def symbolise(self, last_result=None):
self.show_visualisation()
bounds = np.empty(shape=(self.height * self.depth, 2), dtype=np.float64)
for k in range(self.height):
bounds[k * 3] = [0.001, 1] # wavelength as a fraction of the window width
bounds[k * 3 + 1] = [-1, 2] # offset where window is 0,1
bounds[k * 3 + 2] = [-1, 1] # amplitude. Can be negative
self._working_results = np.zeros(shape=(self.width, self.height), dtype=np.float64)
# cuda init
self._factors = np.zeros(shape=(self.height * self.depth,), dtype=np.float64)
# cuda.profile_start()
self._d_working_results = cuda.to_device(self._working_results)
grid_dim = self.width / 128, self.height - 1
block_dim = 128, 1, 1
self.compute_samples_configured = compute_sample_gpu.configure(grid_dim, block_dim)
with cuda.pinned(self._factors, self._working_results):
output = scipy.optimize.differential_evolution(self.evaluation_function_cuda, bounds=bounds,
strategy='best2bin', maxiter=20, recombination=0.9,
mutation=(0.0001, 0.3), tol=0.01, init='latinhypercube',
popsize=15, disp=True, callback=self.step_callback)
# cuda.profile_stop()
self.evaluation_function_cuda(output.x)
# raw_input("Press Enter to continue...")
print(output)
def driver(pricer, pinned=False):
paths = np.zeros((NumPath, NumStep + 1), order='F')
paths[:, 0] = StockPrice
DT = Maturity / NumStep
if pinned:
from numbapro import cuda
with cuda.pinned(paths):
ts = timer()
pricer(paths, DT, InterestRate, Volatility)
te = timer()
else:
ts = timer()
pricer(paths, DT, InterestRate, Volatility)
te = timer()
ST = paths[:, -1]
PaidOff = np.maximum(paths[:, -1] - StrikePrice, 0)
print 'Result'
fmt = '%20s: %s'
print fmt % ('stock price', np.mean(ST))
print fmt % ('standard error', np.std(ST) / sqrt(NumPath))
print fmt % ('paid off', np.mean(PaidOff))
optionprice = np.mean(PaidOff) * exp(-InterestRate * Maturity)
print fmt % ('option price', optionprice)
print fmt % ('Paths ', NumPath)
print fmt % ('NumStep ', NumStep)
print 'Performance'
NumCompute = NumPath * NumStep
print fmt % ('Mstep/second', '%.2f' % (NumCompute / (te - ts) / 1e6))
print fmt % ('time elapsed', '%.3fs' % (te - ts))
if '--plot' in sys.argv:
from matplotlib import pyplot
pathct = min(NumPath, 100)
for i in xrange(pathct):
pyplot.plot(paths[i])
print 'Plotting %d/%d paths' % (pathct, NumPath)
pyplot.show()
def driver(pricer, pinned=False):
paths = np.zeros((NumPath, NumStep + 1), order='F')
paths[:, 0] = StockPrice
DT = Maturity / NumStep
if pinned:
from numbapro import cuda
with cuda.pinned(paths):
ts = timer()
pricer(paths, DT, InterestRate, Volatility)
te = timer()
else:
ts = timer()
pricer(paths, DT, InterestRate, Volatility)
te = timer()
ST = paths[:, -1]
PaidOff = np.maximum(paths[:, -1] - StrikePrice, 0)
print 'Result'
fmt = '%20s: %s'
print fmt % ('stock price', np.mean(ST))
print fmt % ('standard error', np.std(ST) / sqrt(NumPath))
print fmt % ('paid off', np.mean(PaidOff))
optionprice = np.mean(PaidOff) * exp(-InterestRate * Maturity)
print fmt % ('option price', optionprice)
print 'Performance'
NumCompute = NumPath * NumStep
print fmt % ('Mstep/second', '%.2f' % (NumCompute / (te - ts) / 1e6))
print fmt % ('time elapsed', '%.3fs' % (te - ts))
if '--plot' in sys.argv:
from matplotlib import pyplot
pathct = min(NumPath, 100)
for i in xrange(pathct):
pyplot.plot(paths[i])
print 'Plotting %d/%d paths' % (pathct, NumPath)
pyplot.show()
d_src = cuda.to_device(src)
d_dst = cuda.device_array_like(dst)
copy_kernel(d_src, out=d_dst)
d_dst.copy_to_host(dst)
te = timer()
print 'regular', te - ts
del d_src, d_dst
assert np.allclose(dst, src)
# Pinned (pagelocked) memory transfer
with cuda.pinned(src, dst):
ts = timer()
stream = cuda.stream() # use stream to trigger async memory transfer
d_src = cuda.to_device(src, stream=stream)
d_dst = cuda.device_array_like(dst, stream=stream)
copy_kernel(d_src, out=d_dst, stream=stream)
d_dst.copy_to_host(dst, stream=stream)
stream.synchronize()
te = timer()
print 'pinned', te - ts
assert np.allclose(dst, src)
def main():
# Build Filter
laplacian_pts = """
-4 -1 0 -1 -4
-1 2 3 2 -1
0 3 4 3 0
-1 2 3 2 -1
-4 -1 0 -1 -4
""".split()
laplacian = np.array(laplacian_pts, dtype=np.float32).reshape(5, 5)
# Build Image
try:
filename = sys.argv[1]
image = ndimage.imread(filename, flatten=True).astype(np.float32)
except IndexError:
image = misc.lena().astype(np.float32)
print("Image size: %s" % (image.shape,))
response = np.zeros_like(image)
response[:5, :5] = laplacian
# CPU
ts = timer()
cvimage_cpu = fftconvolve(image, laplacian, mode="same")
te = timer()
print("CPU: %.2fs" % (te - ts))
# GPU
threadperblock = 32, 8
blockpergrid = best_grid_size(tuple(reversed(image.shape)), threadperblock)
print("kernel config: %s x %s" % (blockpergrid, threadperblock))
# Trigger initialization the cuFFT system.
# This takes significant time for small dataset.
# We should not be including the time wasted here
cufft.FFTPlan(shape=image.shape, itype=np.complex64, otype=np.complex64)
# Start GPU timer
ts = timer()
image_complex = image.astype(np.complex64)
response_complex = response.astype(np.complex64)
stream1 = cuda.stream()
stream2 = cuda.stream()
fftplan1 = cufft.FFTPlan(shape=image.shape, itype=np.complex64, otype=np.complex64, stream=stream1)
fftplan2 = cufft.FFTPlan(shape=image.shape, itype=np.complex64, otype=np.complex64, stream=stream2)
# pagelock memory
with cuda.pinned(image_complex, response_complex):
# We can overlap the transfer of response_complex with the forward FFT
# on image_complex.
d_image_complex = cuda.to_device(image_complex, stream=stream1)
d_response_complex = cuda.to_device(response_complex, stream=stream2)
fftplan1.forward(d_image_complex, out=d_image_complex)
fftplan2.forward(d_response_complex, out=d_response_complex)
stream2.synchronize()
mult_inplace[blockpergrid, threadperblock, stream1](d_image_complex, d_response_complex)
fftplan1.inverse(d_image_complex, out=d_image_complex)
# implicitly synchronizes the streams
cvimage_gpu = d_image_complex.copy_to_host().real / np.prod(image.shape)
te = timer()
print("GPU: %.2fs" % (te - ts))
# Plot the results
plt.subplot(1, 2, 1)
plt.title("CPU")
plt.imshow(cvimage_cpu, cmap=plt.cm.gray)
plt.axis("off")
plt.subplot(1, 2, 2)
plt.title("GPU")
plt.imshow(cvimage_gpu, cmap=plt.cm.gray)
plt.axis("off")
plt.show()
d_src = cuda.to_device(src)
d_dst = cuda.device_array_like(dst)
copy_kernel(d_src, out=d_dst)
d_dst.copy_to_host(dst)
te = timer()
print 'regular', te - ts
del d_src, d_dst
assert np.allclose(dst, src)
# Pinned (pagelocked) memory transfer
with cuda.pinned(src, dst):
ts = timer()
stream = cuda.stream() # use stream to trigger async memory transfer
d_src = cuda.to_device(src, stream=stream)
d_dst = cuda.device_array_like(dst, stream=stream)
copy_kernel(d_src, out=d_dst, stream=stream)
d_dst.copy_to_host(dst, stream=stream)
stream.synchronize()
te = timer()
print 'pinned', te - ts
assert np.allclose(dst, src)
