Ocelescope - Object-Centric Process Mining

// OBJECT-CENTRIC PROCESS MINING

An extensible web framework for object-centric process mining.

Upload an OCEL, filter it down, discover models, and extend everything with plugins. Spin it up with Docker and analyze in minutes - entirely in your browser.

Get started Documentation

Open source Runs locally Docker RWTH PADS

bash

# two commands and you are running $ curl -LO https://www.ocelescope.org/docker-compose.yaml $ docker compose up -d ok backend ghcr.io/promi4s/ocelescope-backend ok frontend http://localhost:3000 Ocelescope is ready.

OCEL 2.0 PM4PY-compatible Plugin library Custom React modules Self-hosted

LOG OVERVIEW

Understand any OCEL at a glance.

Upload a log and Ocelescope profiles it instantly: events and objects, every type, attribute value ranges and the time span they cover. Know what you are working with before you write a single query.

eventsobjectsobject & event typesvalue rangestime ranges

localhost:3000/logOverview

FILTER CONSTRUCTOR

Get the perspective you need.

Compose filters visually to shrink an OCEL. Stack conditions across object and event types, attribute values, time windows and the existence of O2O or E2O relations - then carry the reduced log straight into discovery.

object type & activity selectionattribute-value conditionstime range windowsO2O & E2O existence filters

localhost:3000/filter

DISCOVERY

Discover models - and add your own methods.

Out of the box, the discovery tab renders object-centric Petri nets and directly-follows graphs. Need a different algorithm? Define it plugin-style, upload the zip, and it appears as a selectable method right in the tab.

OC Petri netsdirectly-follows graphs+ your method

localhost:3000/discovery

WHY OCELESCOPE

Ocelescope is where research code becomes reusable software.

Ocelescope provides a plugin system to integrate new functionality at runtime. A method should not disappear into a one-off prototype for publication. Package it, upload it, run it - and let others build on it.

BUILD A PLUGIN

Declare your parameters. Ocelescope renders the UI.

No frontend work. Type your plugin inputs in Python and the form on the right is generated for you.

Read the plugin guide Follow the tutorial →

localhost:3000/plugin/ocel-graph

from typing import Annotated

from ocelescope import (
    OCEL,
    OCELAnnotation,
    Plugin,
    plugin_method,
)

from .input import OCELGraphInput
from .resource import OCELGraph
from .util import mine_ocel_graph


class OcelGraphDiscovery(Plugin):
    label = "Ocel Graph"
    description = "Discovers a Object-Centric event log graph"
    version = "1.0.3"

    @plugin_method(label="Mine OCEL Graph", description="Mines a OCEL Graph")
    def mine_ocel_graph(
        self,
        ocel: Annotated[
            OCEL,
            OCELAnnotation(label="Event Log", description="The log from which the ocel graph should be mined from"),
        ],
        input: OCELGraphInput,
    ) -> OCELGraph:
        return mine_ocel_graph(ocel, input)

from ocelescope import OCEL_FIELD, PluginInput
from pydantic import BaseModel, Field


class ObjectRoot(BaseModel):
    class Config:
        title = "Object"

    object_id: str = OCEL_FIELD(
        field_type="object_id",
        title="Object Id",
        ocel_id="ocel",
        description="The ID of the Event which is the root of the OcelGraph",
    )


class EventRoot(BaseModel):
    class Config:
        title = "Event"

    event_id: str = OCEL_FIELD(
        field_type="event_id",
        title="Event Id",
        ocel_id="ocel",
        description="The ID of the Event which is the root of the OcelGraph",
    )


class OCELGraphInput(PluginInput):
    root: ObjectRoot | EventRoot
    depth: int = Field(
        title="OCEL Graph Depth", description="The maximum depth of the ocel graph", default=3, gt=0, le=10
    )
    max_neighbours: int = Field(
        title="Maximum Neighbours", description="The maximum amount of neighbours a node can have", gt=0, default=5
    )

from ocelescope import Graph, GraphEdge, GraphNode, GraphvizLayoutConfig, Resource, generate_color_map
from pydantic import BaseModel


class EventNode(BaseModel):
    id: str
    activity: str


class ObjectNode(BaseModel):
    id: str
    object_type: str


class Relation(BaseModel):
    qualifier: str
    source: str
    target: str
    object_type: str | None = None


class OCELGraph(Resource):
    label = "Ocel Graph"
    description = "A Ocel graph"

    events: list[EventNode] = []
    objects: list[ObjectNode] = []
    relations: list[Relation] = []

    @property
    def event_ids(self) -> list[str]:
        return [event.id for event in self.events]

    @property
    def object_ids(self) -> list[str]:
        return [object.id for object in self.objects]

    def visualize(self) -> Graph:
        color_map = generate_color_map(list(set([object.object_type for object in self.objects])))

        object_nodes = [
            GraphNode(
                id=object_node.id, shape="rectangle", label=object_node.id, color=color_map[object_node.object_type]
            )
            for object_node in self.objects
        ]

        event_nodes = [GraphNode(id=event.id, shape="rectangle", label=event.id) for event in self.events]

        edges = [
            GraphEdge(
                source=edge.source,
                target=edge.target,
                label=edge.qualifier,
                color=color_map[edge.object_type] if edge.object_type else None,
            )
            for edge in self.relations
        ]

        return Graph(
            type="graph",
            nodes=object_nodes + event_nodes,
            edges=edges,
            layout_config=GraphvizLayoutConfig(engine="neato", graphAttrs={"overlap": "prism"}),
        )

from typing import cast

import pandas as pd
from ocelescope import OCEL

from .input import EventRoot, OCELGraphInput
from .resource import EventNode, ObjectNode, OCELGraph, Relation


def group_relation_entity(
    df: pd.DataFrame,
    entity_ids: list[str],
    id_column: str,
    type_column: str,
    target_id_column: str,
):
    """
    Count how many 'target' entities are linked to each entity (event or object).

    Returns:
        DataFrame with columns: id, type, count
    """
    return (
        df[df[id_column].isin(entity_ids)]
        .groupby([id_column, type_column])[target_id_column]
        .size()
        .reset_index()
        .rename(columns={id_column: "id", type_column: "type", target_id_column: "count"})
    )


def mine_ocel_graph(ocel: OCEL, input: OCELGraphInput):
    graph = OCELGraph()

    events_to_visit = []
    objects_to_visit = []

    if isinstance(input.root, EventRoot):
        root_id = input.root.event_id
        root = ocel.events.df[ocel.events.df[ocel.ocel.event_id_column] == input.root.event_id].iloc[0]
        events_to_visit.append(EventNode(id=input.root.event_id, activity=root[ocel.ocel.event_activity]))
    else:
        root_id = input.root.object_id
        root = ocel.objects.df[ocel.objects.df[ocel.ocel.object_id_column] == input.root.object_id].iloc[0]
        objects_to_visit.append(ObjectNode(id=input.root.object_id, object_type=root[ocel.ocel.object_type_column]))

    for _ in range(input.depth):
        # Get current frontier IDs
        event_ids_to_visit = [event.id for event in events_to_visit]
        object_ids_to_visit = [obj.id for obj in objects_to_visit]

        # Get event-object relations using XOR and not already in the graph
        relations: pd.DataFrame = cast(
            pd.DataFrame,
            ocel.e2o.df[
                (
                    (ocel.e2o.df[ocel.ocel.event_id_column].isin(event_ids_to_visit))
                    ^ (ocel.e2o.df[ocel.ocel.object_id_column].isin(object_ids_to_visit))
                )
                & ~(ocel.e2o.df[ocel.ocel.event_id_column].isin(graph.event_ids))
                & ~(ocel.e2o.df[ocel.ocel.object_id_column].isin(graph.object_ids))
            ],
        )

        # Count object neighbors per event
        events = group_relation_entity(
            df=relations,
            entity_ids=event_ids_to_visit,
            id_column=ocel.ocel.event_id_column,
            type_column=ocel.ocel.event_activity,
            target_id_column=ocel.ocel.object_id_column,
        )

        # Count event neighbors per object
        e2o_objects = group_relation_entity(
            df=relations,
            entity_ids=object_ids_to_visit,
            id_column=ocel.ocel.object_id_column,
            type_column=ocel.ocel.object_type_column,
            target_id_column=ocel.ocel.event_id_column,
        )

        # Get object-object (o2o) relations using XOR
        o2o = cast(
            pd.DataFrame,
            ocel.o2o.typed_df[
                (
                    (ocel.o2o.df["ocel:oid_1"].isin(object_ids_to_visit))
                    ^ (ocel.o2o.df["ocel:oid_2"].isin(object_ids_to_visit))
                )
                & ~(ocel.o2o.df["ocel:oid_1"].isin(graph.object_ids))
                & ~(ocel.o2o.df["ocel:oid_2"].isin(graph.object_ids))
            ],
        )

        # Normalize and mirror o2o (treat as undirected)
        o2o = o2o.rename(
            columns={"ocel:oid_1": "id", "ocel:type_1": "type", "ocel:oid_2": "target_id", "ocel:type_2": "target_type"}
        )

        mirrored = o2o.rename(
            columns={"target_id": "id", "target_type": "type", "id": "target_id", "type": "target_type"}
        )

        mirrored_o2o = pd.concat([o2o, mirrored], ignore_index=True).rename(columns={"ocel:qualifier": "qualifier"})

        # Count o2o neighbours for each object
        o2o_objects = group_relation_entity(
            df=mirrored_o2o,
            entity_ids=object_ids_to_visit,
            id_column="id",
            type_column="type",
            target_id_column="target_id",
        )

        # Combine object neighbour counts
        objects = (
            pd.concat([o2o_objects, e2o_objects], ignore_index=True)
            .groupby(["id", "type"], as_index=False)["count"]
            .sum()
        )

        # Update graph with this layer
        graph.objects = graph.objects + objects_to_visit
        graph.events = graph.events + events_to_visit

        # Prepare for next layer
        events_to_visit = []
        objects_to_visit = []

        object_id_with_neighbours = [
            row["id"] for _, row in objects.iterrows() if row["count"] <= input.max_neighbours or row["id"] == root_id
        ]
        event_id_with_neighbours = [
            row["id"] for _, row in events.iterrows() if row["count"] <= input.max_neighbours or row["id"] == root_id
        ]

        for _, row in (
            cast(pd.DataFrame, mirrored_o2o[mirrored_o2o["id"].isin(object_id_with_neighbours)])
            .drop_duplicates(subset=["target_id"], keep="first")
            .iterrows()
        ):
            graph.relations.append(
                Relation(source=str(row["id"]), target=str(row["target_id"]), qualifier=str(row["qualifier"]))
            )
            objects_to_visit.append(ObjectNode(id=str(row["target_id"]), object_type=str(row["target_type"])))

        for _, row in (
            cast(pd.DataFrame, relations[relations["ocel:oid"].isin(object_id_with_neighbours)])
            .drop_duplicates(subset=["ocel:eid"], keep="first")
            .iterrows()
        ):
            graph.relations.append(
                Relation(
                    source=str(row["ocel:eid"]),
                    target=str(row["ocel:oid"]),
                    qualifier=str(row["ocel:qualifier"]),
                    object_type=str(row["ocel:type"]),
                )
            )
            events_to_visit.append(EventNode(id=str(row["ocel:eid"]), activity=str(row["ocel:activity"])))

        for _, row in (
            cast(
                pd.DataFrame,
                relations[
                    relations["ocel:eid"].isin(event_id_with_neighbours)
                    & ~relations["ocel:oid"].isin([obj.id for obj in objects_to_visit])
                ],
            )
            .drop_duplicates(subset=["ocel:oid"], keep="first")
            .iterrows()
        ):
            graph.relations.append(
                Relation(
                    source=str(row["ocel:eid"]),
                    target=str(row["ocel:oid"]),
                    qualifier=str(row["ocel:qualifier"]),
                    object_type=str(row["ocel:type"]),
                )
            )
            objects_to_visit.append(ObjectNode(id=str(row["ocel:oid"]), object_type=str(row["ocel:type"])))

    graph.objects = graph.objects + objects_to_visit
    graph.events = graph.events + events_to_visit

    return graph

EXTENSIBLE UI

When a form is not enough, ship a full module.

Each plugin gets a generated page. For richer workflows, custom React modules mount as first-class views. This way, the OCELOT module was integrated.

per-plugin pagescustom React modulesOcelot module

localhost:3000/modules/ocelot

PLUGIN ECOSYSTEM

A growing library of analyses.

PD PM4PY Discovery v1.0.5 by Görkem Öztürk A discovery plugin for ocelescope using the pm4py library 1 star updated Jun 2026 Download ->

TM TOTeM v1.7.2 by Görkem Öztürk A plugin to generate Temporal Object Type Models 1 star updated Jun 2026 Download ->

OD OC-DECLARE v1.0.5 by Görkem Öztürk OC Declare 2 stars updated Jun 2026 Download ->

OG OCEL Graph v1.0.5 by Görkem Öztürk A Ocel graph plugin for ocelescope 1 star updated Jun 2026 Download ->

Browse all plugins Submit your plugin →

RESOURCES

Everything to get unstuck.

Documentation

START Get started guide

Install Ocelescope and load your first OCEL log.

PLUGINS Plugin guide

Build, package, and load reusable analysis methods.

GitHub

SUBMIT Submit a plugin

Share a plugin with the Ocelescope ecosystem.

PACKAGES Request an environment package

Ask for a Python package to be added to the plugin execution environment.

ISSUES Report an issue

Open a focused bug report with reproduction details.

IDEAS Request a feature

Suggest improvements for the framework or docs.

SOURCE GitHub & troubleshooting

Browse the source code and project activity.

GET STARTED

Running in two commands.

Download the compose file, start the stack, and open the frontend on localhost.

bash

$ curl -LO https://www.ocelescope.org/docker-compose.yaml $ docker compose up -d ok backend ghcr.io/promi4s/ocelescope-backend ok frontend http://localhost:3000 Ocelescope is ready.

Step-by-Step Guide Download compose file