Telekom NLU Bridge

Development • Documentation • Support • Contribute • Licensing

The goal of this project is to provide a unified API to several popular intent recognition applications.

About this component

Installation

The core package including NLUdataset and Baseline vendors can be installed for Python>=3.8 using pip

pip install nlubridge

Note that some vendors come with restrictions regarding the Python version, e.g. Rasa3 requires Python\<3.11.

To include optional dependencies for the vendors, e.g. Watson Assistant, type

pip install nlubridge[watson]

Following install options are available:

• watson
• fasttext
• luis
• rasa2
• rasa3
• spacy
• huggingface

Development tools can be installed with option develop.

Some vendors require access credentials like API tokens, URLs etc. These can be passed on construction of the objects. Alternatively, such arguments can be passed as environment variables, where the vendor will look for variables named variable VENDORNAME_PARAM_NAME.

Some vendors require additional dependencies. E.g., Spacy requires a model that can be downloaded (for the model de_core_news_sm) with

python -m spacy download de_core_news_sm

Migration from v0

With realease 1.0.0 we introduce a couple of changes to the names of files and vendor classes(see also https://github.com/telekom/nlu-bridge/issues/18).

Most notably:

• datasets.NLUdataset -> nlu_dataset.NluDataset
• vendors.vendors.Vendor -> - vendors.vendor.Vendor
• new supackage dataloaders that holds all functions for loading data into an NluDataset
• new function nlu_dataset.concat to concatenate NluDatasets passed in a list
• can load dataloaders, NluDataset, Vendor, OUT_OF_SCOPE_TOKEN, EntityKeys, concat, directly from nlubridge like from nlubridge import Vendor
• Load vendors like from nlubridge.vendors import Rasa3
• former TelekomModel now called CharNgramIntentClassifier
• Some vendor names changed for clarity and consistency (see "List of supported vendors" for the new names)

Usage

Here is an example for the TfidfIntentClassifier:

import os

import pandas as pd

from nlubridge.vendors import TfidfIntentClassifier
from nlubridge import NluDataset

dataset = NluDataset(texts, intents)
dataset = dataset.shuffle()

classifier = TfidfIntentClassifier()

train, test = dataset.train_test_split(test_size=0.25, random_state=0)
classifier = classifier.train_intent(train)
predicted = classifier.test_intent(test)
res = pd.DataFrame(list(zip(test.intents, predicted)), columns=['true', 'predicted'])

If you need to configure stratification, use the stratification parameter (defaults to "intents" and uses the intents in the dataset as stratification basis; whatever else you pass along has to conform to sklearn.model_selection.train_test_split(stratify=):

train, test = dataset.train_test_split(test_size=0.25, random_state=0, stratification=None)    # deactivate stratification (sklearn default for train_test_split)

To compare your own vendor or algorithm to existing vendors in this package, you can write a Vendor Subclass for your vendor, and possibly a dataloader function. Feel free to share your implementation using this repo. Similarly, fixes and extensions for the existing vendors are always welcome.

Logging

Most of the code uses python logging to report its progress. To get logs printed out to console or Jupyter notebook, a logger needs to be configured, before the nlutests code. Usually, log messages are on INFO level. This can be configured like this:

import logging

logger = logging.getLogger()
logger.setLevel(logging.INFO)
logger.addHandler(logging.StreamHandler())

Concepts / Architecture

• Vendors\ The vendors subpackage implements standardized interfaces to the specific vendors. A specific Vendor instance is in charge of dealing with converting the data to the required format, uploading data to the cloud if applicable, training models and making predictions.

• Datasets\ The nlu_dataset module provides a standard interface to NLU data. Data stored in different vendor's custom format can be loaded as a dataset and provided to any different vendor.

• Data Loaders\ The dataloaders subpackage provides functions to load data that are in a vendor-specific format as NluDataset.

List of supported vendors

Features

• Abstract class for Vendors with convenience methods (ex: scoring and scikit-learn compatibility)
• Abstract class for datasets with convenience methods (ex: train_test_split, indexing, iteration)
• Rate limiting to comply with cloud providers requirements

Development

TBD

Build

TBD

Code of Conduct

This project has adopted the Contributor Covenant in version 2.0 as our code of conduct. Please see the details in our CODE_OF_CONDUCT.md. All contributors must abide by the code of conduct.

Working Language

We decided to apply English as the primary project language.

Consequently, all content will be made available primarily in English. We also ask all interested people to use English as language to create issues, in their code (comments, documentation etc.) and when you send requests to us. The application itself and all end-user facing content will be made available in other languages as needed.

Documentation

The full documentation for the telekom nlu-bridge can be found in TBD

Support and Feedback

The following channels are available for discussions, feedback, and support requests:

Type	Channel
Issues
Other Requests

How to Contribute

Contribution and feedback is encouraged and always welcome. For more information about how to contribute, the project structure, as well as additional contribution information, see our Contribution Guidelines. By participating in this project, you agree to abide by its Code of Conduct at all times.

Licensing

Copyright (c) 2021 Deutsche Telekom AG.

Licensed under the MIT License (the "License"); you may not use this file except in compliance with the License.

You may obtain a copy of the License by reviewing the file LICENSE in the repository.

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the LICENSE for the specific language governing permissions and limitations under the License.