Discovery and predictive simulation of software-based processes
An embodiment may involve obtaining a log regarding execution of a software application; obtaining indications of availabilities of resources related to the software application; determining, from the log and the indications of availabilities of the resources, a time series of software application activities; and training a prediction engine with the time series of software application activities, wherein the prediction engine as trained is configured to receive an input time series of further software application activities and generate an output time series that predicts additional software application activities. Another embodiment may involve obtaining an input time series of software application activities, wherein the input time series is based on a log regarding execution of a software application and includes indications of availabilities of resources associated with the software applications; and generating, using a prediction engine, an output time series based on the input time series that predicts additional software application activities.
1 . A method comprising:
obtaining a log regarding execution of a software application on a computing system;
obtaining indications of availabilities of resources related to the software application;
determining, from the log and the indications of availabilities of the resources, a workflow graph representing a workflow of software application activities, wherein the workflow graph includes a plurality of workflow states, transitions between workflow states in the plurality of workflow states, and transition probabilities associated with each of the transitions, wherein each workflow state corresponds to a node of the workflow graph, wherein each transition between workflow states corresponds to an edge of the workflow graph;
determining, from the log, the workflow graph, the transition probabilities, and the indications of availabilities of the resources, a training time series of the software application activities;
training a prediction engine comprising a long short-term memory (LSTM) machine learning model stored in a memory with the training time series of the software application activities using an LSTM trainer, wherein the LSTM trainer produces the LSTM machine learning model with a preconfigured number of cells based on the training time series of the software application activities, wherein the prediction engine as trained is configured to receive an input time series of further software application activities and generate an output time series that predicts additional software application activities;
obtaining an altered input time series of the software application activities, wherein the altered input time series of the software application activities represents an alteration to at least one of the software application activities, the workflow, the transition probabilities, or the resources; and
generating, using the prediction engine, a predicted output time series that predicts additional software application activities based on the altered input time series of the software application activities.
2 . The method of claim 1 , wherein the log includes data representing state changes of the software application and timestamps of when the state changes occurred.
3 . The method of claim 2 , wherein the software application executes the workflow that includes a plurality of states, and wherein the state changes are between pairs of the plurality of states.
4 . The method of claim 3 , wherein the resources are computing hardware, computing software, or human agents assigned to carry out tasks associated with at least some of the plurality of states.
5 . The method of claim 1 , wherein the indications of availabilities of the resources include a schedule of resource availability.
6 . The method of claim 1 , wherein the determining the training time series of software application activities comprises:
decomposing an initially-determined time series into variables representing trends, periodic changes, and aperiodic changes; and
using the variables to extend the initially-determined time series, by way of extrapolation, forward or backward in time.
7 . A non-transitory computer-readable medium storing program instructions that, when executed by one or more processors of a computing system, cause the computing system to perform operations comprising:
obtaining a log regarding execution of a software application;
obtaining indications of availabilities of resources related to the software application;
determining, from the log and the indications of availabilities of the resources, a workflow graph representing a workflow of software application activities, wherein the workflow graph includes a plurality of workflow states, transitions between workflow states in the plurality of workflow states, and transition probabilities associated with each of the transitions, wherein each workflow state corresponds to a node of the workflow graph, wherein each transition between workflow states corresponds to an edge of the workflow graph;
determining, from the log, the workflow graph, the transition probabilities, and the indications of availabilities of the resources, a training time series of the software application activities;
training a prediction engine comprising a long short-term memory (LSTM) machine learning model stored in a memory with the training time series of the software application activities using an LSTM trainer, wherein the LSTM trainer produces the LSTM machine learning model with a preconfigured number of cells based on the training time series of the software application activities, wherein the prediction engine as trained is configured to receive an input time series of further software application activities and generate an output time series that predicts additional software application activities;
obtaining an altered input time series of the software application activities, wherein the altered input time series of the software application activities represents an alteration to at least one of the software application activities, the workflow, the transition probabilities, or the resources; and
generating, using the prediction engine, a predicted output time series that predicts additional software application activities based on the altered input time series of the software application activities.
8 . The non-transitory computer-readable medium of claim 7 , wherein the log includes data representing state changes of the software application and timestamps of when the state changes occurred.
9 . The non-transitory computer-readable medium of claim 8 , wherein the software application executes the workflow that includes a plurality of states, and wherein the state changes are between pairs of the plurality of states.
10 . The non-transitory computer-readable medium of claim 7 , wherein the indications of availabilities of the resources include a schedule of resource availability.
11 . The non-transitory computer-readable medium of claim 7 , wherein the determining the training time series of software application activities comprises:
decomposing an initially-determined time series into variables representing trends, periodic changes, and aperiodic changes; and
using the variables to extend the initially-determined time series, by way of extrapolation, forward or backward in time.
12 . The non-transitory computer-readable medium of claim 9 , wherein the resources are computing hardware, computing software, or human agents assigned to carry out tasks associated with at least some of the plurality of states.
13 . A system comprising:
one or more processors; and
memory, containing program instructions that, upon execution by the one or more processors, cause the system to perform operations comprising:
obtaining a log regarding execution of a software application;
obtaining indications of availabilities of resources related to the software application;
determining, from the log and the indications of availabilities of the resources, a workflow graph representing a workflow of software application activities, wherein the workflow graph includes a plurality of workflow states, transitions between workflow states in the plurality of workflow states, and transition probabilities associated with each of the transitions, wherein each workflow state corresponds to a node of the workflow graph, wherein each transition between workflow states corresponds to an edge of the workflow graph;
determining, from the log, the workflow graph, the transition probabilities, and the indications of availabilities of the resources, a training time series of the software application activities;
training a prediction engine comprising a long short-term memory (LSTM) machine learning model stored in a memory with the training time series of the software application activities using an LSTM trainer, wherein the LSTM trainer produces the LSTM machine learning model with a preconfigured number of cells based on the training time series of the software application activities, wherein the prediction engine as trained is configured to receive an input time series of further software application activities and generate an output time series that predicts additional software application activities;
obtaining an altered input time series of the software application activities, wherein the altered input time series of the software application activities represents an alteration to at least one of the software application activities, the workflow, the transition probabilities, or the resources; and
generating, using the prediction engine, a predicted output time series that predicts additional software application activities based on the altered input time series of the software application activities.
14 . The system of claim 13 , wherein the log includes data representing state changes of the software application and timestamps of when the state changes occurred.
15 . The system of claim 14 , wherein the software application executes the workflow that includes a plurality of states, and wherein the state changes are between pairs of the plurality of states.
16 . The system of claim 15 , wherein the resources are computing hardware, computing software, or human agents assigned to carry out tasks associated with at least some of the plurality of states.
17 . The system of claim 13 , wherein the indications of availabilities of the resources include a schedule of resource availability.
18 . The system of claim 13 , wherein the determining the training time series of software application activities comprises:
decomposing an initially-determined time series into variables representing trends, periodic changes, and aperiodic changes; and
using the variables to extend the initially-determined time series, by way of extrapolation, forward or backward in time.