def plot_meshy(x,y,m,tune=None, interp=1, k=1, eps=0.01, cvx_solver=0,xlab='X',ylab='Y',ftrue=None): # Generating the true function without error for plotting and mse. s = np.linspace(min(x)-eps,max(x)+eps,400) r = ftrue(s) ytrue = ftrue(x) order = np.argsort(x) xs = np.array(x)[order] ys = np.array(y)[order] # Create the coefficients if tune != None: model = meshy.meshy_one(x,y,k, tune = tune,eps=eps,cvx_solver=cvx_solver) else: models = meshy.meshy(x=x,y=y,m=m,k=k,interp=interp,ftrue=ytrue) model = models['minmse.fits'] tune = models['minmse.lam'] modelmse = meshy.meshy_mse(model,y=ytrue) print "Pars: n=%s, lambda = %s, MSE = %s" % (len(y), tune,modelmse) fullfits = meshy.meshy_predict(meshy_one_object=model,x=s) plt.plot(s,r,'k-') plt.scatter(xs,ys,c='k',alpha=0.1,s=10) plt.plot(s,fullfits) #plt.legend(bbox_to_anchor=(0.,1.02,1.,.102),borderaxespad=0.,loc=3,ncol=tsize,mode='expand') plt.xlabel(xlab) plt.ylabel(ylab) plt.show()
def plot_lambdas(x,y,lambdas,m,k,ftrue,mesh=None,interp=1,eps=0.01,lnorm=1,xlab='X',ylab='Y'): # In this function, we plot discrete approximations to # functional minimization problems knowing the truth. # This function is for personal use only. # Single number of cuts and k, multiple tuners. # Generating the true function without error for plotting and mse. s = np.linspace(min(x)-eps,max(x)+eps,400) r = ftrue(s) ytrue = ftrue(x) order = np.argsort(x) xs = np.array(x)[order] ys = np.array(y)[order] # determine number of problems to solve tsize = len(lambdas) # Create empty list to store fitted values #fits = MultiDimList((ncutsize, ksize, tsize)) fits = [] # Create the coefficients for i in range(tsize): model = meshy.meshy_one(x,y,m, k, interp=interp,mesh=mesh,tune = lambdas[i],eps=eps, lnorm=lnorm) modelmse = meshy.meshy_mse(meshy_object = model,y=ytrue) print "Pars: n=%s, lambda = %s, MSE = %s" % (len(y), lambdas[i],modelmse) fits.append(model) if mesh==None: mesh = np.linspace(min(x)-eps,max(x)+eps,m) plt.plot(s,r,'k-') plt.scatter(xs,ys,c='k',alpha=0.1,s=10) plt.vlines(mesh,min(y)-eps,max(y)+eps,alpha=0.5) #if k==0: # for j in range(tsize): # delta = fits[j]['mesh'][1]-fits[j]['mesh'][0] # plt.plot(fits[j]['mesh']+delta/2,fits[j]['theta.hat'],label='$\lambda = $ %s' % lambdas[j]) #else: # for j in range(tsize): # plt.plot(xs,fits[j]['fitted'][order],label='$\lambda = $ %s' % lambdas[j]) for j in range(tsize): full_fitted = meshy.meshy_predict(meshy_one_object=fits[j],x=s) plt.plot(s,full_fitted,label='$\lambda = $ %s' % lambdas[j]) plt.legend(bbox_to_anchor=(0.,1.02,1.,.102),borderaxespad=0.,loc=3,ncol=tsize,mode='expand') plt.xlabel(xlab) plt.ylabel(ylab) plt.show()
