def test_basic_batch():
""" Test calling twed_batch using GPUarrays. """
# Call TWED
Res = twed_batch(AA, TAA, BB, TBB, nu, lamb, degree)
print('Python Device Batch cuTWED distances:')
print(Res)
# print("Ref\n",Ref)
assert np.allclose(Ref, Res)
def test_batch():
""" Test running the synthetic control dataset"""
DistanceMatrix = twed_batch(TS, TT, TS, TT, nu, lamb, degree)
name = 'synthetic_distance_matrix_cutwed_batch'
with open(f'{name}.npy', 'wb') as fh:
np.save(fh, DistanceMatrix)
# with sns.axes_style("white"):
# sns.heatmap(DistanceMatrix, square=True, cmap="YlGnBu")
# plt.savefig(f'{name}.png')
return DistanceMatrix
def test_basic_batch_float():
""" Test calling twed_batch using GPUarrays. """
AAf = AA.astype(np.float32)
TAAf = TAA.astype(np.float32)
BBf = BB.astype(np.float32)
TBBf = TBB.astype(np.float32)
# Call TWED
Res = twed_batch(AAf, TAAf, BBf, TBBf, nu, lamb, degree)
print('Python Device Batch cuTWED distances (single precision):')
print(Res)
# print("Ref\n",Ref)
assert np.allclose(Ref, Res)
def test_batch():
D = np.zeros(sz)
print("Computing twed batch")
D = twed_batch(AA, TT, BB, TT, nu, lamb, degree)
assert np.allclose(np.triu(D), DIST)
# In[5]:
# Set algo params
nu = 1.
lamb = 1.
degree = 2
# In[6]:
# Compute all TWED between series in the synthetic index dataset, and plot
# This takes a while for the full dataset, about enough time to go for a walk,
# but fits fine; uses only about 150MiB of GPU memory for the full size..
# I doubt this was reasonably computable at all prior to cuTWED....
DistanceMatrix_batch = twed_batch(TS, TT, TS, TT, nu, lamb, degree)
with sns.axes_style("white"):
sns.heatmap(np.triu(DistanceMatrix_batch), square=True, cmap="YlGnBu")
plt.plot()
# In[7]:
DistanceMatrix_batch[0, -1]
# In[8]:
# NOTE YOU WILL NEED A HIGH MEM MACHINE TO RUN THE FULL DATASET.....
from psutil import virtual_memory
mem = virtual_memory()
ram = mem.total