def getPDFfromSetup(pdfoption,lonlatPframe,xMeshP,yMeshP): row,col = np.shape(xMeshP) if pdfoption.lower()=='kde' or pdfoption.lower()=='kernel': ZZpdfPframe = SK.serialK(int(nSamples),lonlatPframe,row,col,xMeshP,yMeshP,.0,0) elif pdfoption=='normal' or pdfoption=='gaussian' or pdfoption=='gauss': nSamples,dims = np.shape(lonlatPframe) if dims!=2: print 'Error in lonlatPframe, dimensions are not correct' print 'Check inputs to getPDFfromSetup in casualtyEstimate.py' exit() mean,covar = meancov.meancovmatrix(lonlatPframe,nSamples) ZZpdfPframe = meancov.normalbivariate(mean,covar,xMeshP,yMeshP,col,row) #fortran version return ZZpdfPframe
def getPDFfromSetup(pdfoption, lonlatPframe, xMeshP, yMeshP): row, col = np.shape(xMeshP) if pdfoption.lower() == 'kde' or pdfoption.lower() == 'kernel': ZZpdfPframe = SK.serialK(int(nSamples), lonlatPframe, row, col, xMeshP, yMeshP, .0, 0) elif pdfoption == 'normal' or pdfoption == 'gaussian' or pdfoption == 'gauss': nSamples, dims = np.shape(lonlatPframe) if dims != 2: print 'Error in lonlatPframe, dimensions are not correct' print 'Check inputs to getPDFfromSetup in casualtyEstimate.py' exit() mean, covar = meancov.meancovmatrix(lonlatPframe, nSamples) ZZpdfPframe = meancov.normalbivariate(mean, covar, xMeshP, yMeshP, col, row) #fortran version return ZZpdfPframe
def getPDFfromSetup(nSamples,pdfoption,lonlatPframe,lonOrMesh,latOrMesh,U,xMeshP,yMeshP): row,col = np.shape(xMeshP) #print xMeshP if pdfoption.lower()=='kde' or pdfoption.lower()=='kernel': ZZpdfPframe = SK.serialK(int(nSamples),lonlatPframe,row,col,xMeshP,yMeshP,.0,0) elif pdfoption=='normal' or pdfoption=='gaussian' or pdfoption=='gauss': ZZpdfPframe = meancov.normalbivariate(mean,covar,xMeshP,yMeshP,col,row) #fortran version #ZZpdf on P frame #xlen,ylen = np.shape(xMeshP) return (xMeshP,yMeshP,ZZpdfPframe,lonOrMesh,latOrMesh)
def calculatePDF(refLonLat, xMesh, yMesh, pdfoption): ''' plt.figure() plt.plot(refLonLat[:,0],refLonLat[:,1],'x') plt.show() exit() ''' xyz, x, y, z = pdf.getxyz(refLonLat) xmean = np.mean(x) ymean = np.mean(y) zmean = np.mean(z) mag = (xmean**2 + ymean**2 + zmean**2) #**.5 lonlat0 = pdf.getlonlat(xmean / mag, ymean / mag, zmean / mag) lon0 = lonlat0[0, 0] lat0 = lonlat0[0, 1] beta = 0 # initial guess nSamples, colvals = np.shape(refLonLat) row, col = np.shape(xMesh) beta = pdf.optimizeHeadingAngle(refLonLat, nSamples, lon0, lat0, beta) R1 = pdf.Rotz(lon0 * np.pi / 180.) # generating rotation matrices R2 = pdf.Roty(-lat0 * np.pi / 180.) R3 = pdf.Rotx(beta * np.pi / 180.) transformDetails = [R1, R2, R3, lon0, lat0, beta] Uint = np.dot(R2, R1) U = np.dot(R3, Uint) lonlatPframe = pdf.originalFrame2Pframe(refLonLat, U) lonlatMesh = np.zeros((np.size(xMesh), 2)) lonlatMesh[:, 0] = np.reshape(xMesh, np.size(xMesh)) lonlatMesh[:, 1] = np.reshape(yMesh, np.size(yMesh)) lonlatPMesh = pdf.originalFrame2Pframe(lonlatMesh, U) ylen, xlen = np.shape(xMesh) lonMeshP = np.reshape(lonlatPMesh[:, 0], [ylen, xlen]) latMeshP = np.reshape(lonlatPMesh[:, 1], [ylen, xlen]) if pdfoption.lower() == 'kde' or pdfoption.lower() == 'kernel': ZZpdfPframe = SK.serialK(int(nSamples), lonlatPframe, row, col, lonMeshP, latMeshP, .0, 0) elif pdfoption == 'normal' or pdfoption == 'gaussian' or pdfoption == 'gauss': ZZpdfPframe = meancov.normalbivariate(mean, covar, lonMeshP, latMeshP, col, row) #fortran version ''' plt.figure() plt.contour(lonMeshP,latMeshP,ZZpdfPframe,30) plt.plot(lonlatPframe[:,0],lonlatPframe[:,1],'x') plt.figure() plt.plot(lonlatPMesh[:,0],lonlatPMesh[:,1],'x') exit() ''' pqrVec, pVec, qVec, rVec = pdf.getxyz( lonlatPMesh) # mesh locations in P frame pdfN = pdf.transformPDF(U, pqrVec, np.pi / 180 * lonlatMesh[:, 0], np.pi / 180 * lonlatMesh[:, 1], np.reshape(ZZpdfPframe, xlen * ylen)) ZZpdfN = np.reshape(pdfN, [ylen, xlen]) lonMesh = np.reshape(lonlatMesh[:, 0], [ylen, xlen]) latMesh = np.reshape(lonlatMesh[:, 1], [ylen, xlen]) ''' plt.figure() plt.contour(lonMeshP,latMeshP,ZZpdfPframe,30) plt.plot(refLonLat[:,0],refLonLat[:,1],'x') plt.show() plt.figure() plt.contour(lonMesh,latMesh,ZZpdfN,30) plt.plot(refLonLat[:,0],refLonLat[:,1],'x') plt.show() ''' return (lonMesh, latMesh, ZZpdfN)
def getPDF(lonlat,nSamples,delta,nsigma,pdfoption): # lon0, lat0 : main vehicle lat and lon used as initial conditions for debris propagation # beta : main vehicle heading angle (counterclockwise starting East) [deg] xyz,x,y,z = getxyz(lonlat) xmean = np.mean(x) ymean = np.mean(y) zmean = np.mean(z) mag = (xmean**2+ymean**2+zmean**2)**.5 lonlat0 = getlonlat(xmean/mag,ymean/mag,zmean/mag) lon0 = lonlat0[0,0] lat0 = lonlat0[0,1] beta = 0 # initial guess beta = optimizeHeadingAngle(lonlat,nSamples,lon0,lat0,beta) R1 = Rotz(lon0*np.pi/180.) # generating rotation matrices R2 = Roty(-lat0*np.pi/180.) R3 = Rotx(beta*np.pi/180.) transformDetails = [R1,R2,R3,lon0,lat0,beta] Uint = np.dot(R2,R1) U = np.dot(R3,Uint) lonlatPframe = originalFrame2Pframe(lonlat,U) # error check for this case not yet implemented #mean,covar = meancov.meancovmatrix(lonlatPframe,nSamples) #D,Ulocal = np.linalg.eig(covar) #lonlatAframe = np.dot(lonlatPframe,Ulocal) # ZZ = meancov.normalbivariate(mean,covar,XX,YY,xlen,ylen) #fortran version if pdfoption.lower()=='kde' or pdfoption.lower()=='kernel': xMeshP,yMeshP,xlen,ylen,areaInt = statsPython.areaOfInterestKDE(lonlatPframe,delta) # A frame could also be used elif pdfoption=='normal' or pdfoption=='gaussian' or pdfoption=='gauss': mean,covar = meancov.meancovmatrix(lonlatPframe,nSamples) xMeshP,yMeshP,xlen,ylen = statsPython.areaOfInterest(mean,covar,delta,nsigma) lonlatPframeMesh = np.zeros((np.size(xMeshP),2)) lonlatPframeMesh[:,0] = np.reshape(xMeshP,np.size(xMeshP)) lonlatPframeMesh[:,1] = np.reshape(yMeshP,np.size(yMeshP)) lonlatOrFrameMesh = Pframe2originalFrame(lonlatPframeMesh,U) lonMesh,latMesh = getLonLatMesh(lonlatOrFrameMesh[:,0],lonlatOrFrameMesh[:,1],delta) lonVec = np.reshape(lonMesh,np.size(lonMesh)) latVec = np.reshape(latMesh,np.size(latMesh)) lonlatVec = np.zeros((len(lonVec),2)) lonlatVec[:,0] = lonVec lonlatVec[:,1] = latVec lonlatVecP = originalFrame2Pframe(lonlatVec,U) ylen,xlen = np.shape(lonMesh) lonMeshP = np.reshape(lonlatVecP[:,0],[ylen,xlen]) latMeshP = np.reshape(lonlatVecP[:,1],[ylen,xlen]) pqrVec ,pVec,qVec,rVec= getxyz(lonlatVecP) # mesh locations in P frame if pdfoption.lower()=='kde' or pdfoption.lower()=='kernel': #ZZpdf = meancov.kde(lonlatPframe,lonMeshP,latMeshP,[nSamples, xlen,ylen]) print 'Starting quad KDE' ZZpdf = SK.serialK(nSamples,lonlatPframe,ylen,xlen,lonMeshP,latMeshP,.1,1) print 'Done with quad KDE' elif pdfoption=='normal' or pdfoption=='gaussian' or pdfoption=='gauss': ZZpdf = meancov.normalbivariate(mean,covar,lonMeshP,latMeshP,xlen,ylen) #fortran version pdfN = transformPDF(U,pqrVec,np.pi/180*lonlatVec[:,0],np.pi/180*lonlatVec[:,1],np.reshape(ZZpdf,xlen*ylen)) ZZpdfN = np.reshape(pdfN,[ylen,xlen]) #plt.figure() #plt.contourf(lonMesh,latMesh,np.reshape(pdfN,[ylen,xlen])) #plt.show() return (lonMesh,latMesh,ZZpdfN,lonMeshP,latMeshP,xlen,ylen,transformDetails,areaInt) #
def getPDF2(lonlat,nSamples,delta,nsigma,pdfoption): # lon0, lat0 : main vehicle lat and lon used as initial conditions for debris propagation # beta : main vehicle heading angle (counterclockwise starting East) [deg] #import time xyz,x,y,z = getxyz(lonlat) xmean = np.mean(x) ymean = np.mean(y) zmean = np.mean(z) mag = (xmean**2+ymean**2+zmean**2) lonlat0 = getlonlat(xmean/mag,ymean/mag,zmean/mag) lon0 = lonlat0[0,0] lat0 = lonlat0[0,1] beta = 0 # initial guess beta = optimizeHeadingAngle(lonlat,nSamples,lon0,lat0,beta) R1 = Rotz(lon0*np.pi/180.) # generating rotation matrices R2 = Roty(-lat0*np.pi/180.) R3 = Rotx(beta*np.pi/180.) transformDetails = [R1,R2,R3,lon0,lat0,beta] Uint = np.dot(R2,R1) U = np.dot(R3,Uint) lonlatPframe = originalFrame2Pframe(lonlat,U) # error check for this case not yet implemented #mean,covar = meancov.meancovmatrix(lonlatPframe,nSamples) #D,Ulocal = np.linalg.eig(covar) #lonlatAframe = np.dot(lonlatPframe,Ulocal) # ZZ = meancov.normalbivariate(mean,covar,XX,YY,xlen,ylen) #fortran version if pdfoption.lower()=='kde' or pdfoption.lower()=='kernel': xMeshP,yMeshP,xlen,ylen= statsPython.areaOfInterestKDE(lonlatPframe,delta) # A frame could also be used elif pdfoption=='normal' or pdfoption=='gaussian' or pdfoption=='gauss': mean,covar = meancov.meancovmatrix(lonlatPframe,nSamples) xMeshP,yMeshP,xlen,ylen = statsPython.areaOfInterest(mean,covar,delta,nsigma) lonlatPframeMesh = np.zeros((np.size(xMeshP),2)) lonlatPframeMesh[:,0] = np.reshape(xMeshP,np.size(xMeshP)) lonlatPframeMesh[:,1] = np.reshape(yMeshP,np.size(yMeshP)) row,col = np.shape(xMeshP) #print xMeshP if pdfoption.lower()=='kde' or pdfoption.lower()=='kernel': ZZpdfPframe = SK.serialK(int(nSamples),lonlatPframe,row,col,xMeshP,yMeshP,.0,0) elif pdfoption=='normal' or pdfoption=='gaussian' or pdfoption=='gauss': ZZpdfPframe = meancov.normalbivariate(mean,covar,xMeshP,yMeshP,col,row) #fortran version #ZZpdf on P frame lonlatOrFrameMesh = Pframe2originalFrame(lonlatPframeMesh,U) lonOrMesh = np.reshape(lonlatOrFrameMesh[:,0],[ylen,xlen]) latOrMesh = np.reshape(lonlatOrFrameMesh[:,1],[ylen,xlen]) ''' print 'maxABS',np.max(np.abs(ZZpdfPframe-ZZpdfPframe0)) plt.figure() plt.contourf(xMeshP,yMeshP,ZZpdfPframe) plt.figure() plt.contourf(xMeshP,yMeshP,ZZpdfPframe0) plt.figure() plt.contourf(xMeshP,yMeshP,ZZpdfPframe0-ZZpdfPframe) plt.figure() plt.plot(lonlatPframe[:,0],lonlatPframe[:,1],'x') plt.figure() plt.contourf(lonOrMesh,latOrMesh,ZZpdfPframe) plt.plot(lonlat[:,0],lonlat[:,1],'x') plt.show() ''' return (xMeshP,yMeshP,ZZpdfPframe,lonOrMesh,latOrMesh,xlen,ylen,transformDetails) #
def getPDF(lonlat,pdfoption='kde',delta=None,nMesh=50): #lonlat is a nx2 matrix with lon lat locations # nsigma is only used for getting PDF area when using normal distribution # pdf option 'kde' or 'normal' dist # delta is spacing used for mesh in PDF # if delta not specified,then it will use nMesh which is the number of points used in one dim...total mesh will be nMesh x nMesh #this routine rotates the SPHERICAL planet to get a better 'flat planet' approximation by using local lat and lons and then applies the KDE or normal PDF routines xyz,x,y,z = getxyz(lonlat) xmean = np.mean(x) ymean = np.mean(y) zmean = np.mean(z) mag = (xmean**2+ymean**2+zmean**2)**.5 lonlat0 = getlonlat(xmean/mag,ymean/mag,zmean/mag) lon0 = lonlat0[0,0] lat0 = lonlat0[0,1] beta = 0 # initial guess nSamples,dims = np.shape(lonlat) if dims!=2: print 'Error in lonlat, dimensions are not correct' print 'Check inputs to getPDF in casualtyEstimate.py' exit() beta = optimizeHeadingAngle(lonlat,nSamples,lon0,lat0,beta) R1 = Rotz(lon0*np.pi/180.) # generating rotation matrices R2 = Roty(-lat0*np.pi/180.) R3 = Rotx(beta*np.pi/180.) transformDetails = [R1,R2,R3,lon0,lat0,beta] Uint = np.dot(R2,R1) U = np.dot(R3,Uint) lonlatPframe = originalFrame2Pframe(lonlat,U) if pdfoption.lower()=='kde' or pdfoption.lower()=='kernel': xMeshP,yMeshP,xlen,ylen,areaInt = statsPython.areaOfInterestKDE(lonlatPframe,delta=delta,nMesh=nMesh) # A frame could also be used elif pdfoption=='normal' or pdfoption=='gaussian' or pdfoption=='gauss': #mean,covar = meancov.meancovmatrix(lonlatPframe,nSamples) #xMeshP,yMeshP,xlen,ylen = statsPython.areaOfInterest(mean,covar,delta,nsigma) xMeshP,yMeshP,xlen,ylen,areaInt = statsPython.areaOfInterestKDE(lonlatPframe,delta=delta,nMesh=nMesh) # A frame could also be used lonlatPframeMesh = np.zeros((np.size(xMeshP),2)) lonlatPframeMesh[:,0] = np.reshape(xMeshP,np.size(xMeshP)) lonlatPframeMesh[:,1] = np.reshape(yMeshP,np.size(yMeshP)) lonlatOrFrameMesh = Pframe2originalFrame(lonlatPframeMesh,U) lonMesh,latMesh = getLonLatMesh(lonlatOrFrameMesh[:,0],lonlatOrFrameMesh[:,1],delta) lonVec = np.reshape(lonMesh,np.size(lonMesh)) latVec = np.reshape(latMesh,np.size(latMesh)) lonlatVec = np.zeros((len(lonVec),2)) lonlatVec[:,0] = lonVec lonlatVec[:,1] = latVec lonlatVecP = originalFrame2Pframe(lonlatVec,U) ylen,xlen = np.shape(lonMesh) lonMeshP = np.reshape(lonlatVecP[:,0],[ylen,xlen]) latMeshP = np.reshape(lonlatVecP[:,1],[ylen,xlen]) pqrVec ,pVec,qVec,rVec= getxyz(lonlatVecP) # mesh locations in P frame if pdfoption.lower()=='kde' or pdfoption.lower()=='kernel': #ZZpdf = meancov.kde(lonlatPframe,lonMeshP,latMeshP,[nSamples, xlen,ylen]) print 'Starting quad KDE' ZZpdf = SK.serialK(nSamples,lonlatPframe,ylen,xlen,lonMeshP,latMeshP,.1,1) print 'Done with quad KDE' elif pdfoption=='normal' or pdfoption=='gaussian' or pdfoption=='gauss': ZZpdf = meancov.normalbivariate(mean,covar,lonMeshP,latMeshP,xlen,ylen) #fortran version pdfN = transformPDF(U,pqrVec,np.pi/180*lonlatVec[:,0],np.pi/180*lonlatVec[:,1],np.reshape(ZZpdf,xlen*ylen)) ZZpdfN = np.reshape(pdfN,[ylen,xlen]) PframeVals = [lonPMesh,latPMesh,ZZpdf] OrFrameVals = [lonMesh,latMesh,ZZpdfN] return (PframeVals,OrFrameVals,transformDetails,areaInt)
mean = [10, -5] cov = [[1, -.6], [-.6, 1]] x, y = np.random.multivariate_normal(mean, cov, 5000).T xnew = np.array(np.matrix(x).T) ynew = np.array(np.matrix(y).T) xvec = np.linspace(0, 20, 600) yvec = np.linspace(-15, 5, 600) dx = xvec[2] - xvec[1] dy = yvec[2] - yvec[1] xy = np.concatenate((xnew, ynew), 1) #print xy #print np.std(ynew) X, Y = np.meshgrid(xvec, yvec) fhatFAST = SK.serialK(5000, xy, 600, 600, X, Y, 2, 1) print fhatFAST print np.sum(fhatFAST * dx * dy) plt.figure(1) plt.plot(x, y, 'x') plt.axis('equal') plt.figure(2) plt.contour(X, Y, fhatFAST, 50) plt.show()
def calculatePDF(refLonLat,xMesh,yMesh,pdfoption): ''' plt.figure() plt.plot(refLonLat[:,0],refLonLat[:,1],'x') plt.show() exit() ''' xyz,x,y,z = pdf.getxyz(refLonLat) xmean = np.mean(x) ymean = np.mean(y) zmean = np.mean(z) mag = (xmean**2+ymean**2+zmean**2)#**.5 lonlat0 = pdf.getlonlat(xmean/mag,ymean/mag,zmean/mag) lon0 = lonlat0[0,0] lat0 = lonlat0[0,1] beta = 0 # initial guess nSamples,colvals = np.shape(refLonLat) row,col = np.shape(xMesh) beta = pdf.optimizeHeadingAngle(refLonLat,nSamples,lon0,lat0,beta) R1 = pdf.Rotz(lon0*np.pi/180.) # generating rotation matrices R2 = pdf.Roty(-lat0*np.pi/180.) R3 = pdf.Rotx(beta*np.pi/180.) transformDetails = [R1,R2,R3,lon0,lat0,beta] Uint = np.dot(R2,R1) U = np.dot(R3,Uint) lonlatPframe = pdf.originalFrame2Pframe(refLonLat,U) lonlatMesh = np.zeros((np.size(xMesh),2)) lonlatMesh[:,0] = np.reshape(xMesh,np.size(xMesh)) lonlatMesh[:,1] = np.reshape(yMesh,np.size(yMesh)) lonlatPMesh = pdf.originalFrame2Pframe(lonlatMesh,U) ylen,xlen = np.shape(xMesh) lonMeshP = np.reshape(lonlatPMesh[:,0],[ylen,xlen]) latMeshP = np.reshape(lonlatPMesh[:,1],[ylen,xlen]) if pdfoption.lower()=='kde' or pdfoption.lower()=='kernel': ZZpdfPframe = SK.serialK(int(nSamples),lonlatPframe,row,col,lonMeshP,latMeshP,.0,0) elif pdfoption=='normal' or pdfoption=='gaussian' or pdfoption=='gauss': ZZpdfPframe = meancov.normalbivariate(mean,covar,lonMeshP,latMeshP,col,row) #fortran version ''' plt.figure() plt.contour(lonMeshP,latMeshP,ZZpdfPframe,30) plt.plot(lonlatPframe[:,0],lonlatPframe[:,1],'x') plt.figure() plt.plot(lonlatPMesh[:,0],lonlatPMesh[:,1],'x') exit() ''' pqrVec ,pVec,qVec,rVec= pdf.getxyz(lonlatPMesh) # mesh locations in P frame pdfN = pdf.transformPDF(U,pqrVec,np.pi/180*lonlatMesh[:,0],np.pi/180*lonlatMesh[:,1],np.reshape(ZZpdfPframe,xlen*ylen)) ZZpdfN = np.reshape(pdfN,[ylen,xlen]) lonMesh = np.reshape(lonlatMesh[:,0],[ylen,xlen]) latMesh = np.reshape(lonlatMesh[:,1],[ylen,xlen]) ''' plt.figure() plt.contour(lonMeshP,latMeshP,ZZpdfPframe,30) plt.plot(refLonLat[:,0],refLonLat[:,1],'x') plt.show() plt.figure() plt.contour(lonMesh,latMesh,ZZpdfN,30) plt.plot(refLonLat[:,0],refLonLat[:,1],'x') plt.show() ''' return (lonMesh,latMesh,ZZpdfN)
def getPDF(lonlat, pdfoption='kde', delta=None, nMesh=50): #lonlat is a nx2 matrix with lon lat locations # nsigma is only used for getting PDF area when using normal distribution # pdf option 'kde' or 'normal' dist # delta is spacing used for mesh in PDF # if delta not specified,then it will use nMesh which is the number of points used in one dim...total mesh will be nMesh x nMesh #this routine rotates the SPHERICAL planet to get a better 'flat planet' approximation by using local lat and lons and then applies the KDE or normal PDF routines xyz, x, y, z = getxyz(lonlat) xmean = np.mean(x) ymean = np.mean(y) zmean = np.mean(z) mag = (xmean**2 + ymean**2 + zmean**2)**.5 lonlat0 = getlonlat(xmean / mag, ymean / mag, zmean / mag) lon0 = lonlat0[0, 0] lat0 = lonlat0[0, 1] beta = 0 # initial guess nSamples, dims = np.shape(lonlat) if dims != 2: print 'Error in lonlat, dimensions are not correct' print 'Check inputs to getPDF in casualtyEstimate.py' exit() beta = optimizeHeadingAngle(lonlat, nSamples, lon0, lat0, beta) R1 = Rotz(lon0 * np.pi / 180.) # generating rotation matrices R2 = Roty(-lat0 * np.pi / 180.) R3 = Rotx(beta * np.pi / 180.) transformDetails = [R1, R2, R3, lon0, lat0, beta] Uint = np.dot(R2, R1) U = np.dot(R3, Uint) lonlatPframe = originalFrame2Pframe(lonlat, U) if pdfoption.lower() == 'kde' or pdfoption.lower() == 'kernel': xMeshP, yMeshP, xlen, ylen, areaInt = statsPython.areaOfInterestKDE( lonlatPframe, delta=delta, nMesh=nMesh) # A frame could also be used elif pdfoption == 'normal' or pdfoption == 'gaussian' or pdfoption == 'gauss': #mean,covar = meancov.meancovmatrix(lonlatPframe,nSamples) #xMeshP,yMeshP,xlen,ylen = statsPython.areaOfInterest(mean,covar,delta,nsigma) xMeshP, yMeshP, xlen, ylen, areaInt = statsPython.areaOfInterestKDE( lonlatPframe, delta=delta, nMesh=nMesh) # A frame could also be used lonlatPframeMesh = np.zeros((np.size(xMeshP), 2)) lonlatPframeMesh[:, 0] = np.reshape(xMeshP, np.size(xMeshP)) lonlatPframeMesh[:, 1] = np.reshape(yMeshP, np.size(yMeshP)) lonlatOrFrameMesh = Pframe2originalFrame(lonlatPframeMesh, U) lonMesh, latMesh = getLonLatMesh(lonlatOrFrameMesh[:, 0], lonlatOrFrameMesh[:, 1], delta) lonVec = np.reshape(lonMesh, np.size(lonMesh)) latVec = np.reshape(latMesh, np.size(latMesh)) lonlatVec = np.zeros((len(lonVec), 2)) lonlatVec[:, 0] = lonVec lonlatVec[:, 1] = latVec lonlatVecP = originalFrame2Pframe(lonlatVec, U) ylen, xlen = np.shape(lonMesh) lonMeshP = np.reshape(lonlatVecP[:, 0], [ylen, xlen]) latMeshP = np.reshape(lonlatVecP[:, 1], [ylen, xlen]) pqrVec, pVec, qVec, rVec = getxyz(lonlatVecP) # mesh locations in P frame if pdfoption.lower() == 'kde' or pdfoption.lower() == 'kernel': #ZZpdf = meancov.kde(lonlatPframe,lonMeshP,latMeshP,[nSamples, xlen,ylen]) print 'Starting quad KDE' ZZpdf = SK.serialK(nSamples, lonlatPframe, ylen, xlen, lonMeshP, latMeshP, .1, 1) print 'Done with quad KDE' elif pdfoption == 'normal' or pdfoption == 'gaussian' or pdfoption == 'gauss': ZZpdf = meancov.normalbivariate(mean, covar, lonMeshP, latMeshP, xlen, ylen) #fortran version pdfN = transformPDF(U, pqrVec, np.pi / 180 * lonlatVec[:, 0], np.pi / 180 * lonlatVec[:, 1], np.reshape(ZZpdf, xlen * ylen)) ZZpdfN = np.reshape(pdfN, [ylen, xlen]) PframeVals = [lonPMesh, latPMesh, ZZpdf] OrFrameVals = [lonMesh, latMesh, ZZpdfN] return (PframeVals, OrFrameVals, transformDetails, areaInt)