easierai-common-functions

Common Python Helper Functions

This library contains reused code for all the EASIER-AI projects written in Python.

Using the library

Install

This library is available through PIP package manager. To install it, execute

pip install easierai-common-functions

Importing

The library needs to be imported in order to use it:

import common_functions.helpers as helpers

from minio import Minio

from common_functions.logger import Logger

Then, there needs to be some configuration:

helpers.config = helpers.read_config_file(config_file_path)

helpers.minioClient = Minio(minio_host + ':' + minio_port, minio_access, minio_secret, secure=False)

helpers._logger = Logger('helpers', 'helpers.py')

If you wish to check the validity of the configuration provided for an inferencer, you can use this method (it will notify at start if there are no valid models):

helpers.check_initial_config(eslib, False)

Where eslib is a valid started instance of the elasticsearch library.

Necessary environmental variables

This library reads from the following environmental variables:

• LOGSTASH_HOST: IP/hostname hosting the Logstash service to upload the logs
• LOGSTASH_PORT: port of where the Logstash service is listening

Usage

The library has these functions available:

get_data_shape(data_type, num_features, num_samples, algorithm)

Outputs the data_shape required according to the parameters passed.

importer(algorithm, inference_type=constants.ESTIMATOR, lr=0.001)

Returns a predictor instance according to the parameters passed.

load_model_file(eslib, id, inference_type=constants.ESTIMATOR)

Returns the document stored on Elasticsearch, which includes, among other things, the h5 and pkl files where we had saved the model and the scalers, respectively. :param id: id of the entity :param inference_type: this param helps us to download the right model according to its features and parameters :return: dict with the format {"extension": object, "extension2": object2}

read_config_file(config_path)

Initializes the config variable

check_initial_config(eslib, is_classifier)

Checks if there is any model in the database that matches the configuration provided.

scale_dataset(scaler, data, i, ft_range=(-1, 1), training=True)

Scale data (in np.array or list format) using MinMaxScaler and the ft_range given (default is (-1,1) :param data: array of data to be scaled :param i: feature corresponding to the scaler :param ft_range: tuple containing minimum and maximum values of the data already scaled :return: tuple of scaler and data scaled

compose_model_params(is_classifier)

Composes a json object with the parameters of a trained model to store in the database

compose_model_params_filter(is_classifier)

Composes a json object with the parameters in the config file used to look for the models in the database.

save_model(eslib, id, metadata, dict={}, inference_type=constants.ESTIMATOR, _id=None, save_tflite=False, calibration_data=None)

Saves the model related files after training. Has the ability to save a model as tflite format. The parameter dict should come in the format {"extension1": object1, "extension2": object2 ... }. The calibration_data is only used when saving a model as tflite format, and should be a representation of the dataset.

Additional features

Constants file

The constants used on EASIER-AI services are stored in common_functions/constants.py file. It can be imported as:

import common_functions.constants as constants

Advanced logger

This logger has the same syntax as the default logging python library. It needs to be imported and initialized as:

from common_functions.logger import Logger

logger = Logger(service_name, filename)

It then can be used as logger.info(message), logger.debug(message, additional_info), etc.

This logger, apart from printing to console, uploads each log instance to Elasticsearch via Logstash, through a TCP port. To use this functionality it is needed to define the previously mentioned LOGSTASH_HOST and LOGSTASH_PORT environment variables.

Edge toolkit

This class is in charge of converting a tensorflow or keras model into tensorflow lite. It can be used as:

from edge_tools import Edge_Toolkit

edge_toolkit = Edge_Toolkit(logger)

edge_toolkit.convert_model_lite(calibration_data=calibration_data, keras_model_path=filename + '.' + constants.MODEL_EXTENSION)

After executing these lines, the tflite file will be stored in ../storage/ and can be uploaded to a remote filesystem.

Model definitions

The model definitions used by EASIER are also stored in this library. They are imported by the helpers file using the importer( ... ) function.

Copyright

Copyright (C) 2020 ATOS Spain All Rights Reserved.