positions = np.random.randn(len(ids), dimensions)
print("Running OpenCL")
cl_start = time.time()
results = euclidean_embedder_cl.embed(ids, distance_list, positions,
rate=.0005, iterations=1000)
cl_end = time.time()
with open('ranks_cl.csv', 'w') as csvfile:
writer = csv.writer(csvfile)
writer.writerows(results)
print("Running Python")
py_start = time.time()
positions_dict = dict(zip(ids, list(positions)))
result_dict = euclidean_embedder.embed(positions_dict, distance_list,
rate=.0005, iterations=1000)
py_end = time.time()
with open('ranks_py.csv', 'w') as csvfile:
writer = csv.writer(csvfile)
for team, position in result_dict.items():
writer.writerow([team] + list(position))
print("OpenCL run time: " + str(cl_end - cl_start))
print("Python run time: " + str(py_end - py_start))
dimensions = 1
# Here I generate random initial positions, but choosing better
# initial positions should result in quicker or better results
positions = np.random.randn(len(ids), dimensions)
print("Running OpenCL")
cl_start = time.time()
results = euclidean_embedder_cl.embed(ids, distance_list, positions,
rate=.0005, iterations=10000)
cl_end = time.time()
with open('ranks_cl.csv', 'w') as csvfile:
writer = csv.writer(csvfile)
writer.writerows(results)
print("Running Python")
py_start = time.time()
results = euclidean_embedder.embed(ids, distance_list, positions,
rate=.0005, iterations=100)
py_end = time.time()
with open('ranks_py.csv', 'w') as csvfile:
writer = csv.writer(csvfile)
writer.writerows(results)
print("OpenCL run time: " + str(cl_end - cl_start))
print("Python run time: " + str(py_end - py_start))