def generate_data(N, trans, sample_rate, Ngrid, def_param=(shared_input_data, shared_data)):
data = None
if direction == "u":
if not os.path.exists("../../cache/barkley/raw/{0}_{1}.uv.dat.npy".format(N, Ngrid)):
data = bh.generate_data(N, 20000, 5, Ngrid=Ngrid)
np.save("../../cache/barkley/raw/{0}_{1}.uv.dat.npy".format(N, Ngrid), data)
else:
data = np.load("../../cache/barkley/raw/{0}_{1}.uv.dat.npy".format(N, Ngrid))
else:
if not os.path.exists("../../cache/mitchell/raw/{0}_{1}.vh.dat.npy".format(N, Ngrid)):
data = mh.generate_data(N, 20000, 50, Ngrid=Ngrid)
np.save("../../cache/mitchell/raw/{0}_{1}.vh.dat.npy".format(N, Ngrid), data)
else:
data = np.load("../../cache/mitchell/raw/{0}_{1}.vh.dat.npy".format(N, Ngrid))
data = data[0, :]
input_y, input_x, output_y, output_x = hp.create_patch_indices(
(center - (half_inner_size+border_size), center + (half_inner_size+border_size) + right_border_add),
(center - (half_inner_size+border_size), center + (half_inner_size+border_size) + right_border_add),
(center - (half_inner_size), center + (half_inner_size) + right_border_add),
(center - (half_inner_size), center + (half_inner_size) + right_border_add))
input_data = data[:, input_y, input_x].reshape(ndata, -1)
if prediction_mode in ["NN", "RBF"]:
shared_input_data[:] = hp.create_1d_delay_coordinates(input_data, delay_dimension=ddim, tau=tau[direction]).reshape((ndata, -1))
else:
shared_input_data[:] = input_data[:]
shared_data[:] = data[:]
prediction[:] = data[trainLength:trainLength+predictionLength]
prediction[:, output_y, output_x] = 0.0
def gather_data(filename):
f = open(filename, "rb")
viewData = pickle.load(f)
f.close()
print("data loaded")
diff = None
for (name, value) in viewData:
if name.lower() == "diff":
diff = value
break
N = 150
indices = []
for i in range(31):
outer = (N // 2 - 32 + i, N // 2 + 32 - i)
inner = (N // 2 - 31 + i, N // 2 + 31 - i)
outer_y, outer_x, inner_y, inner_x = hp.create_patch_indices(
outer, outer, inner, inner)
indices.append((outer_y, outer_x))
indices.append(([75], [75]))
errors = []
for ind in indices:
mse = np.mean(diff[:, ind[0], ind[1]]**2)
errors.append(mse)
return errors
def mainFunction():
generate_data(ndata, 20000, 50, Ngrid=N)
queue = Queue() # use manager.queue() ?
print("preparing threads...")
pool = Pool(processes=16, initializer=get_prediction_init, initargs=[queue,])
_, _, output_y, output_x = hp.create_patch_indices(
(center - (half_inner_size+border_size), center + (half_inner_size+border_size) + right_border_add),
(center - (half_inner_size+border_size), center + (half_inner_size+border_size) + right_border_add),
(center - (half_inner_size), center + (half_inner_size) + right_border_add),
(center - (half_inner_size), center + (half_inner_size) + right_border_add))
jobs = []
for i in range(len(output_y)):
jobs.append((output_y[i], output_x[i]))
print("fitting...")
process_results_process = Process(target=process_thread_results, args=(queue, len(jobs)))
process_results_process.start()
pool.map(get_prediction, jobs)
pool.close()
process_results_process.join()
print("finished fitting")
prediction[prediction < 0.0] = 0.0
prediction[prediction > 1.0] = 1.0
diff = (shared_data[trainLength:trainLength+predictionLength]-prediction)
mse_validation = np.mean((diff[:predictionLength-testLength, output_y, output_x])**2)
mse_test = np.mean((diff[predictionLength-testLength:predictionLength, output_y, output_x])**2)
print("validation error: {0}".format(mse_validation))
print("test error: {0}".format(mse_test))
model = "barkley" if direction == "u" else "mitchell"
view_data = [("Orig", shared_data[trainLength:trainLength+testLength]), ("Pred", prediction), ("Diff", diff)]
if prediction_mode == "NN":
output_file = open("../../cache/{0}/viewdata/inner_cross_pred/{1}/{2}_viewdata_{3}_{4}_{5}_{6}_{7}.dat".format(
model, direction, prediction_mode.lower(), trainLength, inner_size, border_size, ddim, k), "wb")
elif prediction_mode == "RBF":
output_file = open("../../cache/{0}/viewdata/inner_cross_pred/{1}/{2}_viewdata_{3}_{4}_{5}_{6}_{7}_{8}.dat".format(
model, direction, prediction_mode.lower(), trainLength, inner_size, border_size, ddim, width, basis_points), "wb")
else:
output_file = open("../../cache/{0}/viewdata/inner_cross_pred/{1}/{2}_viewdata_{3}_{4}_{5}_{6}_{7}.dat".format(
model, direction, prediction_mode.lower(), trainLength, inner_size, border_size, regression_parameter, n_units), "wb")
pickle.dump(view_data, output_file)
output_file.close()