Twinn-ml-interface

Twinn-ml-interface is a Python package for data contracts between machine learning code and infrastructure.

Author: Royal HaskoningDHV

Installation

The easiest way to install is package is using pip:

pip install twinn-ml-interface

Model Interface

Properties

model_type_name

An unique name for each model class.

model_category

Whether the model outputs anomalies, predictions or actuals. This determines the format in which the results are expected.

performance_value

The model is expected to calculate some metric value after training to indicate the model performance.

train_data_config

N/A for most cases.

target_tag

For most cases, return a UnitTagLiteral containing the unit code and tagname of the target of the model.

data_config

For most cases, return a list of UnitTagLiterals containing all the data that is used for training the model. The training window can be passed at runtime.

result_tag

For most cases, return a UnitTagLiteral of the result unit and tag of the model.

unit_properties_template

N/A for most cases.

unit_hierarchy_template

N/A for most cases.

train_window_finder_config_template

N/A for most cases.

Training

For training, the functions of the model interface are called in the following order:

1. initialize
2. preprocess
3. validate_input_data
4. train
5. dump

Prediction

For prediction, the functions of the model interface are called in the following order:

1. load
2. predict