Provides a unified API to several popular intent recognition applications
Project description
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 calledCharNgramIntentClassifier
- 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
Thevendors
subpackage implements standardized interfaces to the specific vendors. A specificVendor
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
Thenlu_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
Thedataloaders
subpackage provides functions to load data that are in a vendor-specific format as NluDataset.
List of supported vendors
Vendor Class | Status | Intents | Entities | Algorithm |
---|---|---|---|---|
TfidfIntentClassifier | ✓ | ✓ | ✗ | TFIDF on words + SVM |
FastText | ✓ | ✓ | ✗ | fasttext |
Spacy | ✓ | ✓ | ✗ | BoW linear + CNN |
WatsonAssistant | ✓ | ✓ | ✗ | Propietary (probably LR) |
Luis | needs testing | ✓ | ✗ | Propietary (probably LR) |
CharNgramIntentClassifier | ✓ | ✓ | ✗ | tf-idf on char n-grams + SGD |
Rasa2 | ✓ | ✓ | ✓ | configurable |
Rasa3 | ✓ | ✓ | ✓ | configurable |
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.
Project details
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for nlubridge-1.0.2-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | cdf4d29728f005d0f0a0fca89f777de9e18853ce5c198bd8a76def7cdeaa3dba |
|
MD5 | 69e0a99681852d757da386dde7745685 |
|
BLAKE2b-256 | 721e81719ba2af53d2572b184506dc8ef9a65057e63d0f577cb12871d3f98cba |