How to create node class for a tree structure with jitclass

rashid1844 · July 20, 2020, 10:49pm

I am trying to use jitclass with node class that is used for a tree structure. The problem is that I am not able to store the parent object reference. I tried types.pyobject but wasn’t successful.

Reproducing code example:

spec = [('parent', types.optional(types.pyobject)),
       ('name', types.unicode_type)]
@jitclass(spec)
class Nodes:
    def __init__(self, parent, name):
        self.parent = parent
        self.name = name

p = Nodes(None, 'parent')
c1 = Nodes(p, 'child1')
c2 = Nodes(p, 'child2')
c3 = Nodes(p, 'child3')

Error message:

TypingError: Failed in nopython mode pipeline (step: nopython frontend)
Internal error at <numba.typeinfer.CallConstraint object at 0x0000015A14537E80>.
Failed in nopython mode pipeline (step: nopython mode backend)
Cannot cast instance.jitclass.Nodes#15a151cd9b0<parent:OptionalType(pyobject) i.e. the type ‘pyobject or None’,name:unicode_type> to OptionalType(pyobject) i.e. the type ‘pyobject or None’: %".31" = load {i8*, {{i8*, i8}, {i8*, i64, i32, i32, i64, i8*, i8*}}}, {i8, {{i8*, i8}, {i8*, i64, i32, i32, i64, i8*, i8*}}} %“parent”

File “”, line 6:
def init(self, parent, name):
self.parent = parent
^

Hannes · July 21, 2020, 6:54am

Hi rashid,
welcome to the forum!

Now this is a bit of a head scratcher. You would want the type of self.parent to be the type associated with the Nodes class, but that is not (obviously) available when you are defining the type spec needed for the compilation in order to create the type.

Is using jitclass ultimately necessary here or could you get away with a simpler, more basic data structure? (Obviously depends a little on what you are planning to do going from here)

That said, I think there is a possible solution to your problem hidden in the numba test suite

github.com

numba/numba/blob/b4badb5f0ecae44ce3fbc57d83a85d24488699a3/numba/tests/test_jitclasses.py#L251-L285


def test_deferred_type(self):
    node_type = deferred_type()

    spec = OrderedDict()
    spec['data'] = float32
    spec['next'] = optional(node_type)

    @njit
    def get_data(node):
        return node.data

    @jitclass(spec)
    class LinkedNode(object):

        def __init__(self, data, next):
            self.data = data
            self.next = next

        def get_next_data(self):
            # use deferred type as argument

This file has been truncated. show original

It seems like there is a way to delay the type specification so that jitclasses can contain members of their own type. @sklam probably has more insight into this

stuartarchibald · July 21, 2020, 10:36am

Perhaps this example for a linked list using jitclass will help:

github.com

numba/numba-examples/blob/master/legacy/linkedlist.py

#!/usr/bin/env python
# -*- coding: utf-8 -*-

"""
This example demonstrates jitclasses and deferred types for writing a
singly-linked-list.
"""
from __future__ import print_function, absolute_import
from collections import OrderedDict
import numpy as np
from numba import njit
from numba import int32, deferred_type, optional
from numba.core.runtime import rtsys
from numba.experimental import jitclass

node_type = deferred_type()

spec = OrderedDict()
spec['data'] = int32
spec['next'] = optional(node_type)

This file has been truncated. show original

Topic		Replies	Views
Jitclass with a deferred type inside a typed list used as a parameter for another jitclass Support: What is this error message?	1	338	November 8, 2023
Implementing n-tree with list Support: How do I do ...?	3	769	June 21, 2022
Compiling jitclass before first instantiation Support: How do I do ...?	2	656	May 10, 2021
Jitclass with input of list of jitclass Community Support	3	599	January 28, 2021
Numba quadtree jitclass issue Community Support	3	1417	August 20, 2020

How to create node class for a tree structure with jitclass

Reproducing code example:

Error message:

Related topics