def test_batch_forward():
import torch
import malt
representation = malt.models.representation.DGLRepresentation()
portfolio = malt.data.dataset.Dataset(
[malt.Point("C"), malt.Point("CC")]
).featurize_all()
g = next(iter(portfolio.view(batch_size=2, collate_fn="batch_of_g")))
h = representation(g)
assert isinstance(h, torch.Tensor)
assert h.shape[0] == 2
assert h.shape[1] == 1
def test_gp_train():
import torch
import dgl
import malt
net = malt.models.supervised_model.GaussianProcessSupervisedModel(
representation=malt.models.representation.DGLRepresentation(
out_features=128
),
regressor=malt.models.regressor.ExactGaussianProcessRegressor(
in_features=128,
out_features=2,
),
likelihood=malt.models.likelihood.HeteroschedasticGaussianLikelihood(),
)
point = malt.Point("C")
point.featurize()
graph = dgl.batch([point.g])
y = net.loss(graph, torch.tensor([[5.0]]))
y.backward()
y = net.condition(graph)
assert y.mean.item() != 0.0
def test_evolving():
sim = malt.Simulation(
x_dim=200,
y_dim=200,
randomly_generate=True,
num_nodes=40,
scene="scenes/200.out",
)
dr = malt.Drawer(sim.x_dim, sim.y_dim)
dr.draw_risk_grid(sim.scene).update()
r_ref = 100
l_ref = 100
src_x = 20
src_y = 20
for i in xrange(200):
local_instance = sim.step(src_x, src_y, r_ref, l_ref)
dr.draw_nodes(local_instance.sim.node_positions)
# res = local_instance.get_locations()
dr.draw_source(local_instance.get_max_location().get_position(),
(1, 1, 1), -2)
dr.draw_source(malt.Point(src_x, src_y), (0, 1, 0), -3)
dr.update()
dr.clear_all()
# print "Locations:", res
print local_instance.get_error()
def test_forward():
import torch
import malt
representation = malt.models.representation.DGLRepresentation()
point = malt.Point(smiles="C")
point.featurize()
h = representation(point.g)
assert isinstance(h, torch.Tensor)
assert h.shape[0] == 1
assert h.shape[1] == 1
def test_equality():
import torch
import malt
from malt.metrics.supervised_metrics import MSE, MAPE, RMSE, R2
from malt.metrics.base_metrics import mse, mape, rmse, r2
input = torch.randn(5, 1)
target = torch.randn(5, 1)
ds = malt.Dataset()
for idx in range(5):
point = malt.Point(smiles="C", y=target[idx].item())
ds.append(point)
assert MSE()(DumbModel(input), ds) == mse(input, target)
assert MAPE()(DumbModel(input), ds) == mape(input, target)
assert RMSE()(DumbModel(input), ds) == rmse(input, target)
assert R2()(DumbModel(input), ds) == r2(input, target)
def test_forward():
import torch
import malt
net = malt.models.supervised_model.SimpleSupervisedModel(
representation=malt.models.representation.DGLRepresentation(
out_features=128
),
regressor=malt.models.regressor.NeuralNetworkRegressor(
in_features=128, out_features=1
),
likelihood=malt.models.likelihood.HomoschedasticGaussianLikelihood(),
)
point = malt.Point(smiles="C").featurize()
distribution = net.condition(point.g)
assert isinstance(distribution, torch.distributions.Distribution)
assert distribution.batch_shape == torch.Size([1, 1])
net.loss(point.g, 0.0).backward()