead.

- If you passed in a component like `TextCategorizer()`: call `nlp.add_pipe` with the string name instead, e.g. `nlp.add_pipe('textcat')`.

- If you're using a custom component: Add the decorator `@Language.component` (for function components) or `@Language.factory` (for class components / factories) to your custom component and assign it a name, e.g. `@Language.component('your_name')`. You can then run `nlp.add_pipe('your_name')` to add it to the pipeline.

Steps to reproduce:

1. pip install -r medcat_service/requirements.txt

2. pip install https://s3-us-west-2.amazonaws.com/ai2-s2-scispacy/releases/v0.4.0/en_core_sci_md-0.4.0.tar.gz

Which generates dependency error at bottom:

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
medcat 1.0.40 requires spacy==2.3.4, but you have spacy 3.0.7 which is incompatible.
Successfully installed catalogue-2.0.6 en-core-sci-md-0.4.0 pathy-0.6.0 pydantic-1.8.2 spacy-3.0.7 spacy-legacy-3.0.8 srsly-2.4.1 thinc-8.0.8 typer-0.3.2

3. . start-service-prod.sh
curl -XPOST http://localhost:5000/api/process -H 'Content-Type: application/json' -d '{"content":{"text":"The patient was diagnosed with leukemia."}}’

4. Receive the following error:

[2021-08-25 21:41:19,751] [ERROR] medcat_service.app.app: Exception on /api/process [POST]
Traceback (most recent call last):
  File "/opt/conda/envs/medcat/lib/python3.7/site-packages/injector/__init__.py", line 804, in get
    return self._context[key]
KeyError: <class 'medcat_service.nlp_service.nlp_service.NlpService'>

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/opt/conda/envs/medcat/lib/python3.7/site-packages/injector/__init__.py", line 804, in get
    return self._context[key]
KeyError: <class 'medcat_service.nlp_processor.medcat_processor.MedCatProcessor'>

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/opt/conda/envs/medcat/lib/python3.7/site-packages/flask/app.py", line 2447, in wsgi_app
    response = self.full_dispatch_request()
  File "/opt/conda/envs/medcat/lib/python3.7/site-packages/flask/app.py", line 1952, in full_dispatch_request
    rv = self.handle_user_exception(e)
  File "/opt/conda/envs/medcat/lib/python3.7/site-packages/flask/app.py", line 1821, in handle_user_exception
    reraise(exc_type, exc_value, tb)
  File "/opt/conda/envs/medcat/lib/python3.7/site-packages/flask/_compat.py", line 39, in reraise
    raise value
  File "/opt/conda/envs/medcat/lib/python3.7/site-packages/flask/app.py", line 1950, in full_dispatch_request
    rv = self.dispatch_request()
  File "/opt/conda/envs/medcat/lib/python3.7/site-packages/flask/app.py", line 1936, in dispatch_request
    return self.view_functions[rule.endpoint](**req.view_args)
  File "/opt/conda/envs/medcat/lib/python3.7/site-packages/flask_injector/__init__.py", line 95, in wrapper
    return injector.call_with_injection(callable=fun, args=args, kwargs=kwargs)
  File "/opt/conda/envs/medcat/lib/python3.7/site-packages/injector/__init__.py", line 1024, in call_with_injection
    owner_key=self_.__class__ if self_ is not None else callable.__module__,
  File "/opt/conda/envs/medcat/lib/python3.7/site-packages/injector/__init__.py", line 111, in wrapper
    return function(*args, **kwargs)
  File "/opt/conda/envs/medcat/lib/python3.7/site-packages/injector/__init__.py", line 1069, in args_to_inject
    instance = self.get(interface)  # type: Any
  File "/opt/conda/envs/medcat/lib/python3.7/site-packages/injector/__init__.py", line 963, in get
    result = scope_instance.get(interface, binding.provider).get(self)
  File "/opt/conda/envs/medcat/lib/python3.7/site-packages/injector/__init__.py", line 111, in wrapper
    return function(*args, **kwargs)
  File "/opt/conda/envs/medcat/lib/python3.7/site-packages/injector/__init__.py", line 806, in get
    provider = InstanceProvider(provider.get(self.injector))
  File "/opt/conda/envs/medcat/lib/python3.7/site-packages/injector/__init__.py", line 291, in get
    return injector.create_object(self._cls)
  File "/opt/conda/envs/medcat/lib/python3.7/site-packages/injector/__init__.py", line 990, in create_object
    self.call_with_injection(cls.__init__, self_=instance, kwargs=additional_kwargs)
  File "/opt/conda/envs/medcat/lib/python3.7/site-packages/injector/__init__.py", line 1024, in call_with_injection
    owner_key=self_.__class__ if self_ is not None else callable.__module__,
  File "/opt/conda/envs/medcat/lib/python3.7/site-packages/injector/__init__.py", line 111, in wrapper
    return function(*args, **kwargs)
  File "/opt/conda/envs/medcat/lib/python3.7/site-packages/injector/__init__.py", line 1069, in args_to_inject
    instance = self.get(interface)  # type: Any
  File "/opt/conda/envs/medcat/lib/python3.7/site-packages/injector/__init__.py", line 963, in get
    result = scope_instance.get(interface, binding.provider).get(self)
  File "/opt/conda/envs/medcat/lib/python3.7/site-packages/injector/__init__.py", line 111, in wrapper
    return function(*args, **kwargs)
  File "/opt/conda/envs/medcat/lib/python3.7/site-packages/injector/__init__.py", line 806, in get
    provider = InstanceProvider(provider.get(self.injector))
  File "/opt/conda/envs/medcat/lib/python3.7/site-packages/injector/__init__.py", line 291, in get
    return injector.create_object(self._cls)
  File "/opt/conda/envs/medcat/lib/python3.7/site-packages/injector/__init__.py", line 990, in create_object
    self.call_with_injection(cls.__init__, self_=instance, kwargs=additional_kwargs)
  File "/opt/conda/envs/medcat/lib/python3.7/site-packages/injector/__init__.py", line 1030, in call_with_injection
    return callable(*full_args, **dependencies)
  File "/home/elisa/MedCATservice/medcat_service/nlp_processor/medcat_processor.py", line 63, in __init__
    self.cat = self._create_cat()
  File "/home/elisa/MedCATservice/medcat_service/nlp_processor/medcat_processor.py", line 234, in _create_cat
    return CAT(cdb=cdb, config=conf, vocab=vocab, meta_cats=meta_models)
  File "/opt/conda/envs/medcat/lib/python3.7/site-packages/medcat/cat.py", line 75, in __init__
    additional_fields=['is_punct'])
  File "/opt/conda/envs/medcat/lib/python3.7/site-packages/medcat/pipe.py", line 38, in add_tagger
    self.nlp.add_pipe(tagger, name='tag_' + name, first=True)
  File "/opt/conda/envs/medcat/lib/python3.7/site-packages/spacy/language.py", line 755, in add_pipe
    raise ValueError(err)
ValueError: [E966] `nlp.add_pipe` now takes the string name of the registered component factory, not a callable component. Expected string, but got functools.partial(<function tag_skip_and_punct at 0x7ff0b0e12cb0>, config=<medcat.config.Config object at 0x7ff16c125350>) (name: 'tag_skip_and_punct').

- If you created your component with `nlp.create_pipe('name')`: remove nlp.create_pipe and call `nlp.add_pipe('name')` instead.

- If you passed in a component like `TextCategorizer()`: call `nlp.add_pipe` with the string name instead, e.g. `nlp.add_pipe('textcat')`.

- If you're using a custom component: Add the decorator `@Language.component` (for function components) or `@Language.factory` (for class components / factories) to your custom component and assign it a name, e.g. `@Language.component('your_name')`. You can then run `nlp.add_pipe('your_name')` to add it to the pipeline.
[25/Aug/2021:21:41:19 +0000] [ACCESSS] 127.0.0.1 "POST /api/process HTTP/1.1" 500 "-" "curl/7.52.1"

When trying to add MedCAT as a dependency to my project I run into:

Using version ^1.1.3 for medcat

Updating dependencies
Resolving dependencies... (29.8s)

  SolverProblemError

      Because no versions of spacy match >3.0.1,<3.0.2 || >3.0.2,<3.0.3 || >3.0.3,<3.0.4 || >3.0.4,<3.0.5 || >3.0.5,<3.0.6 || >3.0.6,<3.0.7 || >3.0.7,<3.1.0
   and spacy (3.0.1) depends on typer (>=0.3.0,<0.4.0), spacy (>=3.0.1,<3.0.2 || >3.0.2,<3.0.3 || >3.0.3,<3.0.4 || >3.0.4,<3.0.5 || >3.0.5,<3.0.6 || >3.0.6,<3.0.7 || >3.0.7,<3.1.0) requires typer (>=0.3.0,<0.4.0).
      And because spacy (3.0.2) depends on typer (>=0.3.0,<0.4.0), spacy (>=3.0.1,<3.0.3 || >3.0.3,<3.0.4 || >3.0.4,<3.0.5 || >3.0.5,<3.0.6 || >3.0.6,<3.0.7 || >3.0.7,<3.1.0) requires typer (>=0.3.0,<0.4.0).
      And because spacy (3.0.3) depends on typer (>=0.3.0,<0.4.0)
   and spacy (3.0.4) depends on typer (>=0.3.0,<0.4.0), spacy (>=3.0.1,<3.0.5 || >3.0.5,<3.0.6 || >3.0.6,<3.0.7 || >3.0.7,<3.1.0) requires typer (>=0.3.0,<0.4.0).
      And because spacy (3.0.5) depends on typer (>=0.3.0,<0.4.0)
   and spacy (3.0.6) depends on typer (>=0.3.0,<0.4.0), spacy (>=3.0.1,<3.0.7 || >3.0.7,<3.1.0) requires typer (>=0.3.0,<0.4.0).
      Because no versions of medcat match >1.1.3,<2.0.0
   and medcat (1.1.3) depends on spacy (>=3.0.1,<3.1.0), medcat (>=1.1.3,<2.0.0) requires spacy (>=3.0.1,<3.1.0).
      Thus, medcat (>=1.1.3,<2.0.0) requires typer (>=0.3.0,<0.4.0) or spacy (3.0.7).
  (1) So, because spacy (3.0.7) depends on typer (>=0.3.0,<0.4.0), medcat (>=1.1.3,<2.0.0) requires typer (>=0.3.0,<0.4.0).
  
      Because no versions of typer match >0.3.0,<0.3.1 || >0.3.1,<0.3.2 || >0.3.2,<0.4.0
   and typer (0.3.0) depends on click (>=7.1.1,<7.2.0), typer (>=0.3.0,<0.3.1 || >0.3.1,<0.3.2 || >0.3.2,<0.4.0) requires click (>=7.1.1,<7.2.0).
      And because typer (0.3.1) depends on click (>=7.1.1,<7.2.0)
   and typer (0.3.2) depends on click (>=7.1.1,<7.2.0), typer (>=0.3.0,<0.4.0) requires click (>=7.1.1,<7.2.0).
      And because medcat (>=1.1.3,<2.0.0) requires typer (>=0.3.0,<0.4.0) (1), medcat (>=1.1.3,<2.0.0) requires click (>=7.1.1,<7.2.0)
      So, because ehrapy depends on both click (^8.0.2) and medcat (^1.1.3), version solving failed.

From what I can see you are pinning a version of spaCy which requires typer <=0.4.0
The latest spaCy allows for typer dependencies up to 0.5.0. This version has added support for click 8.x : tiangolo/typer@b972981

Could you please upgrade the spaCy version, (test whether it works with Click 8.x) and release a new version?

This would be highly appreciated. Happy to provide more detailed if required. I urgently need this to work with my environment.

Thanks!

Hi @w-is-h ,

Interesting and well-done work!

I was just wondering about F1 metric that was used in this project. It seems that you use simplified mean of Precision and Recall to calculate F1 instead of conventional one: f1 = 2*(Precision * Recall)/(Precision + Recall).

This also reflects in published paper: page 5 Table 1.

It would be great if you could explain, otherwise, using conventional F1 would be good for comparison and communication purposes.

It seems both files in the links https://s3-eu-west-1.amazonaws.com/zkcl/vocab.dat and https://s3-eu-west-1.amazonaws.com/zkcl/cdb-medmen-v1.dat are not available at the moment.

It says access denied.
Could you have a look at this?

Best,
A

Are there results we can leverage to decide how to use the provided 'medcat_acc'? As in, precision-recall performance at various thresholds. Thanks!

Torch==1.4.0 is not available and triggers build errors with Python 3.9, meaning that this application also cannot be installed. Tested on Ubuntu 18. Changing torch version then leads to a failure in TorchVision dependency. Changing that to the latest triggers a tokenizer installation error (because 0.8.0 is not PEP compliant).

Minimum reproducible example:

$ python3.9 -m venv medcat_install
$ source medcat_install/bin/activate
(medcat_issue) $ pip install MedCat
Collecting MedCat
  Downloading medcat-0.4.0.6-py3-none-any.whl (70 kB)
     |████████████████████████████████| 70 kB 650 kB/s
Collecting tokenizers~=0.8.0
  Downloading tokenizers-0.8.1.tar.gz (97 kB)
     |████████████████████████████████| 97 kB 903 kB/s
  Installing build dependencies ... done
  Getting requirements to build wheel ... done
    Preparing wheel metadata ... done
Collecting numpy~=1.18
  Downloading numpy-1.20.1-cp39-cp39-manylinux2010_x86_64.whl (15.4 MB)
     |████████████████████████████████| 15.4 MB 16.3 MB/s
Collecting gensim~=3.7
  Downloading gensim-3.8.3.tar.gz (23.4 MB)
     |████████████████████████████████| 23.4 MB 9.2 MB/s
ERROR: Could not find a version that satisfies the requirement torch~=1.4.0 (from MedCat) (from versions: 0.1.2, 0.1.2.post1, 0.1.2.post2, 1.7.1, 1.8.0)
ERROR: No matching distribution found for torch~=1.4.0 (from MedCat)

Politely take them from some other open-source repo (example)

Tutorials also need an update for new Word2Vec parameter names, e.g. size to vector_size:

Originally posted by @sandertan in #156 (comment)

Hello getting below error while using prebuid models. How to resolve this?

ConfigValidationError:

Config validation error

ner -> incorrect_spans_key extra fields not permitted

{'nlp': <spacy.lang.en.English object at 0x00000138780A7670>, 'name': 'ner', 'incorrect_spans_key': None, 'model': {'@architectures': 'spacy.TransitionBasedParser.v2', 'state_type': 'ner', 'extra_state_tokens': False, 'hidden_width': 64, 'maxout_pieces': 2, 'use_upper': True, 'nO': None, 'tok2vec': {'@architectures': 'spacy.Tok2Vec.v2', 'embed': {'@architectures': 'spacy.MultiHashEmbed.v2', 'width': 96, 'attrs': ['NORM', 'PREFIX', 'SUFFIX', 'SHAPE'], 'rows': [5000, 2500, 2500, 2500], 'include_static_vectors': True}, 'encode': {'@architectures': 'spacy.MaxoutWindowEncoder.v2', 'width': 96, 'depth': 4, 'window_size': 1, 'maxout_pieces': 3}}}, 'moves': None, 'update_with_oracle_cut_size': 100, '@factories': 'n

OSError: [E050] Can't find model 'en_core_sci_md'. It doesn't seem to be a shortcut link, a Python package or a valid path to a data directory.

Fix: Use the latest version of scispacy

Hi,

the MIMIC-IV dataset (triage.csv) uses a couple of abbreviations which are vital for the extraction of diagnosis etc from freetext, but currently are not really well supported by MedCAT.

I was using your hosted instance (https://medcat.rosalind.kcl.ac.uk/) for these quick tests.

Examples:

1. Hypertension, R Leg numbness, R Shoulder pain -> extracts 3 findings, but does not add the "R" as attributes. It is absolutely essential that these findings were only on the right part of the body.
2. Dizziness, Malaise, n/v/d -> extracts 2 findings, but misses the n/v/d which is short for nausea/vomiting/diarrhea

Do you think that you (we) can improve on the support for such abbreviations in reasonable way?

Hey,
I'm using medcat==1.2.8.

I'm trying to create a CAT object from my vocab and cdb like this:

# load vocab
vocab = Vocab.load("./vocab.dat")
# load cdb
cdb = CDB.load("./cdb-medmen-v1.dat")
# create model
cat = CAT(cdb=concept_db, vocab=vocabulary)

...

However this results in :

/home/myname/anaconda3/envs/poetry3_8/lib/python3.8/site-packages/medcat/cat.py:100  │
│ in __init__                                                                               │
│                                                                                           │
│     97 │   │   │   self.config = config                                                   │
│     98 │   │   │   self.cdb.config = config                                               │
│     99 │   │   self._meta_cats = meta_cats                                                │
│ ❱  100 │   │   self._create_pipeline(self.config)                                         │
│    101 │                                                                                  │
│    102 │   def _create_pipeline(self, config):                                            │
│    103 │   │   # Set log level                                                            │
│                                                                                           │
│ /home/myname/anaconda3/envs/poetry3_8/lib/python3.8/site-packages/medcat/cat.py:107  │
│ in _create_pipeline                                                                       │
│                                                                                           │
│    104 │   │   self.log.setLevel(config.general['log_level'])                             │
│    105 │   │                                                                              │
│    106 │   │   # Build the pipeline                                                       │
│ ❱  107 │   │   self.pipe = Pipe(tokenizer=spacy_split_all, config=config)                 │
│    108 │   │   self.pipe.add_tagger(tagger=tag_skip_and_punct,                            │
│    109 │   │   │   │   │   │   │    name='skip_and_punct',                                │
│    110 │   │   │   │   │   │   │    additional_fields=['is_punct'])                       │
│                                                                                           │
│ /home/myname/anaconda3/envs/poetry3_8/lib/python3.8/site-packages/medcat/pipe.py:40  │
│ in __init__                                                                               │
│                                                                                           │
│    37 │   log = add_handlers(log)                                                         │
│    38 │                                                                                   │
│    39 │   def __init__(self, tokenizer: Tokenizer, config: Config) -> None:               │
│ ❱  40 │   │   self._nlp = spacy.load(config.general['spacy_model'],                       │
│       disable=config.general['spacy_disabled_components'])                                │
│    41 │   │   if config.preprocessing['stopwords'] is not None:                           │
│    42 │   │   │   self._nlp.Defaults.stop_words = set(config.preprocessing['stopwords'])  │
│    43 │   │   self._nlp.tokenizer = tokenizer(self._nlp, config)                          │
│                                                                                           │
│ /home/myname/anaconda3/envs/poetry3_8/lib/python3.8/site-packages/spacy/__init__.py: │
│ 51 in load                                                                                │
│                                                                                           │
│   48 │   │   keyed by section values in dot notation.                                     │
│   49 │   RETURNS (Language): The loaded nlp object.                                       │
│   50 │   """                                                                              │
│ ❱ 51 │   return util.load_model(                                                          │
│   52 │   │   name, vocab=vocab, disable=disable, exclude=exclude, config=config           │
│   53 │   )                                                                                │
│   54                                                                                      │
│                                                                                           │
│ /home/myname/anaconda3/envs/poetry3_8/lib/python3.8/site-packages/spacy/util.py:354  │
│ in load_model                                                                             │
│                                                                                           │
│    351 │   │   return load_model_from_path(name, **kwargs)                                │
│    352 │   if name in OLD_MODEL_SHORTCUTS:                                                │
│    353 │   │   raise IOError(Errors.E941.format(name=name, full=OLD_MODEL_SHORTCUTS[name] │
│ ❱  354 │   raise IOError(Errors.E050.format(name=name))                                   │
│    355                                                                                    │
│    356                                                                                    │
│    357 def load_model_from_package(                                                       │
╰───────────────────────────────────────────────────────────────────────────────────────────╯
OSError: [E050] Can't find model 'en_core_sci_lg'. It doesn't seem to be a Python package or 
a valid path to a data directory.

Any ideas on why this is failing? I tried to install the en_core_sci_lg model (just downloaded it) but it cannot be installed from spacy directly.

Any help would be appreciated!

Hi,

great work!
I was working through the full Medcat pipeline tutorial but am bugged about one thing: discerning between whether an entity has a positive connotation (= was indeed found) or a negative connotation (= was not actually found).
I thought that the status (which I cannot see here?) should tell us that?
The output does not allow me to classify this:

[('H',
  {'entities': {1: {'acc': 1.0,
     'context_similarity': 1.0,
     'cui': 'C0035078',
     'detected_name': 'kidney~failure',
     'end': 21,
     'id': 1,
     'meta_anns': {},
     'pretty_name': 'Kidney Failure',
     'source_value': 'kidney failure',
     'start': 7,
     'tuis': ['T047'],
     'types': ['Disease or Syndrome']}},
   'text': 'He has kidney failure',
   'tokens': []}),
 ('S',
  {'entities': {2: {'acc': 1.0,
     'context_similarity': 1.0,
     'cui': 'C0035078',
     'detected_name': 'kidney~failure',
     'end': 32,
     'id': 2,
     'meta_anns': {},
     'pretty_name': 'Kidney Failure',
     'source_value': 'kidney failure',
     'start': 18,
     'tuis': ['T047'],
     'types': ['Disease or Syndrome']}},
   'text': 'She does not have kidney failure',
   'tokens': []}),

Hello,

Does MedCAT have models or use datasets that are not in english but a different language like french or spanish ?

The updated Google colab tutorial uses cat.train = True and cat(text, do_train=True) in a for loop for unsupervised training, but the MedCAT inline code documentation strongly suggests to use cat.train(). It might be good to update the tutorial to reflect the preferred method.

I tried to use the command

cat.spacy_cat.TUI_FILTER = tui_list

that I found in the MedCAT article:
https://towardsdatascience.com/medcat-extracting-diseases-from-electronic-health-records-f53c45b3d1c1

but I got the following error:
AttributeError: 'CAT' object has no attribute 'spacy_cat'

A recent PR introduced a bug with logs, instead of

log.info('something %s and another %s', (1, 2))

we need to do:

log.info('something %s and another %s', 1, 2)

All logs with more than one argument have this problem, we need to find and fix them.

Facing the below issue while trying to train a custom vocab with word2vec vectors. cdb is a the set of core snomed database. While I am trying to train the clinical text of 5000 visits. I have noticed that it throws error for certain text while certain texts are fine

Using the below code.

cat.spacy_cat.PREFER_FREQUENT = True
cat.spacy_cat.PREFER_ICD10 = False
cat.spacy_cat.WEIGHTED_AVG = True
cat.spacy_cat.MIN_CONCEPT_LENGTH = 3 # Ignore concepts (diseases) <= 3 characters
cat.spacy_cat.MIN_ACC = 0.2 # Confidence cut-off, everything bellow will not be displayed

for i, text in enumerate(data['text'].values):
# This will now run the training in the background
try:
_ = cat(text)
# So we know how things are moving
#if i % 100 == 0:
# print("Finished {} - text blocks".format(i))
except KeyboardInterrupt:
print('Manually Exited')
break
except:
print(data[data['id'] == i]['name'] )
continue

Print statistics on the CDB after training

cat.cdb.print_stats()

Disable the training mode

cat.train = False

File "", line 11, in
_ = cat(text)

File "C:\ProgramData\Anaconda3\lib\site-packages\medcat\cat.py", line 92, in call
return self.nlp(text)

File "C:\ProgramData\Anaconda3\lib\site-packages\medcat\utils\spacy_pipe.py", line 53, in call
return self.nlp(text)

File "C:\ProgramData\Anaconda3\lib\site-packages\spacy\language.py", line 439, in call
doc = proc(doc, **component_cfg.get(name, {}))

File "C:\ProgramData\Anaconda3\lib\site-packages\medcat\spacy_cat.py", line 470, in call
self.cat_ann.add_ann(raw_name, tkns, doc, self.to_disamb, doc_words)

File "C:\ProgramData\Anaconda3\lib\site-packages\medcat\basic_cat_ann.py", line 40, in add_ann
self._cat._add_ann(cui, doc, tkns, acc=1, name=name)

File "C:\ProgramData\Anaconda3\lib\site-packages\medcat\spacy_cat.py", line 373, in _add_ann
self._add_cntx_vec(cui, doc, tkns)

File "C:\ProgramData\Anaconda3\lib\site-packages\medcat\spacy_cat.py", line 291, in _add_cntx_vec
negs = self.vocab.get_negative_samples(n=self.CNTX_SPAN * 2)

File "C:\ProgramData\Anaconda3\lib\site-packages\medcat\utils\vocab.py", line 99, in get_negative_samples
inds = np.random.randint(0, len(self.unigram_table), n)

File "mtrand.pyx", line 745, in numpy.random.mtrand.RandomState.randint

File "_bounded_integers.pyx", line 1363, in numpy.random._bounded_integers._rand_int32

ValueError: low >= high

Hi @w-is-h

Great work on the updated meta_cat, not sure what the functional changes were but the performance on our test sets significantly increased.

One thing I noticed was that when doing a run with auto_save_model set to True, to save the Config, I still needed to do a manual MetaCAT.save(). To prevent this, it might be nice to use this save function:

instead of the auto_save_model's own save function:

Hi @w-is-h ,
I need ICD10 code for the disease. As define in Unsupervised training and NER+L
tutorial 3.2 when we run the code

Let's test the multi processing function first

in_data = [(3, "cancer")]
results = cat.multi_processing(in_data, nproc=2)
print(results)

we got
[(3, {'entities': [{'cui': 'C0006826', 'tui': 'T191', 'type': 'Neoplastic Process', 'source_value': 'cancer', 'acc': '0.7050448862662964', 'start': 0, 'end': 6, 'id': '0', 'pretty_name': 'malignant tumours', 'icd10': '', 'umls': '', 'snomed': ''}], 'text': 'cancer'})]

In this output we are not getting 'icd10','umls',&'snomed' values. How we can access them ? Please suggest...
Thanks

From paper Kraljevic, Zeljko, et al. "MedCAT--medical concept annotation tool." arXiv preprint arXiv:1912.10166 (2019), "An annotation by MedCAT is considered correct only if the exact text value was found and the annotation was linked to the correct concept in the CDB."

My understand is that the exact text value would include the start and end index of prediction, but from the cat.py, it seems that the end index is not included, only the start and the cui? Thank you.

Could we gave a way to set/unset the CUDA flag for the metacat models. Maybe this could be in the config for the model pack somewhere?

Hey,

First of all thanks for the great package.

I'm using medcat 1.2.8 and I noticed the following issue:

Example:

text = "Patient suffers from diabetes. Denies hypertension, psychosis and glaucoma"

# let cat be the CAT object, that has been trained and initialized using the model pack/example data from the docs
annotated_text = cat.get_entities(text)

This results in:

{'entities': {2: {'pretty_name': 'Diabetes',
   'cui': 'C0011847',
   'type_ids': ['T047'],
   'types': ['Disease or Syndrome'],
   'source_value': 'diabetes',
   'detected_name': 'diabetes',
   'acc': 0.6452550625169893,
   'context_similarity': 0.6452550625169893,
   'start': 20,
   'end': 28,
   'icd10': [],
   'ontologies': [],
   'snomed': [],
   'id': 2,
   'meta_anns': {'Status': {'value': 'Affirmed',
     'confidence': 0.999997079372406,
     'name': 'Status'}}},
  3: {'pretty_name': 'Hypertensive disease',
   'cui': 'C0020538',
   'type_ids': ['T047'],
   'types': ['Disease or Syndrome'],
   'source_value': 'hypertension',
   'detected_name': 'hypertension',
   'acc': 0.6790682188733697,
   'context_similarity': 0.6790682188733697,
   'start': 37,
   'end': 49,
   'icd10': [],
   'ontologies': [],
   'snomed': [],
   'id': 3,
   'meta_anns': {'Status': {'value': 'Other',
     'confidence': 0.9918639063835144,
     'name': 'Status'}}},
  4: {'pretty_name': 'Psychotic Disorders',
   'cui': 'C0033975',
   'type_ids': ['T048'],
   'types': ['Mental or Behavioral Dysfunction'],
   'source_value': 'psychosis',
   'detected_name': 'psychosis',
   'acc': 0.3484492297815132,
   'context_similarity': 0.3484492297815132,
   'start': 51,
   'end': 60,
   'icd10': [],
   'ontologies': [],
   'snomed': [],
   'id': 4,
   'meta_anns': {'Status': {'value': 'Affirmed',
     'confidence': 0.8026704788208008,
     'name': 'Status'}}},
  5: {'pretty_name': 'Glaucoma',
   'cui': 'C0017601',
   'type_ids': ['T047'],
   'types': ['Disease or Syndrome'],
   'source_value': 'glaucoma',
   'detected_name': 'glaucoma',
   'acc': 0.3833850208933218,
   'context_similarity': 0.3833850208933218,
   'start': 65,
   'end': 73,
   'icd10': [],
   'ontologies': [],
   'snomed': [],
   'id': 5,
   'meta_anns': {'Status': {'value': 'Affirmed',
     'confidence': 0.9999270439147949,
     'name': 'Status'}}}},
 'tokens': []}

As one can see, medcat correctly gets, that there is a diabetes but no hypertension diagnosis. But the "denies" context seems to get lost/ignored in the enumeration after hypertension so psychosis and glaucoma are labeled as "affirmed" although, they should also be "Other" (like negative).

Is this a known Issue? Are there any approaches to solve such issues?

Many thanks in advance ;)

• MedCAT CDB maker uses the name type_ids
• MedCAT documentation (https://github.com/CogStack/MedCAT/blob/master/examples/README.md) calls it semantic_type_ids (and also mentions group, but I think this is not used)
• UMLS and a previous MedCAT version call it tui
• MedCATTrainer uses both the name tui and semantic_type in the API result

It would be nice to align the naming.

I envisage timestamp being related to the document rather than the relationship.
It could be stored with both node and edge of course, but storing on the node would be more succint.

medcat/neo/data_preparation.py
line 80: create_documents_csv needs timestamp: toInteger(row.timestamp), \n' moved to
line 184: create_document2conceot_csv could have timestamp removed if necessary

Thank you for sharing the files. If it is possible could you answer the following question?

1. Are the context vectors that are used for disambiguation included in the CDB file? I am asking as I am not sure that when we will use the CDB and the VOCAB file we will be using the pre-trained model that was trained on the MIMIC-III?

2. How many words are included in the Vocabulary of Umls?

3. Is there only one MetaCAT Status pre-trained model (the one that you include in the README.md) or there is another (umls) model ? and does it only use the lstm.dat model?

Hi there,

Whenever I attempt to use the Snomed preprocess utility set, I have file not found errors:

from medcat.utils.preprocess_snomed import Snomed
snomed = Snomed("C:/path/to/dir/uk_sct2cl_32.7.0_20211124000001Z/")
cdf = snomed.to_concept_df()

Returns

---------------------------------------------------------------------------
FileNotFoundError                         Traceback (most recent call last)
<ipython-input-23-5eb639e435ed> in <module>
----> 1 cdf = snomed.to_concept_df()

~\Projects\nlp\env\lib\site-packages\medcat\utils\preprocess_snomed.py in to_concept_df(self)
     50                     snomed_v = m.group(1)
     51 
---> 52             int_terms = parse_file(f'{contents_path}/sct2_Concept_Snapshot_{snomed_v}_{snomed_release}.txt')
     53             active_terms = int_terms[int_terms.active == '1']
     54             del int_terms

~\Projects\nlp\env\lib\site-packages\medcat\utils\preprocess_snomed.py in parse_file(filename, first_row_header, columns)
      7 
      8 def parse_file(filename, first_row_header=True, columns=None):
----> 9     with open(filename, encoding='utf-8') as f:
     10         entities = [[n.strip() for n in line.split('\t')] for line in f]
     11         return pd.DataFrame(entities[1:], columns=entities[0] if first_row_header else columns)

FileNotFoundError: [Errno 2] No such file or directory: 'C:/path/to/dir/uk_sct2cl_32.7.0_20211124000001Z/SnomedCT_UKClinicalRefsetsRF2_PRODUCTION_20211124T000001Z\\Snapshot\\Terminology/sct2_Concept_Snapshot_INT_20211124.txt'

Where the file is named sct2_Concept_UKCRSnapshot_GB1000000_20211124.txt

Best wishes,

Keiron

Hey,

consider the following example:

# some setup code here, cat is the medcat object
cat.create_model_pack("~/MyDir/MyProjects/MyModelPack)

This code results in having a directory named ~ inside the current working directory, which itsself contains MyDir, which contains MyProjects which contains the actual model pack named MyModelPack.
I expected this code to save the actual model pack in the path instead of creating this path in the cwd. I guess that's due to the line os.makedirs(save_dir_path, exist_ok=True) in the create_model_pack function. Is this intended behaviour?

Best ;)

When I attempt to call the multiprocessing function, I get a pickling error. My code is as follows:

import pandas as pd
import numpy as np
from medcat.vocab import Vocab
from medcat.cdb import CDB
from medcat.cat import CAT
from medcat.config import Config
from tqdm.notebook import tqdm
from medcat.meta_cat import MetaCAT

data = pd.read_csv('mri_reports.csv')
vocab = Vocab.load('C:/Users/Tom/Documents/Christie/MRI Medcat/Vocab and CBD/vocab.dat')

Config

config = Config()
config.general['spacy_model'] = 'en_core_sci_md'

tui_filter = ['T047'] # Detect only Disease and Mental Disorders
cui_filters = set()
for tui in tui_filter:
cui_filters.update(cdb.addl_info['type_id2cuis'][tui])
config.linking['filters']['cuis'] = cui_filters

Get the status model for meta_annotations

mc_status = MetaCAT(save_dir='C:/Users/Tom/Documents/Christie/MRI Medcat/Vocab and CBD/mc_status')
mc_status.load()
cdb = CDB.load("test_cdb.dat")
cat = CAT(cdb=cdb, config=cdb.config, vocab=vocab)

batch_size = 10
batch = []
cnt = 0
for id, row in data.iterrows():
text = row['report_formatted']
# Skip text if under 10 characters, not really necessary as we have filtered before,
#but I like to be sure.
if len(text) > 10:
batch.append((id, text))

if len(batch) > batch_size or id == len(data) - 1:
    # Update the number of processors depending on your machine.
    results = cat.multiprocessing(batch, nproc=2)

When I get to calling the multiprocessing function, the code errors with the following:

PicklingError Traceback (most recent call last)
in
11 if len(batch) > batch_size or id == len(data) - 1:
12 # Update the number of processors depending on your machine.
---> 13 results = cat.multiprocessing(batch, nproc=2)
14
15 for pair in results:

~\AppData\Roaming\Python\Python38\site-packages\medcat\cat.py in multiprocessing(self, in_data, nproc, batch_size_chars, max_ram_percentage, only_cui, addl_info)
747 p = Process(target=self._mp_cons, kwargs={'in_q': in_q, 'out_dict': out_dict, 'pid': i, 'only_cui': only_cui,
748 'addl_info': addl_info, 'max_ram_percentage': max_ram_percentage})
--> 749 p.start()
750 procs.append(p)
751

~.conda\envs\medcat\lib\multiprocessing\process.py in start(self)
119 'daemonic processes are not allowed to have children'
120 _cleanup()
--> 121 self._popen = self._Popen(self)
122 self._sentinel = self._popen.sentinel
123 # Avoid a refcycle if the target function holds an indirect

~.conda\envs\medcat\lib\multiprocessing\context.py in _Popen(process_obj)
222 @staticmethod
223 def _Popen(process_obj):
--> 224 return _default_context.get_context().Process._Popen(process_obj)
225
226 class DefaultContext(BaseContext):

~.conda\envs\medcat\lib\multiprocessing\context.py in _Popen(process_obj)
325 def _Popen(process_obj):
326 from .popen_spawn_win32 import Popen
--> 327 return Popen(process_obj)
328
329 class SpawnContext(BaseContext):

~.conda\envs\medcat\lib\multiprocessing\popen_spawn_win32.py in init(self, process_obj)
91 try:
92 reduction.dump(prep_data, to_child)
---> 93 reduction.dump(process_obj, to_child)
94 finally:
95 set_spawning_popen(None)

~.conda\envs\medcat\lib\multiprocessing\reduction.py in dump(obj, file, protocol)
58 def dump(obj, file, protocol=None):
59 '''Replacement for pickle.dump() using ForkingPickler.'''
---> 60 ForkingPickler(file, protocol).dump(obj)
61
62 #

PicklingError: Can't pickle <function at 0x000001A8DBFC8DC0>: attribute lookup on medcat.config failed

Seems to be an error with the python's native multiprocessing module not supporting lambda functions?

I'm running python 3.8.8.

Any help would be much appreciated.

Hi,
I am running some experiments with medcat. I have a UMLS license and was wondering whether there are instructions for running the build process anywhere? I've noticed the colab on custom vocabs and perhaps the process for UMLS is the same?

Thanks

I am using the get_entities() method from the CAT class on some arbitrary text (i.e. cat.get_entities(text)) to extract annotations. Sometimes, the "pretty_name" field is empty even though the rest of the fields are populated. Example below. Has anyone else encountered this and know if it's something that would be happening on my end?

Entity 1
{   'acc': 0.2766362802720986,
    'context_similarity': 0.2766362802720986,
    'cui': 'C0542502',
    'detected_name': 'iodination',
    'end': 684,
    'icd10': [],
    'id': 59,
    'meta_anns': {},
    'ontologies': ['CSP', 'MTH', 'MSH', 'AOD'],
    'pretty_name': 'Iodination reaction',
    'snomed': [],
    'source_value': 'iodination',
    'start': 674,
    'tuis': ['T070'],
    'types': ['Natural Phenomenon or Process']}

Entity 2
{   'acc': 1.0,
    'context_similarity': 1.0,
    'cui': 'C0459768',
    'detected_name': 'mercurial',
    'end': 714,
    'icd10': [],
    'id': 60,
    'meta_anns': {},
    'ontologies': ['CHV', 'SNOMEDCT_US'],
    'pretty_name': '',
    'snomed': ['S-280907001'],
    'source_value': 'mercurial',
    'start': 705,
    'tuis': ['T121'],
    'types': ['Pharmacologic Substance']}

Hi,

your 4.3 tutorial https://colab.research.google.com/drive/1apaFscR1a5shzuhg6nLM4lWxgvVbn8f1#scrollTo=TvvCIyv0afMZ fails due to:

FileNotFoundError                         Traceback (most recent call last)

<ipython-input-7-1892cf65204d> in <module>()
     18 
     19 # Get the status model for meta_annotations
---> 20 mc_status = MetaCAT.load("mc_status")
     21 
     22 # Create the full pipeline with models for meta-annotations

1 frames

/content/MedCAT/medcat/config.py in load(cls, save_path)
    105 
    106         # Read the jsonpickle string
--> 107         with open(save_path) as f:
    108             config_dict = jsonpickle.decode(f.read())
    109 

FileNotFoundError: [Errno 2] No such file or directory: 'mc_status/config.json'

at

# Get the status model for meta_annotations
mc_status = MetaCAT.load("mc_status")

Thanks!

Thanks for addressing my previous question.

It seems that the mc_status file is not available.

Also the Meta Annotations with MedCAT tutorial seems not complete. I did not find the code for training a model for meta annotations.

In practice, would you suggest that I just use mc_status, if available later, for contextual detection (experiencer, negation, temporality, etc.)?

Best wishes,
A

Many functions in MedCAT seem to be tailored for English words, in which diacritics are quite rare. However, in Dutch, and I think in some other languages as well, they can be quite common. Because MedCAT often does not take into account words containing diacritics (for example in spell checker

Perhaps some of these issues can be resolved by replacing usage of A-Za-z with [A-Za-zÀ-ÖØ-öø-ÿ] (see https://stackoverflow.com/a/26900132/4141535). I checked some of these occurrences in the MedCAT codebase but it might be better to have expert @w-is-h look into this :)

For an English text and concept test example we can use Ménière's disease:

Ménière's disease (MD) is a disorder of the inner ear that is characterized by episodes of vertigo, tinnitus, hearing loss, and a fullness in the ear. (https://en.wikipedia.org/wiki/Ménière%27s_disease)

Hi,

Currently having an issue installing the medcat package due to the dependencies it's installing first.

Running the pip install medcat:

Collecting medcatNote: you may need to restart the kernel to use updated packages.
  Using cached medcat-0.4.0.2-py3-none-any.whl (70 kB)
Collecting sklearn~=0.0
  Using cached sklearn-0.0.tar.gz (1.1 kB)
Requirement already satisfied: scipy~=1.4 in c:\users\mcheng\anaconda3\lib\site-packages (from medcat) (1.5.0)
Collecting spacy==2.2.4
  Using cached spacy-2.2.4-cp38-cp38-win_amd64.whl (10.1 MB)

ERROR: Could not find a version that satisfies the requirement torch~=1.4.0 (from medcat) (from versions: 0.1.2, 0.1.2.post1, 0.1.2.post2)
ERROR: No matching distribution found for torch~=1.4.0 (from medcat)

Looking into this, there's only mac and linux versions of torch 1.4.0.

Not sure on how I can resolve this on my end. Would a switch from torch to pytorch work?

Running this as a user on a Windows Server 2012 (R2 Standard) with Intel Xeon CPU E5-2650 v4 @ 2.20Ghz.
(I'm currently trying to figure out docker to see if I can use iCAT instead - it's looking unlikely)

Hello,

How to download the complete UML model. I have UML license but unable to find it.

Additionally, Is there any way that we map the synonymous information to the correct pretty name?

E.g. In term "calcaneal fracture" is it identifies it as different entities and not as one might be because it doesn't recognize the work calcaneus to calcaneus. but it maps "Heal bone fracture" correctly to calcaneus. Will this synonym resolution is possible if we try to integrate the complete UML here.

Hey everyone,

great work with MedCAT!

I do have one issue, I can't figure out. Could you help me out how to load the status model for meta_annotations?

Im getting the same error, both local and in the colab (https://colab.research.google.com/drive/1apaFscR1a5shzuhg6nLM4lWxgvVbn8f1#scrollTo=TvvCIyv0afMZ):

FileNotFoundError: [Errno 2] No such file or directory: 'mc_status/config.json'

I have downloaded all required files as per the tutorials/colabs. All the paths are existing paths. It seems there is not config.json file in the mc_status folder after unzipping it.

Any suggestions?

Hi,
While working through your TDS article the first Collab notebook under the Building custom models heading contains csv files that use the "name" header instead of the expected "str" header. This caused an error and required a manual edit and reupload. Thought you might want to know about that.

The previous version of MedCAT (<1.0) showed useful info (CUI, primary name of concept, TUI, name of TUI, context similarity) when rendering a MedCAT doc with displaCy. The latest version just shows "CONCEPT".

Any plans of returning this or similar functionality? It's great for generating screenshots for presentations :)

Additionally, it would be nice to configure what data is displayed, and add colors (e.g. per type)

Quick question: are there any recommendations for how to incorporate approximate string search into the pipeline? I noticed MetCAT doesn't quite handle cases like couhg

Is there any wiki/help guide/Readme on the cdb.config parameters (eg. cdb.config.ner , cdb.config.linking, etc.).
Just want to know what these parameters do, and how to use them

Hi again,

This issue is a little more serious. The 2 .dat files you link in the readme are encoded as bytes and cannot be read in by the Vocab class. I could not decode them using the code below and a range of variants I found from Stack overflow. Code to decode: [line.decode() for line in io.open('vocab.dat', 'rb')]

I was not able to get passed this today so I don't know if the CDB file is any different. Am i doing something wrong? missing a step?