huggingface-sentiment-analysis-to-mlflow

This repo contains a python script that can be used to log the huggingface sentiment-analysis task as a model in MLflow. It can then be registered and available for use by the rest of the MLflow users.

1. Running this script to load the model into MLflow

Ensure that MLFLOW_TRACKING_URI is set correctly in your environment. For example, if using InfinStor's MLflow service, MLFLOW_TRACKING_URI must be set to infinstor://infinstor.com/

The conda environment that you are working in must have pytorch, mlflow and transformers installed.

$ python ./log_model.py

2. Serving the model locally:

Note the id of the run that was generated from the run of log_model.py in step (1) above. Suppose it is 0-351e1a1e91334d9ca5cb704b0792d9b3, then the following command serves that model

$ mlflow models serve -m runs:/0-351e1a1e91334d9ca5cb704b0792d9b3/model

3. Input:

Input is a pandas dataframe with a column titled 'text' that has the text for which we are estimating sentiment. This is supplied to the inference process that was started in step (2) above in the pandas-split format. The following is an example of such input text:

{"columns":["text"],"data":[["This is meh weather"], ["This is great wuther"]]}

4. Example run of inference:

curl -X POST -H "Content-Type:application/json; format=pandas-split" --data '{"columns":["text"],"data":[["This is meh weather"], ["This is great wuther"]]}' http://127.0.0.1:5000/invocations

5. Example output

[{"text": "This is meh weather", "label": "NEGATIVE", "score": 0.753343403339386}, {"text": "This is great wuther", "label": "POSITIVE", "score": 0.9984261989593506}]

As evident, output is a pandas datafrom with three columns - 'text', 'label' and 'score', where 'text' is the same text that was passed in as input, 'label' is the classification (POSITIVE or NEGATIVE) and 'score' is the confidence score.