argv_idx = task_id * 5 + 1
dir_in = sys.argv[argv_idx]
dir_out = sys.argv[argv_idx + 1]
fn = sys.argv[argv_idx + 2]
nruns_f = int(sys.argv[argv_idx + 3])
nruns_J = int(sys.argv[argv_idx + 4])
fn_in = dir_in + fn
fn_out = dir_out + fn
alphas, means, icf, x, wishart_gamma, wishart_m = gmm_io.read_gmm_instance(
fn_in + ".txt", replicate_point)
tf, err = utils.timer(f, (alphas, means, icf, x, wishart_gamma, wishart_m),
nruns=nruns_f,
limit=time_limit,
ret_val=True)
print("err:")
print(err)
name = "Theano"
if nruns_J > 0:
tJ, J = utils.timer(fgrad,
(alphas, means, icf, x, wishart_gamma, wishart_m),
nruns=nruns_J,
limit=time_limit,
ret_val=True)
tJ += tf # !!!!!!!!! adding this because no function value is returned by fgrad
gmm_io.write_J(fn_out + "_J_" + name + ".txt", J)
model_dir = dir_in + "model/"
fn_in = dir_in + fn
fn_out = dir_out + fn
params, us, data = hand_io.read_hand_instance(model_dir, fn_in + ".txt",
True)
all_params = np.append(us.flatten(), params)
if data.model.is_mirrored:
mirror_factor = -1.
else:
mirror_factor = 1.
tf, err = utils.timer(
f, (all_params, data.model.nbones, data.model.base_relatives,
data.model.parents, data.model.inverse_base_absolutes,
data.model.base_positions, data.model.weights, mirror_factor,
data.points, data.correspondences, data.model.triangles),
nruns=nruns_f,
limit=time_limit,
ret_val=True)
print("err:")
# print(err)
name = "Theano"
ntheta = params.shape[0]
npts = us.shape[0]
seed = np.zeros((2 + ntheta, all_params.shape[0]), dtype=all_params.dtype)
for i in range(npts):
seed[0][2 * i] = 1.
seed[1][2 * i + 1] = 1.
for i in range(ntheta):
fn_in = dir_in + fn
fn_out = dir_out + fn
params, us, data = hand_io.read_hand_instance(model_dir, fn_in + ".txt",
True)
all_params = np.append(us.flatten(), params)
if data.model.is_mirrored:
mirror_factor = -1.
else:
mirror_factor = 1.
tf, err = utils.timer(
torch_wrapper.torch_func,
(hand_objective_complicated, (all_params, ),
(data.model.nbones, data.model.base_relatives, data.model.parents,
data.model.inverse_base_absolutes, data.model.base_positions,
data.model.weights, mirror_factor, data.points, data.correspondences,
data.model.triangles), False),
nruns=nruns_f,
limit=time_limit,
ret_val=True)
# print("err:")
# print(err)
name = "PyTorch"
ntheta = params.shape[0]
npts = us.shape[0]
seed = torch.zeros((2 + ntheta, all_params.shape[0]), dtype=torch.float64)
for i in range(npts):
seed[0][2 * i] = 1.
seed[1][2 * i + 1] = 1.
nruns_f = int(sys.argv[argv_idx + 3])
nruns_J = int(sys.argv[argv_idx + 4])
model_dir = dir_in + "model/"
fn_in = dir_in + fn
fn_out = dir_out + fn
params, data = hand_io.read_hand_instance(model_dir, fn_in + ".txt", False)
if data.model.is_mirrored:
mirror_factor = -1.
else:
mirror_factor = 1.
tf, err = utils.timer(f, (
params, data.model.nbones, data.model.base_relatives, data.model.parents,
data.model.inverse_base_absolutes, data.model.base_positions,
data.model.weights, mirror_factor, data.points,
data.correspondences
), nruns=nruns_f, limit=time_limit, ret_val=True)
print("err:")
# print(err)
name = "Theano"
if nruns_J > 0:
tJ, J = utils.timer(fjac, (
params, data.model.nbones, data.model.base_relatives, data.model.parents,
data.model.inverse_base_absolutes, data.model.base_positions,
data.model.weights, mirror_factor, data.points,
data.correspondences
), nruns=nruns_J, limit=time_limit, ret_val=True)
tJ += tf # !!!!!!!!! adding this because no function value is returned by fjac
return (reproj_err_d, w_err_d)
dir_in = sys.argv[1]
dir_out = sys.argv[2]
fn = sys.argv[3]
nruns_f = int(sys.argv[4])
nruns_J = int(sys.argv[5])
time_limit = int(sys.argv[6]) if len(sys.argv) >= 7 else float("inf")
fn_in = dir_in + fn
fn_out = dir_out + fn
cams, X, w, obs, feats = ba_io.read_ba_instance(fn_in + ".txt")
tf = utils.timer(ba_objective, (cams, X, w, obs, feats),
nruns=nruns_f,
limit=time_limit)
name = "Autograd"
if nruns_J > 0:
tJ = utils.timer(compute_ba_J, (cams, X, w, obs, feats),
nruns=nruns_J,
limit=time_limit)
# gmm.write_J(fn_out + "_J_" + name + ".txt",grad[1])
else:
tJ = 0
utils.write_times(fn_out + "_times_" + name + ".txt", tf, tJ)
replicate_point = (len(sys.argv) >= 8 and sys.argv[7] == "-rep")
fn_in = dir_in + fn
fn_out = dir_out + fn
def gmm_objective_wrapper(params, x, wishart_gamma, wishart_m):
return gmm.gmm_objective(params[0], params[1], params[2], x, wishart_gamma,
wishart_m)
alphas, means, icf, x, wishart_gamma, wishart_m = gmm_io.read_gmm_instance(
fn_in + ".txt", replicate_point)
tf = utils.timer(gmm.gmm_objective,
(alphas, means, icf, x, wishart_gamma, wishart_m),
nruns=nruns_f,
limit=time_limit)
name = "Autograd"
if nruns_J > 0:
# k = alphas.size
grad_gmm_objective_wrapper = value_and_grad(gmm_objective_wrapper)
tJ, grad = utils.timer(grad_gmm_objective_wrapper,
((alphas, means, icf), x, wishart_gamma, wishart_m),
nruns=nruns_J,
limit=time_limit,
ret_val=True)
gmm_io.write_J(fn_out + "_J_" + name + ".txt", grad[1])
else:
tJ = 0
import ba_objective as ba
dir_in = sys.argv[1]
dir_out = sys.argv[2]
fn = sys.argv[3]
nruns_f = int(sys.argv[4])
nruns_J = int(sys.argv[5])
time_limit = int(sys.argv[6]) if len(sys.argv) >= 7 else float("inf")
fn_in = dir_in + fn
fn_out = dir_out + fn
cams, X, w, obs, feats = ba_io.read_ba_instance(fn_in + ".txt")
tf = utils.timer(torch_wrapper.torch_func,
(ba.ba_objective, (cams, X, w), (obs, feats), False),
nruns=nruns_f,
limit=time_limit)
name = "PyTorch"
if nruns_J > 0:
tJ, res = utils.timer(torch_wrapper.torch_func,
(ba.ba_objective, (cams, X, w), (obs, feats), True),
nruns=nruns_J,
limit=time_limit)
J = res[1]
# TODO write J to file for comparison testing
# gmm.write_J(fn_out + "_J_" + name + ".txt",grad[1])
else:
tJ = 0
utils.write_times(fn_out + "_times_" + name + ".txt", tf, tJ)
print("task_id: %i" % task_id)
argv_idx = task_id * 5 + 1
dir_in = sys.argv[argv_idx]
dir_out = sys.argv[argv_idx + 1]
fn = sys.argv[argv_idx + 2]
nruns_f = int(sys.argv[argv_idx + 3])
nruns_J = int(sys.argv[argv_idx + 4])
fn_in = dir_in + fn
fn_out = dir_out + fn
cams, X, w, obs, feats = ba_io.read_ba_instance(fn_in + ".txt")
tf, err = utils.timer(f, (cams, X, w, obs, feats),
nruns=nruns_f,
limit=time_limit,
ret_val=True)
# print("err:")
# print(err)
name = "Theano"
if nruns_J > 0:
tJ, J = utils.timer(f_compute_ba_J, (cams, X, w, obs, feats),
nruns=nruns_J,
limit=time_limit,
ret_val=True)
tJ += tf # !!!!!!!!! adding this because no function value is returned by fJ
# ba_io.write_J(fn_out + "_J_" + name + ".txt",J)
else:
tJ = 0
model_dir = dir_in + "model/"
fn_in = dir_in + fn
fn_out = dir_out + fn
params, data = hand_io.read_hand_instance(model_dir, fn_in + ".txt", False)
if data.model.is_mirrored:
mirror_factor = -1.
else:
mirror_factor = 1.
tf, err = utils.timer(
torch_wrapper.torch_func,
(hand_objective, (params, ),
(data.model.nbones, data.model.base_relatives, data.model.parents,
data.model.inverse_base_absolutes, data.model.base_positions,
data.model.weights, mirror_factor, data.points,
data.correspondences), False),
nruns=nruns_f,
limit=time_limit,
ret_val=True)
# print("err:")
# print(err)
name = "PyTorch"
if nruns_J > 0:
tJ, res = utils.timer(
torch_wrapper.torch_func,
(hand_objective, (params, ),
(data.model.nbones, data.model.base_relatives, data.model.parents,
data.model.inverse_base_absolutes, data.model.base_positions,
dir_in = sys.argv[1]
dir_out = sys.argv[2]
fn = sys.argv[3]
nruns_f = int(sys.argv[4])
nruns_J = int(sys.argv[5])
time_limit = int(sys.argv[6]) if len(sys.argv) >= 7 else float("inf")
fn_in = dir_in + fn
fn_out = dir_out + fn
main_params, extra_params, state, text_data = lstm_io.read_lstm_instance(
fn_in + ".txt")
tf = utils.timer(torch_wrapper.torch_func,
(lstm.loss, (main_params, extra_params),
(state, text_data), False, True),
nruns=nruns_f,
limit=time_limit)
name = "PyTorch"
if nruns_J > 0:
# k = alphas.size
tJ, res = utils.timer(torch_wrapper.torch_func,
(lstm.loss, (main_params, extra_params),
(state, text_data), True, True),
nruns=nruns_J,
limit=time_limit,
ret_val=True)
lstm_io.write_J(fn_out + "_J_" + name + ".txt", res[1])
else:
tJ = 0
from python_common import gmm_io
import torch_wrapper
import gmm_objective as gmm
dir_in = sys.argv[1]
dir_out = sys.argv[2]
fn = sys.argv[3]
nruns_f = int(sys.argv[4])
nruns_J = int(sys.argv[5])
time_limit = int(sys.argv[6]) if len(sys.argv) >= 7 else float("inf")
replicate_point = (len(sys.argv) >= 8 and sys.argv[7] == "-rep")
fn_in = dir_in + fn
fn_out = dir_out + fn
alphas, means, icf, x, wishart_gamma, wishart_m = gmm_io.read_gmm_instance(fn_in + ".txt", replicate_point)
tf = utils.timer(torch_wrapper.torch_func, (gmm.gmm_objective, (alphas, means, icf), (x, wishart_gamma, wishart_m), False), nruns=nruns_f, limit=time_limit)
name = "PyTorch"
if nruns_J > 0:
# k = alphas.size
tJ, res = utils.timer(torch_wrapper.torch_func, (gmm.gmm_objective, (alphas, means, icf), (x, wishart_gamma, wishart_m), True), nruns=nruns_J, limit=time_limit, ret_val=True)
gmm_io.write_J(fn_out + "_J_" + name + ".txt", res[1])
else:
tJ = 0
utils.write_times(fn_out + "_times_" + name + ".txt", tf, tJ)
