qrl-graph 0.0.11

Reinforcement Learning Algorithms for the quantum speed up in graphs

QRL_graph

Reinforcement Learning for the quantum speedup in the graph

Given a graph, we try to compute the classical and quantum critical time. The definition of the criticial time is defined as the hitting time of the endpoints with the probility bigger than $p_0$.

Install

pip install qrl_graph==0.0.8

Usage

import numpy as np
from scipy.sparse.csgraph import laplacian
import networkx as nx
import matplotlib.pyplot as plt
import matplotlib 
from qrl_graph.graph_env.graph import Graph

g = np.array([[0, 1, 1, 0],
              [1, 0, 0, 1],
              [1, 0, 0, 1],
              [0, 1, 1, 0]])

g_env = Graph(g=g)
print('Laplacian matrix:\n', g_env.laplacian)

t_cl = g_env.get_classical_time(p0=0.1)
t_q = g_env.get_quantum_time(p0=0.1)

print('Classical time:', t_cl)
print('Quantum time:', t_q)
print('Speed up:', t_cl / t_q)


# uncomment to show the graph
# g_env.show_graph()

The results are

Laplacian matrix:
 [[ 2 -1 -1  0]
 [-1  2  0 -1]
 [-1  0  2 -1]
 [ 0 -1 -1  2]]
Classical time: 0.25000000000000006
Quantum time: 0.6000000000000003
Speed up: 0.4166666666666665

Side notes

for the policy gradient, First, the probability is parametrized through $$\sigma(x) = \frac 1 {1+e^{-x}}$$, and its derivative is $$\sigma(x) (1 - \sigma(x))$$.

According to the policy gradient, the policy gradient is equal to $$\mathbb E( \sigma(\theta) (1 - \sigma(\theta)) r ).$$