QuantumClassifier: QuantumClassifier

In QuantumOps: Performs Common Linear Algebra Operations Used in Quantum Computing and Implements Quantum Algorithms

Description

Quantum classifier which was proposed by Maria Schuld (2018). Consists of code blocks which have parallel single qubit quantum gates followed by controlled qubit gates. Takes as input samples and a corresponding list of labels indicating the correct output value of each sample. Will update the parameters of the gates in order to correctly identify the samples.

Usage

QuantumClassifier(	n=8,B=2,r=c(1,3),
			data=NULL,labels=NULL,digit=0,
			eta=1,decay=1,bsc=1,t=20,tag="",pl=TRUE,train=TRUE,
			validT=FALSE,vdata=NULL,vlabels=NULL,
			pretrained=FALSE,alpha=NULL,beta=NULL,gamma=NULL,bias=NULL,
		       	writeParameters=FALSE,outputPath=NULL	)

Arguments

n	Number of qubits that this the classifier will use.
B	Number of blocks in the circuit. A block consists of n single qubit gates applied to each of the qubits in parallel and then n/(gcd,r) controlled qubit gates to perform a maximal entanglement. r is the specified range of the controlled gates
r	Vector containing the range for each block. length(r) should be equal to B. r of 1 means controlled gates are performed on adjacent qubits
data	matrix containing input training data. Rows are individual samples. The number of columns should be equal to 2^n
labels	Vector containing labels of digits. Length must be the same as the number of rows in data
digit	Individual output to identify. The network will attempt to differentiate between inputs that are labelled as digit (in the labels vector) and inputs that are labelled as any other number
eta	learning rate for parameter updates
decay	Multiplier for learning rate after each training iteration. If set to less than 1, the learning rate decays in time
bsc	Scaler for the learning rate of the bias. Setting to a low value will result in other parameters updating faster than the bias
t	Number of training iterations to perform. Total runs is equal to this value multiplied by the number of samples provided
tag	String to attach to name of output files
pl	Boolean indicating whether training output should be plotted
train	Boolean specifying if network should trained on training data, only false if passing in pretrained model
validT	Boolean specifying if the network should be tested on validation data while being trained
vdata	Validation data, necessary if validT is set to true in which case network is tested on this data while being trained. Can be set to same as data.
vlabels	Validation labels, necessary if validT is set to true in which case network is tested on vdata while being trained. Can be set to same as labels.
pretrained	Boolean specifying if a pretrained model is being passed in. If so, alpha, beta, and gamme will be set to intputs, rather than randomized.
alpha	alpha values for gates if pretrained is set to TRUE, should be a vector of length equal to the number of gates in the circuit.
beta	beta values for gates if pretrained is set to TRUE, should be a vector of length equal to the number of gates in the circuit.
gamma	gamma values for gates if pretrained is set to TRUE, should be a vector of length equal to the number of gates in the circuit.
bias	Bias applied to the output of the circuit.
writeParameters	Boolean specifying whether function should write the parameters as it trains. Useful when training takes a long time.
outputPath	String which specifies path to write output parameters to if writeParameters is TRUE. Must have write priveleges in this directory. The function will create two directories inside outputPath, named 0 and 1, and will alternate output to each folder. This prevents corruption of output if interrupted.

Value

List containing a list of the 33 gates and the matrix representing the entire circuit of the trained classifier

Examples

 ## Not run: 
 QuantumClassifier(n=8,B=2,r=c(1,3),
    matrix(sample(256,replace=TRUE),nrow=1),
	array(1),0,1,1,.001,1,"test")
 
## End(Not run)

QuantumOps documentation built on Feb. 3, 2020, 5:07 p.m.