Mapomatic - Automatic mapping of compiled circuits to low-noise sub-graphs

Overview

mapomatic

Automatic mapping of compiled circuits to low-noise sub-graphs

Overview

One of the main painpoints in executing circuits on IBM Quantum hardware is finding the best qubit mapping. For a given circuit, one typically tries to pick the best initial_layout for a given target system, and then SWAP maps using that set of qubits as the starting point. However there are a couple of issues with that execution model. First, an initial_layout seletected, for example with respect to the noise characteristics of the system, need not be optimal for the SWAP mapping. In practice this leads to either low-noise layouts with extra SWAP gates inserted in the circuit, or optimally SWAP mapped circuits on (possibly) lousy qubits. Second, there is no way to know if the system you targeted in the compilation is actually the best one to execute the compiled circuit on. With 20+ quantum systems, it is hard to determine which device is actually ideal for a given problem.

mapomatic tries to tackle these issues in a different way. mapomatic is a post-compilation routine that finds the best low noise sub-graph on which to run a circuit given one or more quantum systems as target devices. Once compiled, a circuit has been rewritten so that its two-qubit gate structure matches that of a given sub-graph on the target system. mapomatic then searches for matching sub-graphs using the VF2 mapper in Qiskit (retworkx actually), and uses a heuristic to rank them based on error rates determined by the current calibration data. That is to say that given a single target system, mapomatic will return the best set of qubits on which to execute the compiled circuit. Or, given a list of systems, it will find the best system and set of qubits on which to run your circuit. Given the current size of quantum hardware, and the excellent performance of the VF2 mapper, this whole process is actually very fast.

Usage

To begin we first import what we need and load our IBM Quantum account.

import numpy as np
from qiskit import *
import mapomatic as mm

IBMQ.load_account()

Second we will select a provider that has one or more systems of interest in it:

provider = IBMQ.get_provider(group='deployed')

We then go through the usual step of making a circuit and calling transpile on a given backend:

qc = QuantumCircuit(5)
qc.h(0)
qc.cx(0,1)
qc.cx(0,2)
qc.cx(0,3)
qc.cx(0,4)
qc.measure_all()

Here we use optimization_level=3 as it is the best overall. It is also not noise-aware though, and thus can select lousy qubits on which to do a good SWAP mapping

trans_qc = transpile(qc, provider.get_backend('ibm_auckland'),optimization_level=3)

Now, a call to transpile inflates the circuit to the number of qubits in the target system. For small problems like the example here, this prevents us from finding the smaller sub-graphs. Thus we need to deflate the circuit down to just the number of active qubits:

small_qc = mm.deflate_circuit(trans_qc)

This deflated circuit, along with one or more backends can now be used to find the ideal system and mapping. Here we will look over all systems in the provider:

backends = provider.backends()

mm.best_mapping(small_qc, backends)

that returns a tuple with the target layout, system, and the computed error score:

([2, 1, 3, 5, 8], 'ibm_auckland', 0.09518597703355036)

You can then use the best layout in a new call to transpile which will then do the desired mapping for you. Alternatively, we can ask for the best mapping on all systems, yielding a list sorted in order from best to worse:

mm.best_mapping(small_qc, backends, successors=True)
[([2, 1, 3, 5, 8], 'ibm_auckland', 0.09518597703355036),
 ([7, 10, 4, 1, 0], 'ibm_hanoi', 0.11217956761629977),
 ([5, 6, 3, 1, 2], 'ibm_lagos', 0.1123755285308975),
 ([7, 6, 10, 12, 15], 'ibmq_mumbai', 0.13708593236124922),
 ([3, 2, 5, 8, 9], 'ibmq_montreal', 0.13762962991865924),
 ([2, 1, 3, 5, 8], 'ibm_cairo', 0.1423752001642351),
 ([1, 2, 3, 5, 6], 'ibmq_casablanca', 0.15623594190953083),
 ([4, 3, 5, 6, 7], 'ibmq_brooklyn', 0.16468576058762707),
 ([7, 6, 10, 12, 15], 'ibmq_guadalupe', 0.17186581811649904),
 ([5, 3, 8, 11, 14], 'ibmq_toronto', 0.1735555283027388),
 ([5, 4, 3, 1, 0], 'ibmq_jakarta', 0.1792325518776976),
 ([2, 3, 1, 0, 14], 'ibm_washington', 0.2078576175452339),
 ([1, 0, 2, 3, 4], 'ibmq_bogota', 0.23973220166838316),
 ([1, 2, 3, 5, 6], 'ibm_perth', 0.31268969778002176),
 ([3, 4, 2, 1, 0], 'ibmq_manila', 0.3182338194159915),
 ([1, 0, 2, 3, 4], 'ibmq_santiago', 1.0)]

Because of the stochastic nature of the SWAP mapping, the optimal sub-graph may change over repeated compilations.

Getting optimal results

Because the SWAP mappers in Qiskit are stochastic, the number of inserted SWAP gates can vary with each run. The spread in this number can be quite large, and can impact the performance of your circuit. It is thus beneficial to transpile many instances of a circuit and take the best one. For example:

trans_qc_list = transpile([qc]*20, provider.get_backend('ibm_auckland'), optimization_level=3)

best_cx_count = [circ.count_ops()['cx'] for circ in trans_qc_list]
best_cx_count
[10, 13, 10, 7, 7, 10, 10, 7, 10, 7, 10, 10, 10, 10, 5, 7, 6, 13, 7, 10]

We obviously want the one with minimum CNOT gates here:

best_idx = np.where(best_cx_count == np.min(best_cx_count))[0][0]
best_qc = trans_qc_list[best_idx] 

We can then use this best mapped circuit to find the ideal qubit candidates via mapomatic.

best_small_qc = mm.deflate_circuit(best_qc)
mm.best_mapping(best_small_qc, backends, successors=True)
[([11, 13, 14, 16, 19], 'ibm_auckland', 0.07634155667667142),
 ([2, 0, 1, 4, 7], 'ibm_hanoi', 0.0799012562006044),
 ([4, 6, 5, 3, 1], 'ibm_lagos', 0.09374259142721897),
 ([10, 15, 12, 13, 14], 'ibm_cairo', 0.0938958618334792),
 ([5, 9, 8, 11, 14], 'ibmq_montreal', 0.09663069814643488),
 ([10, 6, 7, 4, 1], 'ibmq_mumbai', 0.10253149958591112),
 ([10, 15, 12, 13, 14], 'ibmq_guadalupe', 0.11075230351892806),
 ([11, 5, 4, 3, 2], 'ibmq_brooklyn', 0.13179514610612808),
 ([0, 2, 1, 3, 5], 'ibm_perth', 0.13309987649094324),
 ([4, 6, 5, 3, 1], 'ibmq_casablanca', 0.13570907147053013),
 ([2, 0, 1, 3, 5], 'ibmq_jakarta', 0.14449169384159954),
 ([5, 9, 8, 11, 14], 'ibmq_toronto', 0.1495199193756318),
 ([2, 0, 1, 3, 4], 'ibmq_quito', 0.16858894163955718),
 ([0, 2, 1, 3, 4], 'ibmq_belem', 0.1783430267967986),
 ([0, 2, 1, 3, 4], 'ibmq_lima', 0.20380730100751476),
 ([23, 25, 24, 34, 43], 'ibm_washington', 0.23527393065514557)]
Owner
Qiskit Partners
Qiskit Partners
Handout for the tutorial "Creating publication-quality figures with matplotlib"

Handout for the tutorial "Creating publication-quality figures with matplotlib"

JB Mouret 1.9k Jan 02, 2023
ipyvizzu - Jupyter notebook integration of Vizzu

ipyvizzu - Jupyter notebook integration of Vizzu. Tutorial · Examples · Repository About The Project ipyvizzu is the Jupyter Notebook integration of V

Vizzu 729 Jan 08, 2023
Chem: collection of mostly python code for molecular visualization, QM/MM, FEP, etc

chem: collection of mostly python code for molecular visualization, QM/MM, FEP,

5 Sep 02, 2022
OpenStats is a library built on top of streamlit that extracts data from the Github API and shows the main KPIs

Open Stats Discover and share the KPIs of your OpenSource project. OpenStats is a library built on top of streamlit that extracts data from the Github

Pere Miquel Brull 4 Apr 03, 2022
Learn Data Science with focus on adding value with the most efficient tech stack.

DataScienceWithPython Get started with Data Science with Python An engaging journey to become a Data Scientist with Python TL;DR Download all Jupyter

Learn Python with Rune 110 Dec 22, 2022
Create 3d loss surface visualizations, with optimizer path. Issues welcome!

MLVTK A loss surface visualization tool Simple feed-forward network trained on chess data, using elu activation and Adam optimizer Simple feed-forward

7 Dec 21, 2022
Gaphas is the diagramming widget library for Python.

Gaphas Gaphas is the diagramming widget library for Python. Gaphas is a library that provides the user interface component (widget) for drawing diagra

Gaphor 144 Dec 14, 2022
nptsne is a numpy compatible python binary package that offers a number of APIs for fast tSNE calculation.

nptsne nptsne is a numpy compatible python binary package that offers a number of APIs for fast tSNE calculation and HSNE modelling. For more detail s

Biomedical Visual Analytics Unit LUMC - TU Delft 29 Jul 05, 2022
Rockstar - Makes you a Rockstar C++ Programmer in 2 minutes

Rockstar Rockstar is one amazing library, which will make you a Rockstar Programmer in just 2 minutes. In last decade, people learned C++ in 21 days.

4k Jan 05, 2023
A simple interpreted language for creating basic mathematical graphs.

graphr Introduction graphr is a small language written to create basic mathematical graphs. It is an interpreted language written in python and essent

2 Dec 26, 2021
Farhad Davaripour, Ph.D. 1 Jan 05, 2022
🎨 Python3 binding for `@AntV/G2Plot` Plotting Library .

PyG2Plot 🎨 Python3 binding for @AntV/G2Plot which an interactive and responsive charting library. Based on the grammar of graphics, you can easily ma

hustcc 990 Jan 05, 2023
This is a super simple visualization toolbox (script) for transformer attention visualization ✌

Trans_attention_vis This is a super simple visualization toolbox (script) for transformer attention visualization ✌ 1. How to prepare your attention m

Mingyu Wang 3 Jul 09, 2022
Missing data visualization module for Python.

missingno Messy datasets? Missing values? missingno provides a small toolset of flexible and easy-to-use missing data visualizations and utilities tha

Aleksey Bilogur 3.4k Dec 29, 2022
🗾 Streamlit Component for rendering kepler.gl maps

streamlit-keplergl 🗾 Streamlit Component for rendering kepler.gl maps in a streamlit app. 🎈 Live Demo 🎈 Installation pip install streamlit-keplergl

Christoph Rieke 39 Dec 14, 2022
PolytopeSampler is a Matlab implementation of constrained Riemannian Hamiltonian Monte Carlo for sampling from high dimensional disributions on polytopes

PolytopeSampler PolytopeSampler is a Matlab implementation of constrained Riemannian Hamiltonian Monte Carlo for sampling from high dimensional disrib

9 Sep 26, 2022
DrawBot lets you draw images taken from the internet on Skribbl.io, Gartic Phone and Paint

DrawBot You don't speak french? No worries, english translation is over here. C'est quoi ? DrawBot est un logiciel codé par V2F qui va prendre possess

V2F 205 Jan 01, 2023
Extract and visualize information from Gurobi log files

GRBlogtools Extract information from Gurobi log files and generate pandas DataFrames or Excel worksheets for further processing. Also includes a wrapp

Gurobi Optimization 56 Nov 17, 2022
Apache Superset is a Data Visualization and Data Exploration Platform

Superset A modern, enterprise-ready business intelligence web application. Why Superset? | Supported Databases | Installation and Configuration | Rele

The Apache Software Foundation 50k Jan 06, 2023