add the Unit-Disk Mapping Module #125

nez0b · 2025-04-03T10:15:47Z

The added module udm is integrated to Qamomile in the below structure:

├── qamomile
│   ├── core (ising_qubo.py modified)
│   ├── pennylane
│   ├── qiskit
│   ├── quri_parts
│   ├── qutip
│   └── udm
├── tests
│   ├── core
		├── udm
│   ├── pennylane
│   ├── qiskit
│   └── quri_parts
├── poetry.lock
├── pyproject.toml
└── README.md

The updated Qamomilepackage is capable of:

Map QUBO/Ising problem to Unit-Disk Graph (UDG)
Solve Ising Problem by either (a) Brute Force (b) Use scipy/milp to find MWIS solutions
Use Bloqade or Pulser Emulators to solve MWIS-UDG problem.

Jacomichi

Can you change the structure of your code first?

Jacomichi · 2025-04-04T06:26:58Z

qamomile/core/ising_qubo.py

+        J = np.zeros((n, n))
+        h = np.zeros(n)
+
+        for (i, j), value in self.ising_model.quad.items():
+            J[i, j] = value
+            J[j, i] = value
+
+        for i, value in self.ising_model.linear.items():
+            h[i] = value


Is there big reasons to create matrix version of Ising model?
If so, I think it is better to use this instead of your own implementation.

Qamomile/qamomile/core/post_process/local_search.py

Line 30 in 4c77c13

class IsingMatrix:

Ok. I use IsingMatrix Class instead. Create the numpy matrix is easier for using scipy solver.

Jacomichi · 2025-04-04T06:29:25Z

qamomile/core/ising_qubo.py

+
+        Args:
+            use_brute_force: Whether to use brute force enumeration for small problems
+            binary_variables: Whether to use {0,1} variables (True) or {-1,1} variables (False)


I think in Qamomile workflow, library only deals with Ising model (spin variables), so you need to care about {0,1} case.
Or if binary {0,1} is better than spin {-1,1} when converting, you can only deal with better formulation.

Ok I removed the {0,1} encoding option. It was for other type of problem.

Jacomichi · 2025-04-04T06:33:49Z

qamomile/udm/copyline.py

+        if isinstance(indices, tuple) and len(indices) == 2:
+            i, j = indices
+
+            if not (1 <= i <= self.height and 1 <= j <= self.width):


i and j start from 1, not 0?

Yes it was easier to compare to the original Julia version. There julia use 1-indexed array.

You might also add a short note in the class or method docstring to indicate that indexing is 1-based.

Jacomichi · 2025-04-04T06:38:14Z

qamomile/core/ising_qubo.py

 import numpy as np
+from ..udm import map_qubo, solve_qubo, qubo_result_to_networkx, QUBOResult


I think udm is independent subpackage, so it should not use in core.
It is better to move this to inside of udm

Ok. Now the entire udm related files are outside of the core. I created a Ising_UnitDiskGraph in the udm module instead.

…sp core.ising_qubo)

ahao27

A few points should be addressed. I’ve included inline comments with suggestions.
let me know if you'd like help implementing any of them.

ahao27 · 2025-04-07T07:42:49Z

qamomile/udm/__init__.py

+"""
+
+
+from .dragondrop import (


The module name dragondrop doesn't clearly show its purpose or functionality. If it's related to solving QUBO problems or MWIS on graphs, a name like graph_solver, qubo_tools, or mwis_utils might be more appropriate.

Changed from dragondrop do mwis_solver

ahao27 · 2025-04-07T07:47:27Z

qamomile/udm/copyline.py

+        if isinstance(indices, tuple) and len(indices) == 2:
+            i, j = indices
+
+            if not (1 <= i <= self.height and 1 <= j <= self.width):


You might also add a short note in the class or method docstring to indicate that indexing is 1-based.

ahao27 · 2025-04-07T07:49:33Z

qamomile/udm/core.py

+SHOW_WEIGHT = False
+
+# ONE type - a singleton to represent the constant 1 for unweighted cases
+class ONE:


Consider renaming ONE to something more descriptive like UnweightedMarker or UnweightedSingleton.

Renamed from ONE to UNWEIGHTED

ahao27 · 2025-04-07T07:57:26Z

qamomile/udm/core.py

+        """Get all vertex indices."""
+        return range(self.num_vertices())
+
+    def cell_matrix(self):


In .cell_matrix, it may be safer to add a bounds check to ensure that node.loc is still in the grid size, to prevent index errors if invalid nodes are passed in.

Added following bound checks:

for node in self.nodes: i, j = node.loc assert 0 <= i < self.size[0], f"i={i} is out of bounds [0, {self.size[0]})" assert 0 <= j < self.size[1], f"j={j} is out of bounds [0, {self.size[1]})" matrix[i][j] = SimpleCell(occupied=True, weight=node.weight)

ahao27 · 2025-04-07T07:59:07Z

qamomile/udm/utils.py

+    return (loc[1], loc[0])
+
+# Apply transformations with a center
+def apply_transform(loc: Tuple[int, int], center: Tuple[int, int], transform_func):


Suggest adding type hints to apply_transform for clarity. For example:

Suggested change

def apply_transform(loc: Tuple[int, int], center: Tuple[int, int], transform_func):

def apply_transform(loc: Tuple[int, int], center: Tuple[int, int], transform_func: Callable[[Tuple[int, int]], Tuple[int, int]]) -> Tuple[int, int]:

Applied your suggestions

ahao27 · 2025-04-07T08:01:49Z

qamomile/udm/utils.py

+
+    return True, diff if diff is not None else 0
+
+def is_unit_disk_graph(grid_graph) -> bool:


Consider adding a verbose=False argument to is_unit_disk_graph, so if a mismatch is found, it can print the problematic pair for easier debugging.

Added a verbose option:

def is_unit_disk_graph(grid_graph, verbose=False) -> bool: # ---------------------------omitted--------------------------------- # If there's a mismatch, this is not a valid unit disk graph if should_be_connected != is_connected: if verbose: print(f"({i}, {j}) should be connected but is not") return False

ahao27 · 2025-04-07T08:04:24Z

qamomile/udm/utils.py

+def simple_graph_from_edgelist(edgelist: List[Tuple[int, int]]) -> nx.Graph:
+    """Create a simple graph from an edge list."""
+    g = nx.Graph()
+    for i, j in edgelist:
+        g.add_edge(i, j)
+    return g


The function simple_graph_from_edgelist can be simplified using:

Suggested change

def simple_graph_from_edgelist(edgelist: List[Tuple[int, int]]) -> nx.Graph:

"""Create a simple graph from an edge list."""

g = nx.Graph()

for i, j in edgelist:

g.add_edge(i, j)

return g

def simple_graph_from_edgelist(edgelist: List[Tuple[int, int]]) -> nx.Graph:

"""Create a simple graph from an edge list."""

g = nx.Graph()

return nx.Graph(edgelist)

ahao27 · 2025-04-07T08:12:47Z

qamomile/udm/dragondrop.py

+    integrality = np.ones(n, dtype=bool)
+
+    # Solve the MILP
+    res = milp(c=c, constraints=constraints, bounds=bounds, integrality=integrality)


solve_mwis_scipy should check res.success to ensure the MILP solver didn't fail such like the case res.success == False

Added solver status message block:

# Solve the MILP res = milp(c=c, constraints=constraints, bounds=bounds, integrality=integrality) # Check solution status if not res.success: status_messages = { 1: "Iteration or time limit reached", 2: "Problem is infeasible", 3: "Problem is unbounded", 4: f"Other error: {res.message}" } error_msg = status_messages.get(res.status, f"Unknown error: {res.message}") raise RuntimeError(f"Failed to find optimal solution: {error_msg}") # The solution vector res.x contains the binary decisions. selected_nodes = [node for node, idx in node_to_idx.items() if round(res.x[idx]) == 1] return selected_nodes, -res.fun, res.x

ahao27 · 2025-04-07T08:15:07Z

qamomile/udm/dragondrop.py

+    gg, pins = post_process_grid(grid, weights, np.zeros_like(weights))
+
+    # Calculate overhead
+    mis_overhead = (n - 1) * n * 4 + n - 4 - 2 * graph.number_of_edges()


Maybe a short docstring to explain fomulation of overhead.

this overhead calculation is for a general mwis instance, which is not fully implmented/tested yet. So comment this line out for now.

nez0b and others added 2 commits April 1, 2025 22:24

init commit for adding the udm module

b171769

Merge branch 'Jij-Inc:main' into main

93035f0

Jacomichi requested review from Jacomichi, yuyamashiro and ahao27 and removed request for Jacomichi April 3, 2025 12:46

upload tutorial to /docs/tutorial/usage/

c39c025

Jacomichi requested changes Apr 7, 2025

View reviewed changes

nez0b added 2 commits April 6, 2025 23:05

udm branch refactor the udm module to be entirely outside the core (e…

54ec5b6

…sp core.ising_qubo)

remove the alternate encoding {0, 1} in solve_qubo.

4fa5c48

ahao27 requested changes Apr 7, 2025

View reviewed changes

nez0b added 2 commits April 7, 2025 22:39

Fix binary encoding bug

37059e7

Address Wei-Hao's PR comments.

143bd61

yuyamashiro merged commit 143bd61 into Jij-Inc:main May 9, 2025

		import numpy as np
		from ..udm import map_qubo, solve_qubo, qubo_result_to_networkx, QUBOResult

	def apply_transform(loc: Tuple[int, int], center: Tuple[int, int], transform_func):
	def apply_transform(loc: Tuple[int, int], center: Tuple[int, int], transform_func: Callable[[Tuple[int, int]], Tuple[int, int]]) -> Tuple[int, int]:


		return True, diff if diff is not None else 0

		def is_unit_disk_graph(grid_graph) -> bool:

add the Unit-Disk Mapping Module #125

add the Unit-Disk Mapping Module #125

Uh oh!

Conversation

nez0b commented Apr 3, 2025

Uh oh!

Jacomichi left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ahao27 left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!