package stat;

import java.awt.*;
import java.io.*;
import parameters.Parameter;
import parameters.ParameterSet;
import parameters.ParameterSetArray;


/** Saves the cross-correlation coefficient between each two parameters in the
	wheel plot. At the bottom of the file, also prints the largest x correlation
	coefficients where x is set in the code (currently set to 20). Does not test for
	significance yet. */
public class CrossCorrStat extends Stat implements Runnable {
	public CrossCorrStat() {}
	
	public void run() {
		int		i, j, k, l;
		int		num_largest = 20;
		String	[] largest_names = new String[num_largest];
		float [] largest_values = new float[num_largest];
		
		for(k = 0; k < num_largest; k++) largest_values[k] = 0;
		
		if(paramSets == null) {
			System.out.println("No Cam loaded - Stat did nothing");
			return;
		}
		FileDialog fd = new FileDialog(new Frame(), "Save into file:", FileDialog.SAVE);
		fd.show();
		if(fd.getFile() == null) return;
		try {
			PrintWriter ps = new PrintWriter(new FileOutputStream(fd.getDirectory() + fd.getFile()));
			double [] mat = makeParameterMatrix();
	
			if(mat == null) {
				System.out.println("Problem making matrix of parameters");
				ps.close();
				return;
			}
			if(numSets < 2) {
				System.out.println("You need at least two sets of parameters in the cam in order to get a correlation");
				ps.close();
				return;
			}
	
		// First print out labels
			ps.print("\t");
			printParameterLabels(prototype, ps);
			ps.println("");

			for(i = 0; i < prototype.getNumParams(); i++) {
				ps.print(prototype.getName(i) + "\t");	// Print the param label for this row
				for(j = 0; j <= i; j++) {
					float	val = getCorrelation(mat, i, j);
					ps.print(val + "\t");
					if(i != j) {
						k = 0;
						while(k < num_largest && Math.abs(val) < Math.abs(largest_values[k])) k++;
						if(k < num_largest) {
							// Move everything else down
							for(l = num_largest - 1; l > k; l--) {
								largest_names[l] = largest_names[l - 1];
								largest_values[l] = largest_values[l - 1];
							}
							largest_names[k] = new String(prototype.getName(i) + "-" + prototype.getName(j));
							largest_values[k] = val;
						}
					}
				}
				ps.println("");
			}

			for(k = 0; k < num_largest; k++) {
		System.out.println(largest_names[k] + "\t" + largest_values[k]);
				if(Math.abs(largest_values[k]) > -1) ps.println(largest_names[k] + "\t" + largest_values[k]);
			}
		} catch(Exception e) { System.out.println("Error in CrossCorrStat: " + e.toString()); }
		System.out.println("Finished calculating cross correlations");
	}
	
	/** Returns the correlation coefficient between two parameters given the data in the input matrix.
		Not sure what stats book I pulled this formula from. */
	float getCorrelation(double [] mat, int param1, int param2) {
		double	sum1 = 0, sum2 = 0, sum_sqr1 = 0, sum_sqr2 = 0;
		int		i, num_params = prototype.getNumParams();
		
		for(i = 0; i < numSets; i++) {
			sum1 += mat[i * num_params + param1];
			sum_sqr1 += mat[i * num_params + param1] * mat[i * num_params + param1];
			sum2 += mat[i * num_params + param2];
			sum_sqr2 += mat[i * num_params + param2] * mat[i * num_params + param2];
		}
		
		double	var1 = Math.sqrt((sum_sqr1 - (sum1 * sum1 / numSets)) / numSets);	
		double	var2 = Math.sqrt((sum_sqr2 - (sum2 * sum2 / numSets)) / numSets);	
		sum1 /= numSets;  sum2 /= numSets;
		
		float	p = 0;
		for(i = 0; i < numSets; i++) {
			p += (mat[i * num_params + param1] - sum1) * (mat[i * num_params + param2] - sum2);
		}
		p /= numSets * var1 * var2;
		return p;
	}
	
}