def test_FastNewtonTR():
warnings.filterwarnings("ignore")
finder = autocrit.FastNewtonTR
finder_str = "FastNewtonTR"
problem_str = "least squares"
finder_kwargs = {"step_size": 0.5}
num_iters = 25
random_least_squares_problem, random_init = \
shared.generate_random_least_squares()
shared.convergence_test(finder, finder_str, finder_kwargs,
random_least_squares_problem.loss,
random_least_squares_problem, problem_str,
random_init, num_iters)
problem_str = "shallow regression"
finder_kwargs = {"step_size": 0.1}
num_iters = 250
random_regression_problem, network, random_init = \
shared.generate_random_shallow_regression()
shared.convergence_test(finder,
finder_str,
finder_kwargs,
network.loss,
random_regression_problem,
problem_str,
random_init,
num_iters,
test_soln_converge=False)
def test_FastNewtonMR():
warnings.filterwarnings("ignore")
finder = autocrit.FastNewtonMR
finder_str = "FastNewtonMR"
problem_str = "least squares"
finder_kwargs = {"alpha": 0.5, "beta": 0.99, "check_pure": True}
num_iters = 500
random_least_squares_problem, random_init = \
shared.generate_random_least_squares()
shared.convergence_test(finder, finder_str, finder_kwargs,
random_least_squares_problem.loss,
random_least_squares_problem, problem_str,
random_init, num_iters)
problem_str = "shallow regression"
finder_kwargs = {"alpha": 0.5, "beta": 0.9, "rho": 1e-6}
num_iters = 250
random_regression_problem, network, random_init = \
shared.generate_random_shallow_regression()
shared.convergence_test(finder,
finder_str,
finder_kwargs,
network.loss,
random_regression_problem,
problem_str,
random_init,
num_iters,
test_soln_converge=False)
def test_NewtonMethod():
warnings.filterwarnings("ignore")
finder = autocrit.finders.newtons.NewtonMethod
finder_str = "NewtonMethod"
problem_str = "least squares"
finder_kwargs = {}
num_iters = 1
random_least_squares_problem, random_init = \
shared.generate_random_least_squares()
shared.convergence_test(finder, finder_str, finder_kwargs,
random_least_squares_problem.loss,
random_least_squares_problem, problem_str,
random_init, num_iters)
def test_deep_classification():
warnings.filterwarnings("ignore")
problem_str = "deep classification"
finder = autocrit.FastNewtonMR
finder_str = "FastNewtonMR"
finder_kwargs = {"alpha": 1., "beta": 0.5, "rho": 1e-6, "check_pure": True}
num_iters = 250
random_classification_problem, network, random_init = \
shared.generate_random_deep_classification()
shared.convergence_test(finder,
finder_str,
finder_kwargs,
network.loss,
random_classification_problem,
problem_str,
random_init,
num_iters,
test_function_converge=False,
test_soln_converge=False)
finder = autocrit.FastNewtonTR
finder_str = "FastNewtonTR"
finder_kwargs = {"step_size": 0.05}
num_iters = 250
random_classification_problem, network, random_init = \
shared.generate_random_deep_classification(seed=shared.SEED + 1)
shared.convergence_test(finder,
finder_str,
finder_kwargs,
network.loss,
random_classification_problem,
problem_str,
random_init,
num_iters,
test_function_converge=False,
test_soln_converge=False)
def test_GradientNormMinimizer():
warnings.filterwarnings("ignore")
finder = autocrit.GradientNormMinimizer
finder_str = "GradientNormMinimizer"
problem_str = "least squares"
finder_kwargs = {
"minimizer_str": "momentum",
"minimizer_params": {
"lr": 1e-2,
"momentum": 0.9
}
}
num_iters = 1000
random_least_squares_problem, random_init = \
shared.generate_random_least_squares()
shared.convergence_test(finder, finder_str, finder_kwargs,
random_least_squares_problem.loss,
random_least_squares_problem, problem_str,
random_init, num_iters)
def test_GradientDescentOptimizer():
optimizer = autocrit.optimizers.GradientDescentOptimizer
optimizer_str = "GradientDescentOptimizer"
problem_str = "least squares"
optimizer_kwargs = {}
num_iters = 1000
random_least_squares_problem, random_init = \
shared.generate_random_least_squares()
shared.convergence_test(optimizer, optimizer_str, optimizer_kwargs,
random_least_squares_problem.loss,
random_least_squares_problem, problem_str,
random_init, num_iters)
problem_str = "shallow regression"
optimizer_kwargs = {}
num_iters = 10000
random_regression_problem, network, random_init = \
shared.generate_random_shallow_regression()
shared.convergence_test(optimizer, optimizer_str, optimizer_kwargs,
network.loss, random_regression_problem,
problem_str, random_init, num_iters)
problem_str = "shallow classification"
optimizer_kwargs = {}
num_iters = 12500
random_classification_problem, network, random_init = \
shared.generate_random_shallow_classification()
shared.convergence_test(optimizer,
optimizer_str,
optimizer_kwargs,
network.loss,
random_classification_problem,
problem_str,
random_init,
num_iters,
test_soln_converge=False)
def test_BackTrackingLineSearchOptimizer(dim=25):
optimizer = autocrit.optimizers.BackTrackingLineSearchOptimizer
optimizer_str = "BackTrackingLineSearchOptimizer"
problem_str = "least squares"
optimizer_kwargs = {}
num_iters = 1000
random_least_squares_problem, random_init = \
shared.generate_random_least_squares()
shared.convergence_test(optimizer, optimizer_str, optimizer_kwargs,
random_least_squares_problem.loss,
random_least_squares_problem, problem_str,
random_init, num_iters)
problem_str = "shallow regression"
optimizer_kwargs = {"gamma": 1 - 1e-3}
num_iters = 100
random_regression_problem, network, random_init = \
shared.generate_random_shallow_regression()
shared.convergence_test(optimizer, optimizer_str, optimizer_kwargs,
network.loss, random_regression_problem,
problem_str, random_init, num_iters)
problem_str = "shallow classification"
optimizer_kwargs = {"gamma": 1 - 1e-3}
num_iters = 100
random_classification_problem, network, random_init = \
shared.generate_random_shallow_classification()
shared.convergence_test(optimizer,
optimizer_str,
optimizer_kwargs,
network.loss,
random_classification_problem,
problem_str,
random_init,
num_iters,
test_soln_converge=False)
def test_MomentumOptimizer():
warnings.filterwarnings("ignore")
optimizer = autocrit.optimizers.MomentumOptimizer
optimizer_str = "MomentumOptimizer"
problem_str = "least squares"
optimizer_kwargs = {}
num_iters = 1000
random_least_squares_problem, random_init = \
shared.generate_random_least_squares()
shared.convergence_test(optimizer, optimizer_str, optimizer_kwargs,
random_least_squares_problem.loss,
random_least_squares_problem, problem_str,
random_init, num_iters)
problem_str = "shallow regression"
optimizer_kwargs = {}
num_iters = 1000
random_regression_problem, network, random_init = \
shared.generate_random_shallow_regression()
shared.convergence_test(optimizer, optimizer_str, optimizer_kwargs,
network.loss, random_regression_problem,
problem_str, random_init, num_iters)
problem_str = "shallow classification"
optimizer_kwargs = {}
num_iters = 1000
random_classification_problem, network, random_init = \
shared.generate_random_shallow_classification()
shared.convergence_test(optimizer,
optimizer_str,
optimizer_kwargs,
network.loss,
random_classification_problem,
problem_str,
random_init,
num_iters,
test_soln_converge=False)
problem_str = "convolutional classification"
optimizer_kwargs = {"momentum": 0.99}
num_iters = 1000
test_classification_problem, network, random_init = \
shared.generate_test_conv_classification()
shared.convergence_test(optimizer,
optimizer_str,
optimizer_kwargs,
network.loss,
test_classification_problem,
problem_str,
random_init,
num_iters,
test_soln_converge=False)