Models

class PREDICT.Models.BayesianModel(priors, input_data=None, predictColName='prediction', outcomeColName='outcome', dateCol='date', verbose=True, plot_idata=False, draws=100, tune=100, cores=12, chains=4, model_formula=None)

Bases: PREDICTModel

A class which uses Bayesian regression to update the coefficients of a logistic regression model.

Parameters:

  • priors (dict) – Dictionary of predictors (key) and their mean and stds of the coefficients (values) e.g. {“blood_pressure”: (2.193, 0.12)}, this must include “Intercept” as a dictionary key. If any of the prior keys are None then the prior coefficients are estimated using a logistic regression model.

  • input_data (pd.DataFrame) – The input data for creating the model to calculate initial priors if none are provided.

  • predictColName (str) – The name of the column in the dataframe containing the predictions (default=’prediction’).

  • outcomeColName (str) – The name of the column in the dataframe containing the outcomes (default=’outcome’).

  • dateCol (str) – The name of the column in the dataframe containing the dates (default=’date’).

  • verbose (bool) – Whether to print the priors and posteriors of the model (default=True).

  • plot_idata (bool) – Whether to plot the idata trace plot (default=False).

  • draws (int) – Number of draws to use in the Bayesian model (default=100).

  • tune (int) – Number of tuning steps to use in the Bayesian model (default=100).

  • cores (int) – Number of cores to use in the Bayesian model (default=12).

  • chains (int) – Number of chains to use in the Bayesian model (default=4).

  • model_formula (str) – The formula to use in the Bayesian model (default=None, if None then a standard linear model formula is used without interactions).

Raises:

  • ValueError – If the priors are not in a dictionary format.

  • ValueError – If the required ‘Intercept’ is missing from the priors.keys().

  • ValueError – If priors are not provided and input_data is None.

Example

# Create a Bayesian model which refits and gives new coefficients and predictions when triggered. # Full example can be found in Examples/bayesian_example.ipynb model = BayesianModel(input_data=df, priors={“Intercept”: (0.34, 0.1), “age”: (1.56, 0.4), “systolic_bp”: (5.34, 0.2)}) model.trigger = BayesianRefitTrigger(model=model, input_data=df, refitFrequency=1)

get_coefs()

Return the current coefficients of the model for the loghook.

Returns:

Current priors of the model.

Return type:

dict

predict(input_data)

Makes predictions based on the input data.

Parameters:

input_data (any) – The input data for making predictions.

update(input_data)

Updates the model if required based on the input data.

Parameters:

input_data (any) – The input data for updating the model.

class PREDICT.Models.EvaluatePredictions(colName='prediction')

Bases: PREDICTModel

A class used to evaluate the predictions arising from another model which are already in the dataframe.

Parameters:

colName (str) – The name of the column in the dataframe containing the predictions (default=’prediction’).

predict(input_data)

Makes predictions based on the input data.

Parameters:

input_data (any) – The input data for making predictions.

class PREDICT.Models.PREDICTModel

Bases: object

A class used to represent the PREDICT Model.

addPostPredictHook(hook)

Adds a hook to be executed after making predictions.

Parameters:

hook (callable) – A function to be executed after predictions.

addPrePredictHook(hook)

Adds a hook to be executed before making predictions.

Parameters:

hook (callable) – A function to be executed before predictions.

predict(input_data)

Makes predictions based on the input data.

Parameters:

input_data (any) – The input data for making predictions.

trigger(input_data)

Evaluates whether the model needs to be updated based on the input data.

Parameters:

input_data (any) – The input data to evaluate the model update.

Returns:

Returns False indicating no update is required.

Return type:

bool

update(input_data)

Updates the model if required based on the input data.

Parameters:

input_data (any) – The input data for updating the model.

class PREDICT.Models.RecalibratePredictions(predictColName='prediction', outcomeColName='outcome', dateCol='date')

Bases: PREDICTModel

A class which recalibrates the predictions arising from another model based on the trigger function.

Recalibration involves using a logistic regression to adjust the model predictions.

Needs to be followed by setting a trigger function (see example).

Parameters:

  • predictColName (str) – The name of the column in the dataframe containing the predictions (default=’prediction’).

  • outcomeColName (str) – The name of the column in the dataframe containing the outcomes (default=’outcome’).

  • dateCol (str) – The name of the column in the dataframe containing the dates (default=’date’).

Examples

# Create a model which recalibrates predictions when triggered # Full example can be found in Examples/recalibration_example.ipynb model = RecalibratePredictions() model.trigger = AccuracyThreshold(model=model, threshold=0.7)

CalculateControlLimits(input_data, startCLDate, endCLDate, warningCL, recalCL, warningSDs, recalSDs)

Calculate the static control limits of data using either the specific period (startCLDate to endCLDate), the inputted control limits (warningCL, recalCL), or the first X months (nummMonths) since the start of the data.

Parameters:

  • input_data (pd.DataFrame) – The input data for updating the model.

  • startCLDate (str) – Start date to determine control limits from. Defaults to None.

  • endCLDate (str) – End date to determine control limits from. Defaults to None.

  • warningCL (float) – A manually set control limit for the warning control limit.

  • recalCL (float) – A manually set control limit for the recalibration trigger limit.

  • warningSDs (int or float) – Number of standard deviations from the mean to set the warning limit to. Defaults to 2.

  • recalSDs (int or float) – Number of standard deviations from the mean to set the recalibration trigger to. Defaults to 3.

Returns:

Two upper control limits for the warning and danger/recalibration trigger zones.

Return type:

float, float

predict(input_data)

Makes predictions based on the input data.

Parameters:

input_data (any) – The input data for making predictions.

update(input_data)

Updates the model if required based on the input data.

Parameters:

input_data (any) – The input data for updating the model.