def softmax_2_class_problems():
return [
(_Spec(pg.SoftmaxRegression, (10, 2), {}), datasets.random(
10, 1000, random_seed=123, sep=2.0), 100),
(_Spec(pg.SoftmaxRegression, (100, 2), {}), datasets.random(
100, 1000, random_seed=123), 50),
(_Spec(pg.SoftmaxRegression, (200, 2), {}), datasets.random(
200, 1000, random_seed=123, sep=1.5), 20),
(_Spec(pg.SoftmaxRegression, (256, 2), {}), datasets.random(
256, 1000, random_seed=123, sep=1.5), 100),
]
def softmax_2_class_problems():
return [
(_Spec(pg.SoftmaxRegression, (10, 2),
{}), datasets.random(10, 1000, random_seed=123, sep=2.0), 100),
(_Spec(pg.SoftmaxRegression, (100, 2),
{}), datasets.random(100, 1000, random_seed=123), 50),
(_Spec(pg.SoftmaxRegression, (200, 2),
{}), datasets.random(200, 1000, random_seed=123, sep=1.5), 20),
(_Spec(pg.SoftmaxRegression, (256, 2),
{}), datasets.random(256, 1000, random_seed=123, sep=1.5), 100),
]
def test_problems():
"""Test problems for visualizations."""
# Unlike the training problem sets, these test problems are made up of
# length-5 tuples. The final items in the tuple are the name of the problem
# and the initialization random_seed for testing consistency.
tp = [
(_Spec(pg.Quadratic, (20, ),
{"random_seed": 1234}), None, None, "quad_problem", 5678),
(_Spec(pg.Quadratic, (20, ), {
"noise_stdev": 1.0,
"random_seed": 1234
}), None, None, "quad_problem_noise", 5678),
(_Spec(pg.Rosenbrock, (),
{"random_seed": 1234}), None, None, "rosenbrock", 5678),
(_Spec(pg.Rosenbrock, (), {
"random_seed": 1234,
"noise_stdev": 1.0
}), None, None, "rosenbrock_noise", 5678),
(_Spec(pg.SoftmaxRegression, (10, 2),
{}), datasets.random(10, 10000,
random_seed=1234), 100, "softmax", 5678),
(_Spec(pg.SoftmaxRegression, (10, 2), {"noise_stdev": 1.0}),
datasets.random(10, 10000,
random_seed=1234), 100, "softmax_noise", 5678),
(_Spec(pg.FullyConnected, (10, 2),
{}), datasets.random(10, 10000,
random_seed=1234), 100, "mlp_small",
_test_problem_mlp_scaled_init_small()),
(_Spec(pg.FullyConnected, (20, 10),
{}), datasets.random(20, 10000, n_classes=10,
random_seed=1234), 100, "mlp_large",
_test_problem_mlp_scaled_init_large()),
(_Spec(pg.FullyConnected, (784, 10), {
"hidden_sizes": (64, ),
"activation": tf.nn.sigmoid
}), datasets.mnist(), 64, "mlp_mnist_sigmoid",
_test_problem_mlp_scaled_init_mnist()),
(_Spec(pg.FullyConnected, (784, 10), {
"hidden_sizes": (64, ),
"activation": tf.nn.relu
}), datasets.mnist(), 64, "mlp_mnist_relu",
_test_problem_mlp_scaled_init_mnist()),
(_Spec(pg.ConvNet, ((1, 28, 28), 10, [(3, 3, 8), (5, 5, 8)]),
{"activation": tf.nn.sigmoid}), datasets.mnist(), 64,
"convnet_mnist_sigmoid", None),
(_Spec(pg.ConvNet, ((1, 28, 28), 10, [(3, 3, 8), (5, 5, 8)]),
{"activation": tf.nn.relu}), datasets.mnist(), 64,
"convnet_mnist_relu", None),
]
return tp
def test_problems():
"""Test problems for visualizations."""
# Unlike the training problem sets, these test problems are made up of
# length-5 tuples. The final items in the tuple are the name of the problem
# and the initialization random_seed for testing consistency.
tp = [
(_Spec(pg.Quadratic, (20,), {"random_seed": 1234}), None, None,
"quad_problem", 5678),
(_Spec(pg.Quadratic, (20,), {"noise_stdev": 1.0, "random_seed": 1234}),
None, None, "quad_problem_noise", 5678),
(_Spec(pg.Rosenbrock, (), {"random_seed": 1234}), None, None,
"rosenbrock", 5678),
(_Spec(pg.Rosenbrock, (), {"random_seed": 1234, "noise_stdev": 1.0}),
None, None, "rosenbrock_noise", 5678),
(_Spec(pg.SoftmaxRegression, (10, 2), {}), datasets.random(
10, 10000, random_seed=1234), 100, "softmax", 5678),
(_Spec(pg.SoftmaxRegression, (10, 2), {"noise_stdev": 1.0}),
datasets.random(10, 10000, random_seed=1234), 100, "softmax_noise",
5678),
(_Spec(pg.FullyConnected, (10, 2), {}), datasets.random(
10, 10000, random_seed=1234), 100, "mlp_small",
_test_problem_mlp_scaled_init_small()),
(_Spec(pg.FullyConnected, (20, 10), {}), datasets.random(
20, 10000, n_classes=10, random_seed=1234), 100, "mlp_large",
_test_problem_mlp_scaled_init_large()),
(_Spec(pg.FullyConnected, (784, 10),
{"hidden_sizes": (64,), "activation": tf.nn.sigmoid}),
datasets.mnist(), 64, "mlp_mnist_sigmoid",
_test_problem_mlp_scaled_init_mnist()),
(_Spec(pg.FullyConnected, (784, 10),
{"hidden_sizes": (64,), "activation": tf.nn.relu}),
datasets.mnist(), 64, "mlp_mnist_relu",
_test_problem_mlp_scaled_init_mnist()),
(_Spec(pg.ConvNet, ((1, 28, 28), 10, [(3, 3, 8), (5, 5, 8)]),
{"activation": tf.nn.sigmoid}), datasets.mnist(), 64,
"convnet_mnist_sigmoid", None),
(_Spec(pg.ConvNet, ((1, 28, 28), 10, [(3, 3, 8), (5, 5, 8)]),
{"activation": tf.nn.relu}), datasets.mnist(), 64,
"convnet_mnist_relu", None),
]
return tp