Quantum Clustering
———————————


This web app was developed by Roi Dannon in NodeJS, through March-June of 2017, under the supervision of Prof. David Horn.


It was implemented according to “COMPACT” simulation (original Compact version by Roy Varshavsky were implemented by Guy Shaked, February 2012.)


This web app enables a quick way for visualizing quantum clustering algorithm. Follow these steps to use the app:


1) Choose dataset: In this part the user should specify the desired dataset for analysis. The user can choose between an existing file in the server's database, or the user can upload a file (tsv/csv format) of  own. The user can Change the settings in this step if the csv doesn't contain header names or wish to cluster columns instead of rows by ticking the “File contains header row/column” and “Cluster Columns” accordingly.


Hit “Done” button.


2) Features selection: Sort the dataset's columns into relevant "Features" (Mandatory), "Classification" (Truth set, Optional) and "Ignored". Drag the header boxes from side to side.


Hit “Done” button.


3) Quantum Clustering: In this step the user can choose the preprocessing parameters and the algorithm parameters, and show the results.


Preprocessing Parameters:


* Use PCA - check if the user wants to use PCA matrix instead of the original data, one can use the PCA Variance graph at the bottom of the web page to help the user decide. 
* Custom principal components - this option only works if when check “Use PCA” option is on. this allows the user to select specific PCA dimensions. Consult also the graph at the bottom of the web page for choosing the right dimensions.
* Use normalization - this option normalizes the data around the origin (0,0,0,…)


Algorithm Parameters:


* sigma 
* Step size - required for gradient descent (Rule of thumb: sigma/7)
* Cluster Max Distance - At the end of the simulation, the algorithm cluster datapoints that reside in a certain epsilon-ball. this value is the circumference of that ball.
* Max Iterations - The simulation will stops if either all the datapoint reach an equilibrium or the simulation iterated for “maximum iterations” times.


Hit “Run QC Algorithm”


The web app now sends a request to the server to do the calculations the clustering and shows a loading circle. once the loading circle disappear the entire iterations are loaded onto the web browser. The user can now hit “Play” and “Pause” to play through the algorithm evolution.


The user can always hit “Download Initial :& Final States” to store the algorithm initial data and final data.


If the user specified a classification column in Step 2, he can scroll down to see Jaccard Similarity Coefficient results and download the clusters of the data.