deepparse 0.9.9

Here is Deepparse.

Deepparse is a state-of-the-art library for parsing multinational street addresses using deep learning.

Use deepparse to

• parse multinational address using one of our pretrained models with or without attention mechanism,
• parse addresses directly from the command line without code to write,
• parse addresses with our out-of-the-box FastAPI parser,
• retrain our pretrained models on new data to improve parsing on specific country address patterns,
• retrain our pretrained models with new prediction tags easily,
• retrain our pretrained models with or without freezing some layers,
• train a new Seq2Seq addresses parsing models easily using a new model configuration.

Read the documentation at deepparse.org.

Deepparse is compatible with the latest version of PyTorch and Python >= 3.8.

Countries and Results

We evaluate our models on two forms of address data

• clean data which refers to addresses containing elements from four categories, namely a street name, a municipality, a province and a postal code,
• incomplete data which is made up of addresses missing at least one category amongst the aforementioned ones.

You can get our dataset here.

Clean Data

The following table presents the accuracy (using clean data) on the 20 countries we used during training for both our models. Attention mechanisms improve performance by around 0.5% for all countries.

We have also made a zero-shot evaluation of our models using clean data from 41 other countries; the results are shown in the next table.

Moreover, we also tested the performance when using attention mechanism to further improve zero-shot performance on those countries; the result are shown in the next table.

Incomplete Data

The following table presents the accuracy on the 20 countries we used during training for both our models but for incomplete data. We didn't test on the other 41 countries since we did not train on them and therefore do not expect to achieve an interesting performance. Attention mechanisms improve performance by around 0.5% for all countries.

Getting Started:

from deepparse.parser import AddressParser
from deepparse.dataset_container import CSVDatasetContainer

address_parser = AddressParser(model_type="bpemb", device=0)

# you can parse one address
parsed_address = address_parser("350 rue des Lilas Ouest Québec Québec G1L 1B6")

# or multiple addresses
parsed_address = address_parser(
    [
        "350 rue des Lilas Ouest Québec Québec G1L 1B6",
        "350 rue des Lilas Ouest Québec Québec G1L 1B6",
    ]
)

# or multinational addresses
# Canada, US, Germany, UK and South Korea
parsed_address = address_parser(
    [
        "350 rue des Lilas Ouest Québec Québec G1L 1B6",
        "777 Brockton Avenue, Abington MA 2351",
        "Ansgarstr. 4, Wallenhorst, 49134",
        "221 B Baker Street",
        "서울특별시 종로구 사직로3길 23",
    ]
)

# you can also get the probability of the predicted tags
parsed_address = address_parser(
    "350 rue des Lilas Ouest Québec Québec G1L 1B6", with_prob=True
)

# Print the parsed address
print(parsed_address)

# or using one of our dataset container
addresses_to_parse = CSVDatasetContainer(
    "./a_path.csv", column_names=["address_column_name"], is_training_container=False
)
address_parser(addresses_to_parse)

The default predictions tags are the following

• "StreetNumber": for the street number,
• "StreetName": for the name of the street,
• "Unit": for the unit (such as apartment),
• "Municipality": for the municipality,
• "Province": for the province or local region,
• "PostalCode": for the postal code,
• "Orientation": for the street orientation (e.g. west, east),
• "GeneralDelivery": for other delivery information.

Parse Addresses From the Command Line

You can also use our cli to parse addresses using:

parse <parsing_model> <dataset_path> <export_file_name>

Parse Addresses Using Your Own Retrained Model

See here for a complete example.

address_parser = AddressParser(
    model_type="bpemb",
    device=0,
    path_to_retrained_model="path/to/retrained/bpemb/model.p",
)

address_parser("350 rue des Lilas Ouest Québec Québec G1L 1B6")

Parse Address With Our Out-Of-The-Box FastAPI Parse Model

You can use Out-Of-The-Box RESTAPI to parse addresses:

Installation:

First, ensure that you have Docker Engine and Docker Compose installed on your machine. if not, you can install them using the following documentations in the following order:

1. Docker Engine
2. Docker Compose

Once you have Docker Engine and Docker Compose installed, you can run the following command to start the FastAPI application:

docker compose up app

Sentry:

Also, you can monitor your application usage with Sentry by setting the environment variable SENTRY_DSN to your Sentry's project DSN. There is an example of the .env file in the project's root named .env_example. You can copy it using the following command:

cp .env_example .env

#### Request Examples:

Once the application is up and running and the port 8000 is exported on your localhost, you can send request with one
of the following methods:

##### cURL POST request:
```shell
curl -X POST --location "http://127.0.0.1:8000/parse/bpemb-attention" --http1.1 \
    -H "Host: 127.0.0.1:8000" \
    -H "Content-Type: application/json" \
    -d "[
          {\"raw\": \"350 rue des Lilas Ouest Quebec city Quebec G1L 1B6\"},
          {\"raw\": \"2325 Rue de l'Université, Québec, QC G1V 0A6\"}
        ]"

Python POST request:

import requests

url = 'http://localhost:8000/parse/bpemb'
addresses = [
    {"raw": "350 rue des Lilas Ouest Quebec city Quebec G1L 1B6"},
    {"raw": "2325 Rue de l'Université, Québec, QC G1V 0A6"}
    ]

response = requests.post(url, json=addresses)
parsed_addresses = response.json()
print(parsed_addresses)

Retrain a Model

See here for a complete example using Pickle and here for a complete example using CSV.

# We will retrain the fasttext version of our pretrained model.
address_parser = AddressParser(model_type="fasttext", device=0)

address_parser.retrain(training_container, train_ratio=0.8, epochs=5, batch_size=8)

One can also freeze some layers to speed up the training using the layers_to_freeze parameter.

address_parser.retrain(
    training_container,
    train_ratio=0.8,
    epochs=5,
    batch_size=8,
    layers_to_freeze="seq2seq",
)

Or you can also give a specific name to the retrained model. This name will be use as the model name (for print and class name) when reloading it.

address_parser.retrain(
    training_container,
    train_ratio=0.8,
    epochs=5,
    batch_size=8,
    name_of_the_retrain_parser="MyNewParser",
)

Retrain a Model With an Attention Mechanism

See here for a complete example.

# We will retrain the fasttext version of our pretrained model.
address_parser = AddressParser(
    model_type="fasttext", device=0, attention_mechanism=True
)

address_parser.retrain(training_container, train_ratio=0.8, epochs=5, batch_size=8)

Retrain a Model With New Tags

See here for a complete example.

address_components = {"ATag": 0, "AnotherTag": 1, "EOS": 2}
address_parser.retrain(
    training_container,
    train_ratio=0.8,
    epochs=1,
    batch_size=128,
    prediction_tags=address_components,
)

Retrain a Seq2Seq Model From Scratch

See here for a complete example.

seq2seq_params = {"encoder_hidden_size": 512, "decoder_hidden_size": 512}
address_parser.retrain(
    training_container,
    train_ratio=0.8,
    epochs=1,
    batch_size=128,
    seq2seq_params=seq2seq_params,
)

Download Our Models

Here are the URLs to download our pretrained models directly

• FastText,
• FastTextAttention,
• BPEmb,
• BPEmbAttention,
• FastText Light ( using Magnitude Light).

Or you can use our cli to download our pretrained models directly using:

download_model <model_name>

Starting at version 0.9.8, we will also release the weights with the GitHub release note available here.

Installation

Before installing deepparse, you must have the latest version of PyTorch in your environment.

• Install the stable version of deepparse:

pip install deepparse

• Install the stable version of deepparse with the app extra dependencies:

pip install deepparse[app]  # for bash terminal
pip install 'deepparse[app]' # for ZSH terminal

• Install the latest development version of deepparse:

pip install -U git+https://github.com/GRAAL-Research/deepparse.git@dev

Cite

Use the following for the article;

@misc{yassine2020leveraging,
    title={{Leveraging Subword Embeddings for Multinational Address Parsing}},
    author={Marouane Yassine and David Beauchemin and François Laviolette and Luc Lamontagne},
    year={2020},
    eprint={2006.16152},
    archivePrefix={arXiv}
}

and this one for the package;

@misc{deepparse,
    author = {Marouane Yassine and David Beauchemin},
    title  = {{Deepparse: A State-Of-The-Art Deep Learning Multinational Addresses Parser}},
    year   = {2020},
    note   = {\url{https://deepparse.org}}
}

Contributing to Deepparse

We welcome user input, whether it is regarding bugs found in the library or feature propositions ! Make sure to have a look at our contributing guidelines for more details on this matter.

License

Deepparse is LGPLv3 licensed, as found in the LICENSE file.