Zeroshot Classifier

The Zeroshot classifier is a tool allowing to classify a document according to classes defined in configuration file This tools is a rest service. Tokenization depends on annotation model created by the tool stored in tkeir/thot/tasks/tokenizer/createAnnotationResouces.py This tools allows to create typed compound word list.

Zeroshot Classifier API

Zeroshot Classifier configuration

Example of Configuration:

zeroshotclassifier.json

{
    "logger": {
        "logging-level": "{{ project.loglevel }}"
    },
    "zeroshot-classification": {
        "settings":{
            "language":"en",
            "use-cuda":false,
            "cuda-device":0,
            "model-path-or-name":"{{ project.path }}/resources/modeling/net"
        },        
        "classes":[
          {"label":"machine learning", "content":["machine learning","machine learning model", "bias metric","feature selection","data science"]},
          {"label":"deep learning", "content":["deep learning","hidden layer","loss function", "neural network", "convolutional neural network", "perceptron"]},
          {"label":"reinforcement learning", "content":["reinforcement learning","learning policy", "learning state", "advantage function","agent"]},
          {"label":"robotic", "content":["robotic","robot","control engineering","cybernetics","haptic"]},
          {"label":"computer vision", "content":["computer vision","images processing","anchor box","object detection","object segmentation"]},
          {"label":"natural language processing", "content":["natural language processing","tagger","semantic","named entities","classification"]},
          {"label":"internet of things", "content":["internet of things","devices","iot","embedded software","firmware","machine to machine"]},
          {"label":"neuromorphic computing", "content":["neuromorphic computing","bio-inspired","personhood","human brain","biological computation"]}
        ],
        "re-labelling-strategy":"max",
        "network": {
            "host":"0.0.0.0",
            "port":10010,
            "associate-environment": {
                "host":"CLASSIFICATION_HOST",
                "port":"CLASSIFICATION_PORT"
            }
        },
        "runtime":{
            "request-max-size":100000000,
            "request-buffer-queue-size":100,
            "keep-alive":true,
            "keep-alive-timeout":500,
            "graceful-shutown-timeout":15.0,
            "request-timeout":600,
            "response-timeout":600,
            "workers":1
        }
    }
}

Zeroshot Classifier is an aggreation of network configuration, serialize configuration, runtime configuration (in field converter), logger (at top level). The zeroshot-classification configuration is a table containing classes configuration:

settings/language: define classifier language [en | fr]
settings/use-cuda: use cuda device
settings/cuda-device: device number (multiple graphic cards)
[classes]/label: label of classes
[classes]/content: possible value for the classe (can view view as synonyms or sublasses)
re-labelling-strategy : * sum : master class is the sum of the scores of subclasses (synonyms) * mean : master class is the mean of the scores of subclasses (synonyms) * max : master class is the max of the scores of subclasses (synonyms)

Configure classifier logger

Logger is configuration at top level of json in logger field.

Example of Configuration:

logger configuration

{
    "logger": {
        "logging-level": "debug"
    }    
}

The logger fields is:

logging-level

It can be set to the following values:

debug for the debug level and developper information
info for the level of information
warning to display only warning and errors
error to display only error
critical to display only error

Configure classifier Network

Example of Configuration:

network configuration

{
    "network": {
        "host":"0.0.0.0",
        "port":8080,
        "associate-environment": {
            "host":"HOST_ENVNAME",
            "port":"PORT_ENVNAME"
        },
        "ssl":
        {
            "certificate":"path/to/certificate",
            "key":"path/to/key"
        }
    }
}

The network fields:

host : hostname
port : port of the service
associated-environement : is the "host" and "port" associated environment variables that allows to replace the default one. This field is not mandatory.
"host" : associated "host" environment variable
"port" : associated "port" environment variable
ssl : ssl configuration IN PRODUCTION IT IS MANDATORY TO USE CERTIFICATE AND KEY THAT ARE *NOT* SELF SIGNED
cert : certificate file
key : key file

Configure classifier runtime

Example of Configuration:

network configuration

{
    "runtime":{
        "request-max-size":100000000,
        "request-buffer-queue-size":100,
        "keep-alive":true,
        "keep-alive-timeout":5,
        "graceful-shutown-timeout":15.0,
        "request-timeout":60,
        "response-timeout":60,
        "workers":1
    }    
}

The Runtime fields:

request-max-size : how big a request may be (bytes)
request-buffer-queue-size: request streaming buffer queue size
request-timeout : how long a request can take to arrive (sec)
response-timeout : how long a response can take to process (sec)
keep-alive: keep-alive
keep-alive-timeout: how long to hold a TCP connection open (sec)
graceful-shutdown_timeout : how long to wait to force close non-idle connection (sec)
workers : number of workers for the service on a node
associated-environement : if one of previous field is on the associated environment variables that allows to replace the default one. This field is not mandatory.
request-max-size : overwrite with environement variable
request-buffer-queue-size: overwrite with environement variable
request-timeout : overwrite with environement variable
response-timeout : overwrite with environement variable
keep-alive: overwrite with environement variable
keep-alive-timeout: overwrite with environement variable
graceful-shutdown_timeout : overwrite with environement variable
workers : overwrite with environement variable

Zeroshot Classifier service

To run the command type simply from tkeir directory:

python3 thot/zeroshotclassifier_svc.py --config=<path to configuration file>

or if you install tkeir wheel:

python3 tkeir-zeroshotclassifier-svc --config=<path to configuration file>

A light client can be run through the command

python3 thot/zeroshotclassifier_client.py --config=<path to configuration file> --input=<input directory> --output=<output directory>

or if you install tkeir wheel:

tkeir-zeroshotclassifier-client.py --config=<path to configuration file> --input=<input directory> --output=<output directory>

Zeroshot Classifier Tests

The converter service come with unit and functional testing.

Zeroshot Classifier Unit tests

Unittest allows to test Zeroshot Classifier classes only.

python3 -m unittest thot/tests/unittests/TestZeroshotClassifierConfiguration.py
python3 -m unittest thot/tests/unittests/TestZeroshotClassifier.py

Zeroshot Classifier Functional tests

python3 -m unittest thot/tests/functional_tests/TestZeroshotClassifierSvc.py