Export the train data and imported to presidio

[Bamorem]

I have been using presidio_analyzer in my local host.

I used presidio-research to generate a dataset with Fake and then run the train/run/dev.

the output is 3 JSON files.
I used the same generated dataset to generate the .spacy file

Now my question is how to integrate the trained data to my local presidio_analyzer to run with the trained data.

I feel like it's missing the integration steps ^^

[omri374]

Hi, have you trained a spaCy model using those samples, or would you like to evaluate Presidio using those samples?

[Bamorem]

Maybe I don't get how it should work, but I have Presidio that uses en_core_web_trf (spaCy model, right ?)

But Presidio (presidio_analyzer) doesn't perform well in some cases so I wanted to train my presidio_analyzer to perform better.

So that's when I found out about presidio-research that says it would be able to train it. I followed all the steps and I thought about the output.spacy would need to be imported to my presidio_analyzer. I'm I wrong? How can I train my local presidio_analyzer for it to perform better based on my dataset?

I would love some help ^^

[omri374]

What you can consider, is to train a new model using spaCy (or other libraries) and then integrate it into Presidio.
More on training using spaCy: https://spacy.io/usage/training
Once you have the model installed locally, you can configure Presidio to use the custom model instead of the en_core_web_trf one: https://microsoft.github.io/presidio/tutorial/05_languages/

[Bamorem]

Okay so Using the link that you send me (https://spacy.io/usage/training) I did those two steps :

python -m spacy init fill-config ./base_config.cfg ./config.cfg
python -m spacy train config.cfg --output ./output --paths.train ./train.spacy --paths.dev ./dev.spacy

Here again, I guess the file ./train.spacy is the one generated like presidio-research suggests:

from presidio_evaluator import InputSample
dataset = InputSample.read_dataset_json("data/synth_dataset_v2.json")
InputSample.create_spacy_dataset(dataset, output_path="dataset.spacy")

Am I right or here again dataset.spacy and train.spacy are two different things? if yes how do I get this train.spacy file?

I think I can use presidio-research :

• https://github.com/microsoft/presidio-research/blob/master/notebooks/models/Create%20datasets%20for%20Spacy%20training.ipynb
  to generate the train.spacy and the dev.spacy (is it the test or validation)

Tell me if I'm wrong.

[omri374]

Hi, we split the dataset into three, where train.spacy and dev.spacy are used for training/validation, and test.spacy will be used for evaluation. dataset.spacy is just an example on how to create a spacy dataset.

[Bamorem]

After I run the training I get those outputs.

now how do I integrate it into my local presidio_analyzer ?

This is the part missing from the documentation I guess.

[omri374]

To serialize/desrialize, see this spaCy doc: https://spacy.io/usage/saving-loading

To integrate it into Presidio, see this issue: microsoft/presidio#822

[Bamorem]

I'm sorry but I got even more lost...

So the next step after the spacy train is to serialize/desrialize the output model ?

Also, the second link to integrate is still using en_core_web_sm how do I use the modal that I trained?

Sorry for the question but is it possible to get a step-by-step on what to do to train an existing model and integrate it into to Presidio analyzer?

• 1 - Generate data set using presidio research
  https://github.com/microsoft/presidio-research/blob/master/notebooks/1_Generate_data.ipynb

This helps to generate a dataset that will train an existing model to perform better based on the dataset expectation.

• 2 - Used the generated dataset to generate spacy train and dev data
  https://github.com/microsoft/presidio-research/blob/master/notebooks/models/Create%20datasets%20for%20Spacy%20training.ipynb

This will generate two files that will be used for the spacy train

• 3 - Run spacy train with train.spacy and dev.spacy
  https://spacy.io/usage/training

This will generate two output , one call model-best and one call model-last (most of the time the model-best will be the one that we one to use)

• 4 - serialize/desrialize
  what this is for, how to run it?

• 5 - integrate into presidio
  in my case how does it get the model name? is it with a path or need to include it? where do I need to move ? ....

sorry for this but I still don't see how I can do it

[omri374]

The question is more related to spaCy than to Presidio. Let me try to help although I might be wrong as I haven't done this in a while.

• Serialize: save the model to disk.
• Deserialize: Load the model from disk.

does this work?

import spacy

nlp  = spacy.load("PATH_TO_MODEL_BEST")

If yes, then you should use the same approach when creating your NlpEngine, as described before:

from presidio_analyzer import AnalyzerEngine
from presidio_analyzer.nlp_engine import SpacyNlpEngine
import spacy

# Create a class inheriting from SpacyNlpEngine
class LoadedSpacyNlpEngine(SpacyNlpEngine):

    def __init__(self, loaded_spacy_model):
        self.nlp = {"en": loaded_spacy_model}

# Load a model a-priori
nlp = spacy.load("PATH_TO_MODEL_BEST")

# Pass the loaded model to the new LoadedSpacyNlpEngine
loaded_nlp_engine = LoadedSpacyNlpEngine(loaded_spacy_model = nlp)

# Pass the engine to the analyzer
analyzer = AnalyzerEngine(nlp_engine = loaded_nlp_engine)

From here, you can continue with the evaluating Presidio notebook.