def test_obj(amp=1, order_poly=3, num_knots=5, noise=.1, lenxx = 3, lenxy = 2, lenyx=5, lenyy=8, weights=None):
'''
tests total transform object in high_order
'''
x = (np.random.random(1000) -.5 ) * 4096
y = (np.random.random(1000) -.5 ) * 4096
xref, yref = periodic_dist(x,y,amp, lenxx, lenxy, lenyx, lenyy)
xref = xref + noise * np.random.randn(len(xref))
yref = yref + noise * np.random.randn(len(yref))
t = high_order.transform(x, y, xref, yref, order_poly=order_poly, num_knots=num_knots, weights=weights)
x_u = (np.random.random(1000) -.5 ) * 4096
y_u = (np.random.random(1000) -.5 ) * 4096
xref_u, yref_u = periodic_dist(x_u,y_u,amp, lenxx, lenxy, lenyx, lenyy)
xref_u = xref_u + noise * np.random.randn(len(xref_u))
yref_u = yref_u + noise * np.random.randn(len(yref_u))
xref_u_f , yref_u_f = t.evaluate(x_u, y_u)
print 'average residual i x is ', np.sum(np.abs(xref_u_f - xref_u)) / len(xref_u_f)
print 'average residual i y is ', np.sum(np.abs(yref_u_f - yref_u)) / len(yref_u_f)
def mk_data_per(points=100,order=4):
x1 = np.random.rand(1000)*4096 - 2048
y1 = np.random.rand(1000)*4096 - 2048
xs = np.random.rand(1000)*4096 - 2048
ys = np.random.rand(1000)*4096 - 2048
Zx = np.cos(2 * np.pi * x1 / 4) * np.cos(2 * np.pi * y1 / 2)
Zxs = np.cos(2 * np.pi * xs / 4) * np.cos(2 * np.pi * ys / 2)
Zy = np.cos(2 * np.pi * x1 / 3) * np.cos(2 * np.pi * y1 / 1)
Zys = np.cos(2 * np.pi * xs / 3) * np.cos(2 * np.pi * ys / 1)
xref = x1 + Zx
yref = y1 + Zy
xrefs = xs + Zxs
yrefs = ys + Zys
tpoly = high_order.transform(x1,y1, xref, yref, order_poly=5, fit_spline_b=False, leg=False)
xpoly, ypoly = tpoly.evaluate(x1,y1)
xpolys,ypolys = tpoly.evaluate(xs,ys)
print np.sum(np.abs(xpolys-xrefs))
print np.sum(np.abs(xpoly-xref))
print np.sum(np.abs(xs-xrefs))
return xpoly, ypoly
def test_polynomial_obj(num_co=10, order_poly=5, num_knots=4):
'''
test 5th order polynomial on spline and polnyomial fits
'''
'''
keep rotation zmall, maybe translation a few pixels
'''
xref = (np.random.random(1000) -.5 ) * 4096
yref = (np.random.random(1000) -.5 ) * 4096
#coeff_x = (np.random.random(num_co) - .5) * .01
#coeff_y = (np.random.random(num_co) - .5) * .01
#pick static set instead, then tweaking once it works better
coeff_x = np.array([20.0, .5, .2, .02 , .001 , .07, .009, .003, .004, .006, .0005,.0003,.0007,.0001,.0004])
coeff_y = np.array([20.0, .7, .3, .06 , .004 , .09, .004, .007, .002, .005, .0007,.0005,.0003,.0002,.0005])
x_t = high_order.poly(np.array([xref,yref]), coeff_x)
y_t = high_order.poly(np.array([xref,yref]), coeff_y)
t = high_order.transform(xref,yref,x_t,y_t,order_poly=order_poly, num_knots=num_knots)
x_t_ev, y_t_ev = t.evaluate(xref, yref)
x_u = (np.random.random(2000) -.5 ) * 4096
y_u = (np.random.random(2000) -.5 ) * 4096
xt_u = high_order.poly(np.array([x_u,y_u]), coeff_x)
yt_u = high_order.poly(np.array([x_u,y_u]), coeff_y)
xt_u_m, yt_u_m = t.evaluate(x_u, y_u)
print 'REFERENCE LISTS:average difference between transform and input in x', np.sum(np.abs(xt_u_m - xt_u))/len(xt_u)
print 'REFERENCE LISTS: average difference between transform and input in y', np.sum(np.abs(yt_u_m - yt_u))/len(yt_u)
print 'RANDOM LIST:average difference between transform and input in x', np.sum(np.abs(x_t_ev - x_t))/len(x_t)
print 'RANDOM LIST:average difference between transform and input in y', np.sum(np.abs(y_t_ev - y_t))/len(y_t)
print np.max(x_u), np.min(x_u)
print np.max(xt_u), np.min(xt_u)
return t, coeff_x, coeff_y
def mk_fake_data(points = 1000, param=15):
'''
go ahead and make some fake data
'''
x1 = np.random.rand(1000)*4096 - 2048
y1 = np.random.rand(1000)*4096 - 2048
xs = np.random.rand(1000)*4096 - 2048
ys = np.random.rand(1000)*4096 - 2048
co_ran = [1,10**-6,10**-6 ,10**-8,10**-8,10**-8, 10**-9,10**-9,10**-9,10**-9,10**-10,10**-10,10**-10,10**-10,10**-10]
coeff_trans_x = np.random.randn(param)
coeff_trans_y = np.random.randn(param)
for i in range(len(coeff_trans_x)):
if i ==1:
coeff_trans_x[i] = 1.0
elif i < 15:
coeff_trans_x[i] = coeff_trans_x[i] * co_ran[i]
else:
coeff_trans_x[i] = coeff_trans_x[i] * co_ran[-1]
for i in range(len(coeff_trans_y)):
if i ==2:
coeff_trans_y[i] = 1.0
elif i < 15:
coeff_trans_y[i] = coeff_trans_y[i] * co_ran[i]
else:
coeff_trans_y[i] = coeff_trans_y[i] * co_ran[-1]
coeff_trans_x =[ -3.71658981e-01 , 1.00000000e+00 , 0 , -1.20639707e-09, -1.37110890e-08 , 8.39249153e-09 , 1.69986169e-10 , -1.39590886e-10,-2.00797325e-09, -3.08447522e-10, 1.50453620e-11, -3.51821590e-11,-9.86845535e-11, 8.69664940e-11, -3.52721864e-11]
coeff_trans_y = [ 2.58551652e-01, 0, 1.00000000e+00, 1.18005089e-09,-4.63985850e-09 , 1.42579849e-09 , 1.58246425e-09 , -1.40892711e-09,5.57847993e-11 , 1.16534465e-10 , -3.50121458e-11, 1.49194611e-11,5.37039418e-11, -6.21582929e-11, 6.93338443e-12]
# old coeff -4.40296981e-07
#old coeff -1.26099851e-06
xref = high_order.poly(np.array([x1,y1]), coeff_trans_x)
yref = high_order.poly(np.array([x1,y1]), coeff_trans_y)
xrefs = high_order.poly(np.array([xs,ys]), coeff_trans_x)
yrefs = high_order.poly(np.array([xs,ys]), coeff_trans_y)
plt.quiver(x1,y1,xref-x1,yref-y1)
plt.show()
print 'difference after trnasform', np.sum(np.abs(xref-x1)), np.sum(np.abs(yref-y1))
print np.max(xref), np.min(xref)
print np.max(yref), np.min(yref)
#xref = x1 + del_x
#yref = y1 + del_y
tpoly = high_order.transform(x1,y1, xref, yref, order_poly=3, fit_spline_b=False, leg=False)
xpoly, ypoly = tpoly.evaluate(x1,y1)
xpolys,ypolys = tpoly.evaluate(xs,ys)
print np.sum(np.abs(xpolys-xrefs))
print np.sum(np.abs(ypoly-yref))
#print np.sum(np.abs(xs-xrefs))
print coeff_trans_x
print coeff_trans_y
print (coeff_trans_x[0:len(tpoly.coeff_x)]- tpoly.coeff_x) #/ coeff_trans_x[0:len(tpoly.coeff_x)]
print (coeff_trans_y[0:len(tpoly.coeff_x)]- tpoly.coeff_y) #/ coeff_trans_y[0:len(tpoly.coeff_x)]
tpoly = high_order.transform(x1,y1,xref,yref,leg=True)
tpoly.c_x = np.reshape(np.random.random(25), (5,5))
tpoly.c_y = np.reshape(np.random.random(25), (5,5))
tpoly.leg = True
xref, yref = tpoly.evaluate(x1,y1)
xrefs, yrefs = tpoly.evaluate(xs,ys)
print np.max(xref), np.min(xref)
tleg = high_order.transform(x1,y1,xref,yref, order_poly=5, fit_spline_b=False, leg=True)
xleg, yleg = tleg.evaluate(x1,y1)
xlegs, ylegs = tleg.evaluate(xs,ys)
print np.sum(np.abs(xlegs-xrefs)) / len(xlegs)
print np.sum(np.abs(xleg-xref)) / len(xref)
#print np.sum(np.abs(xs-xrefs))
print tleg.c_y
print tleg.c_x
#print coeff_trans_y
#print (coeff_trans_x[0:len(tpoly.coeff_x)]- tpoly.coeff_x) #/ coeff_trans_x[0:len(tpoly.coeff_x)]
#print (coeff_trans_y[0:len(tpoly.coeff_x)]- tpoly.coeff_y) #/ coeff_trans_y[0:len(tpoly.coeff_x)]
#plt.scatter(xpoly,ypoly)
#plt.show()
return xref, yref
