API Reference

This section provides detailed documentation for all modules and functions in the pyoxynet package.

Utilities Module

The utilities module contains the core functionality for data processing, model loading, and inference.

utilities.PrintHello(hello='hello')

This function prints to screen.

Parameters:

name (str) – The name to use.

Returns:

none

utilities.compute_shap(explainer, df, n_inputs=6, past_points=40, shap_stride=20)

Computes SHAP values

Parameters:

• explainer (explainer SHAP) – deep explainer that explains the deep learning model

• data (pd df) – data for the local explanation

Returns:

SHAP values

Return type:

shap (dict)

utilities.create_probabilities(duration=600, VT1=320, VT2=460, training=True, normalization=False, resting=True, resting_duration=60, initial_step=False, y_pm0=1, generator=False, sigmoid_steepness=0.05)

Creates piecewise-linear probabilities for moderate/heavy/severe zones with enforced overlaps at VT1 and VT2.

Behaviour:

• Nodes at t=1, VT1, VT2, duration with values:

  t=1 : (y_pm0, 1-y_pm0, 0) VT1 : (0.5, 0.5, 0) VT2 : (0.0, 0.5, 0.5) end : (0.0, 0.0, 1.0)

• Linear interpolation between these nodes ensures fully linear transitions, exact overlaps at VT1 and VT2, and sum-to-1 (after numerical normalization).

utilities.draw_real_test(resting='random')

Draw a single data file from the directory containing all the real files

Parameters:

none

Returns:

Real data output

Return type:

df (pandas df)

utilities.generate_CPET(generator, plot=False, fitness_group=None, noise_factor=None, resting=False, training=True, normalization=False)

Generate synthetic CPET data using the TensorFlow generator model.

NOTE: This function requires a loaded TensorFlow generator model (not TFLite). Use load_tf_generator() to load the model first. Requires full TensorFlow installation.

Parameters:

• generator – TensorFlow generator model (required). Use load_tf_generator() to load.

• plot (bool) – Whether to plot the generated data. Default False.

• fitness_group (int) – Fitness level: 0=very low, 1=low, 2=medium, 3=high. None=random.

• noise_factor (float) – Noise factor for white noise (1.5-2.5). None=random.

• resting (bool) – Whether to include resting phase. Default False.

• training (bool) – Training mode flag. Default True.

• normalization (bool) – Whether to normalize data. Default False.

Returns:

Pandas dataframe with generated CPET data data (dict): Metadata about the generated CPET test

Return type:

df (pd.DataFrame)

Raises:

• ValueError – If generator model is None or invalid

• ImportError – If required dependencies are missing

Note

The generator uses a hardcoded input shape of [1, 23] which consists of: - 20 dimensions for data noise - 3 dimensions for domain probabilities (moderate, heavy, severe)

utilities.get_sec(time_str)

This function converts strings to time.

Parameters:

time_str (str) – The time str.

Returns:

total seconds (int)

utilities.load_csv_data(csv_file='data_test.csv')

Loads data from csv file (returns test data if no arguments)

Parameters:

• n_inputs (int) – Number of input variables.

• past_points (int) – Number of past inputs in the time-series.

Returns:

Model output example

Return type:

df (pandas df)

utilities.load_exercise_threshold_app_data(data_dict={}, filter_size=20)

Loads data from data dict with format provided by https://www.exercisethresholds.com/

Parameters:

data_dict (dict) – Dictionary with format like test/exercise_threshold_app_test.json

Returns:

Pandas data frame with format that can be used by Pyoxynet for inference (columns needed: ‘VO2_I’, ‘VCO2_I’, ‘VE_I’, ‘PetO2_I’, ‘PetCO2_I’, ‘VEVO2_I’, ‘VEVCO2_I’)

Return type:

df (pandas df)

utilities.load_explainer(tf_model)

Loads a deep explainer from the shap library and automatically loads the data to train it

Parameters:

tf_model (Model class) – deep learning model that deep explainer needs to explain

Returns:

deep explainer SHAP

Return type:

explainer (deep)

utilities.load_tf_generator()

This function loads the saved TensorFlow generator model.

NOTE: This function requires full TensorFlow installation and will NOT work with tflite_runtime only. If you installed the lite version of pyoxynet, this generator functionality is not available.

Parameters:

None

Returns:

Handle on the TensorFlow generator model, or None if unavailable

Return type:

generator (TF model)

utilities.load_tf_model(n_inputs=5, past_points=40, model='CNN')

This function loads the saved tf models.

Parameters:

• n_inputs (int) – Number of input variables.

• past_points (int) – Number of past inputs in the time-series.

Returns:

handle on the TFLite interpreter

Return type:

interpreter (tflite interpreter)

utilities.normalize(df, min_target=-1, max_target=1)

Pandas df normalisation

Parameters:

df (pd df) – input df

Returns:

output df

Return type:

result (pd df)

utilities.optimal_filter(t, y, my_lambda)

A bad ass optimisation filter

Parameters:

• t – array Independent coord array

• y – array Dependent coord array

• my_lambda – float Smoothing factor

Returns:

array

Filtered variable

Return type:

x

utilities.random_walk(length=1, scale_factor=1, variation=1)

Random walk generator

Parameters:

• length (int) – Length of the output list

• scale_factor (float) – Scale factor to be applied to the whole output

• variation (float) – Local variation of the main signal with the random walk

Returns:

none

utilities.test_pyoxynet(tf_model=[], input_df=[], n_inputs=5, past_points=40, model='CNN', plot=False, inference_stride=1)

Runs the pyoxynet inference

Parameters:

• tf_model (TF model) – Model uploaded with load_tf_model

• n_inputs (int) – Number of inputs (deafult to Oxynet configuration)

• past_points (int) – Number of past points in the time series (deafult to Oxynet configuration)

• inference_stride (int) – Stride inference for NN - speed up computation

Returns:

Model output example

Return type:

x (array)

Testing Module

The testing module provides functions for testing and validating the package functionality.

Model Module

The model module contains model-related utilities and configurations.

class model.LSTMGRUModel(*args: Any, **kwargs: Any)

Bases: Model

call(inputs, training=None)

load_model(model_path)

class model.Model(*args: Any, **kwargs: Any)

Bases: Model

call(inputs, training=None)

loadModel(model_path)

class model.TCN(*args: Any, **kwargs: Any)

Bases: Model

build_tcn_block(num_filters, kernel_size, dilation_rate, dropout_rate)

call(inputs, training=None)

loadModel(model_path)

class model.generator(*args: Any, **kwargs: Any)

Bases: Model

call(inputs, training=None)