processpredictR

Description

Means to predict process flow, such as process outcome, next activity, next time, remaining time, and remaining trace. Off-the-shelf predictive models based on the concept of Transformers are provided, as well as multiple ways to customize the models. This package is partly based on work described in Zaharah A. Bukhsh, Aaqib Saeed, & Remco M. Dijkman. (2021). "ProcessTransformer: Predictive Business Process Monitoring with Transformer Network" arXiv:2104.00721.

Author(s)

Maintainer: Gert Janssenswillen gert.janssenswillen@uhasselt.be

Authors:

• Ivan Esin ivan.esin@student.uhasselt

Other contributors:

• Hasselt University [copyright holder]

Pipe operator

Description

See magrittr::%>% for details.

Usage

lhs %>% rhs

Arguments

Value

The result of calling rhs(lhs).

Confusion matrix for predictions

Description

Confusion matrix for predictions

Usage

confusion_matrix(predictions, ...)

Arguments

Value

A table object that can be used for plotting a confusion matrix using plot().

Define transformer model

Description

Defines the model using the keras functional API. The following 4 process monitoring tasks are defined:

• outcome

• next_activity

• next_time

• remaining_time

• remaining_trace

• remaining_trace_s2s

Usage

create_model(
  x_train,
  custom = FALSE,
  num_heads = 4,
  output_dim_emb = 36,
  dim_ff = 64,
  ...
)

Arguments

Value

An object of class ppred_model and list containing a Transformer model (returned by keras::keras_model()) and some additional useful metrics.

Create a vocabulary

Description

Creates a vocabulary of activities and outcome labels.

Usage

create_vocabulary(processed_df)

Arguments

Value

A list consisting of:

• "keys_x": list of activity labels

• "keys_y": list of outcome labels (none for tasks "next_time" and "remaining_time")

Utils

Description

Utils

Usage

get_vocabulary(examples)

Arguments

Calculate the maximum length of a case / number of activities in the longest trace in an event log

Description

Calculate the maximum length of a case / number of activities in the longest trace in an event log

Usage

max_case_length(processed_df)

Arguments

Value

An integer number of the maximum case length (longest trace) in an event log.

Examples

library(processpredictR)
library(eventdataR)

df <- prepare_examples(patients)
max_case_length(df)

Calculate number of outputs (target variables)

Description

Calculate number of outputs (target variables)

Usage

num_outputs(processed_df)

Arguments

Value

an integer number of outputs for supplying as an argument to a Transformer model, i.e. number of unique labels for a specific process monitoring task.

Examples

library(processpredictR)
library(eventdataR)
df <- prepare_examples(patients)
num_outputs(df)

Plot Methods

Description

Visualize metric

Usage

## S3 method for class 'ppred_predictions'
plot(x, ...)

Arguments

Value

A ggplot object, which can be customized further, if deemed necessary.

ppred_examples_df object

Description

object of type ppred_examples_df is a transformed event log returned by prepare_examples_dt().

ppred_model object

Description

object of type ppred_model is a list returned by processpredictR::create_model() containing a custom keras functional (transformer) model and some other useful metrics of an event log.

ppred_predictions object

Description

object of type ppred_predictions is a data.frame with predicted values returned by predict.ppred_model().

Convert a dataset of type log into a preprocessed format.

Description

an event log is converted into a tibble where each row contains a cumulative sequence of activities per case. This sequence will eventually be feeded to the Transformer model's token embedding layer.

Usage

prepare_examples(
  log,
  task = c("outcome", "next_activity", "next_time", "remaining_time", "remaining_trace",
    "remaining_trace_s2s"),
  features = NULL,
  ...
)

Arguments

Value

a preprocessed dataset of class ppred_examples_df.

Examples

library(processpredictR)
library(eventdataR)

prepare_examples(patients, "next_activity")

Print methods

Description

Print methods

Usage

## S3 method for class 'ppred_model'
print(x, ...)

Arguments

Value

prints a Transformer model from a list returned by create_model().

Default compile function for ProcessTransformer model

Description

These objects are imported from other packages. Follow the links below to see their documentation.

keras

compile, evaluate, fit

stats

predict

Arguments

See Also

See keras::fit() for documentation of parameters

Splits the preprocessed data.frame.

Description

Returns train- and test dataframes as a list.

Usage

split_train_test(processed_df, split = 0.7)

Arguments

Value

A list containing the train- and the test set objects.

Examples

library(processpredictR)
library(eventdataR)

df <- prepare_examples(patients, "next_activity")
split_train_test(df, split = 0.8)

Stacks a keras layer on top of existing model

Description

User friendly interface to add a keras layer on top of existing model.

Usage

stack_layers(object, ...)

Arguments

Value

a list containing an adapted Transformer model.

Tokenize features and target of a processed dataset of class ppred_examples_df

Description

Tokenize features and target of a processed ppred_examples_df object to fit the Transformer model.

Usage

tokenize(processed_df)

Arguments

Value

A list of (sequence) tokens and additional numeric or categorical features.

Calculate the vocabulary size, i.e. the sum of number of activities, outcome labels and padding keys

Description

Calculate the vocabulary size, i.e. the sum of number of activities, outcome labels and padding keys

Usage

vocab_size(processed_df)

Arguments

Value

an integer number of vocabulary size to define the Transformer model.

Examples

library(processpredictR)
library(eventdataR)
df <- prepare_examples(patients)
vocab_size(df)