Standard GP¶

In [1]:

import fastgps
import torch
import numpy as np

import fastgps import torch import numpy as np

In [2]:

torch.set_default_dtype(torch.float64)

torch.set_default_dtype(torch.float64)

True Function¶

In [3]:

def f_ackley(x, a=20, b=0.2, c=2*np.pi, scaling=32.768):
    # https://www.sfu.ca/~ssurjano/ackley.html
    assert x.ndim==2
    x = 2*scaling*x-scaling
    t1 = a*torch.exp(-b*torch.sqrt(torch.mean(x**2,1)))
    t2 = torch.exp(torch.mean(torch.cos(c*x),1))
    t3 = a+np.exp(1)
    y = -t1-t2+t3
    return y
d = 1 # dimension
rng = torch.Generator().manual_seed(17)
x = torch.rand((2**7,d),generator=rng) # random testing locations
y = f_ackley(x) # true values at random testing locations
z = torch.rand((2**8,d),generator=rng) # other random locations at which to evaluate covariance
print("x.shape = %s"%str(tuple(x.shape)))
print("y.shape = %s"%str(tuple(y.shape)))
print("z.shape = %s"%str(tuple(z.shape)))

def f_ackley(x, a=20, b=0.2, c=2*np.pi, scaling=32.768): # https://www.sfu.ca/~ssurjano/ackley.html assert x.ndim==2 x = 2*scaling*x-scaling t1 = a*torch.exp(-b*torch.sqrt(torch.mean(x**2,1))) t2 = torch.exp(torch.mean(torch.cos(c*x),1)) t3 = a+np.exp(1) y = -t1-t2+t3 return y d = 1 # dimension rng = torch.Generator().manual_seed(17) x = torch.rand((2**7,d),generator=rng) # random testing locations y = f_ackley(x) # true values at random testing locations z = torch.rand((2**8,d),generator=rng) # other random locations at which to evaluate covariance print("x.shape = %s"%str(tuple(x.shape))) print("y.shape = %s"%str(tuple(y.shape))) print("z.shape = %s"%str(tuple(z.shape)))

x.shape = (128, 1)
y.shape = (128,)
z.shape = (256, 1)

Construct GP¶

In [4]:

fgp = fastgps.StandardGP(d,seed_for_seq=7)
x_next = fgp.get_x_next(2**6)
y_next = f_ackley(x_next)
fgp.add_y_next(y_next)
print("x_next.shape = %s"%str(tuple(x_next.shape)))
print("y_next.shape = %s"%str(tuple(y_next.shape)))

fgp = fastgps.StandardGP(d,seed_for_seq=7) x_next = fgp.get_x_next(2**6) y_next = f_ackley(x_next) fgp.add_y_next(y_next) print("x_next.shape = %s"%str(tuple(x_next.shape))) print("y_next.shape = %s"%str(tuple(y_next.shape)))

x_next.shape = (64, 1)
y_next.shape = (64,)

In [5]:

pmean = fgp.post_mean(x)
print("pmean.shape = %s"%str(tuple(pmean.shape)))
print("l2 relative error = %.2e"%(torch.linalg.norm(y-pmean)/torch.linalg.norm(y)))

pmean = fgp.post_mean(x) print("pmean.shape = %s"%str(tuple(pmean.shape))) print("l2 relative error = %.2e"%(torch.linalg.norm(y-pmean)/torch.linalg.norm(y)))

pmean.shape = (128,)
l2 relative error = 1.70e-01

In [6]:

data = fgp.fit()
list(data.keys())

data = fgp.fit() list(data.keys())

     iter of 5.0e+03 | loss       | term1      | term2     
    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
            0.00e+00 | 3.19e+06   | 6.38e+06   | -5.54e+02 
            5.00e+00 | 2.29e+06   | 4.57e+06   | -5.44e+02 
            1.00e+01 | 7.51e+05   | 1.50e+06   | -4.97e+02 
            1.50e+01 | 1.98e+03   | 3.62e+03   | 2.31e+02  
            2.00e+01 | 1.15e+04   | 2.27e+04   | 1.79e+02  
            2.50e+01 | 2.86e+02   | 1.48e+02   | 3.07e+02  
            3.00e+01 | 2.44e+02   | 1.74e+01   | 3.53e+02  
            3.50e+01 | 2.27e+02   | 2.64e+01   | 3.10e+02  
            4.00e+01 | 2.16e+02   | 7.85e+01   | 2.35e+02  
            4.50e+01 | 2.14e+02   | 6.94e+01   | 2.41e+02  
            5.00e+01 | 2.14e+02   | 6.56e+01   | 2.44e+02  
            5.50e+01 | 2.14e+02   | 6.37e+01   | 2.46e+02  
            6.00e+01 | 2.14e+02   | 6.38e+01   | 2.46e+02  
            6.10e+01 | 2.14e+02   | 6.38e+01   | 2.46e+02  

Out[6]:

['iterations']

In [7]:

pmean,pvar,q,ci_low,ci_high = fgp.post_ci(x,confidence=0.99)
print("pmean.shape = %s"%str(tuple(pmean.shape)))
print("pvar.shape = %s"%str(tuple(pvar.shape)))
print("q = %.2f"%q)
print("ci_low.shape = %s"%str(tuple(ci_low.shape)))
print("ci_high.shape = %s"%str(tuple(ci_high.shape)))
print("l2 relative error = %.2e"%(torch.linalg.norm(y-pmean)/torch.linalg.norm(y)))
pcov = fgp.post_cov(x,x)
print("pcov.shape = %s"%str(tuple(pcov.shape)))
assert torch.allclose(pcov.diagonal(),pvar) and (pvar>=0).all()
pcov2 = fgp.post_cov(x,z)
print("pcov2.shape = %s"%str(tuple(pcov2.shape)))

pmean,pvar,q,ci_low,ci_high = fgp.post_ci(x,confidence=0.99) print("pmean.shape = %s"%str(tuple(pmean.shape))) print("pvar.shape = %s"%str(tuple(pvar.shape))) print("q = %.2f"%q) print("ci_low.shape = %s"%str(tuple(ci_low.shape))) print("ci_high.shape = %s"%str(tuple(ci_high.shape))) print("l2 relative error = %.2e"%(torch.linalg.norm(y-pmean)/torch.linalg.norm(y))) pcov = fgp.post_cov(x,x) print("pcov.shape = %s"%str(tuple(pcov.shape))) assert torch.allclose(pcov.diagonal(),pvar) and (pvar>=0).all() pcov2 = fgp.post_cov(x,z) print("pcov2.shape = %s"%str(tuple(pcov2.shape)))

pmean.shape = (128,)
pvar.shape = (128,)
q = 2.58
ci_low.shape = (128,)
ci_high.shape = (128,)
l2 relative error = 8.29e-02
pcov.shape = (128, 128)
pcov2.shape = (128, 256)

In [8]:

pcmean,pcvar,q,cci_low,cci_high = fgp.post_cubature_ci(confidence=0.99)
print("pcmean = %.3e"%pcmean)
print("pcvar = %.3e"%pcvar)
print("cci_low = %.3e"%cci_low)
print("cci_high = %.3e"%cci_high)

pcmean,pcvar,q,cci_low,cci_high = fgp.post_cubature_ci(confidence=0.99) print("pcmean = %.3e"%pcmean) print("pcvar = %.3e"%pcvar) print("cci_low = %.3e"%cci_low) print("cci_high = %.3e"%cci_high)

pcmean = 1.846e+01
pcvar = 1.564e-03
cci_low = 1.836e+01
cci_high = 1.856e+01

Project and Increase Sample Size¶

In [9]:

pcov_future = fgp.post_cov(x,z,n=2*fgp.n)
pvar_future = fgp.post_var(x,n=2*fgp.n)
pcvar_future = fgp.post_cubature_var(n=2*fgp.n)

pcov_future = fgp.post_cov(x,z,n=2*fgp.n) pvar_future = fgp.post_var(x,n=2*fgp.n) pcvar_future = fgp.post_cubature_var(n=2*fgp.n)

In [10]:

x_next = fgp.get_x_next(2*fgp.n)
y_next = f_ackley(x_next)
fgp.add_y_next(y_next)
print("l2 relative error = %.2e"%(torch.linalg.norm(y-fgp.post_mean(x))/torch.linalg.norm(y)))
assert torch.allclose(fgp.post_cov(x,z),pcov_future)
assert torch.allclose(fgp.post_var(x),pvar_future)
assert torch.allclose(fgp.post_cubature_var(),pcvar_future)

x_next = fgp.get_x_next(2*fgp.n) y_next = f_ackley(x_next) fgp.add_y_next(y_next) print("l2 relative error = %.2e"%(torch.linalg.norm(y-fgp.post_mean(x))/torch.linalg.norm(y))) assert torch.allclose(fgp.post_cov(x,z),pcov_future) assert torch.allclose(fgp.post_var(x),pvar_future) assert torch.allclose(fgp.post_cubature_var(),pcvar_future)

l2 relative error = 1.44e-01

In [11]:

data = fgp.fit(verbose=False)
print("l2 relative error = %.2e"%(torch.linalg.norm(y-fgp.post_mean(x))/torch.linalg.norm(y)))

data = fgp.fit(verbose=False) print("l2 relative error = %.2e"%(torch.linalg.norm(y-fgp.post_mean(x))/torch.linalg.norm(y)))

l2 relative error = 6.82e-02

In [12]:

pcov_16n = fgp.post_cov(x,z,n=16*fgp.n)
pvar_16n = fgp.post_var(x,n=16*fgp.n)
pcvar_16n = fgp.post_cubature_var(n=16*fgp.n)
x_next = fgp.get_x_next(16*fgp.n)
y_next = f_ackley(x_next)
fgp.add_y_next(y_next)
assert torch.allclose(fgp.post_cov(x,z),pcov_16n)
assert torch.allclose(fgp.post_var(x),pvar_16n)
assert torch.allclose(fgp.post_cubature_var(),pcvar_16n)

pcov_16n = fgp.post_cov(x,z,n=16*fgp.n) pvar_16n = fgp.post_var(x,n=16*fgp.n) pcvar_16n = fgp.post_cubature_var(n=16*fgp.n) x_next = fgp.get_x_next(16*fgp.n) y_next = f_ackley(x_next) fgp.add_y_next(y_next) assert torch.allclose(fgp.post_cov(x,z),pcov_16n) assert torch.allclose(fgp.post_var(x),pvar_16n) assert torch.allclose(fgp.post_cubature_var(),pcvar_16n)

In [ ]: