def on_estimateButton_clicked(self):
import torch
import pickle
import gat
# import gcn
self.ui.statusbar.showMessage("estimate")
if self.model is None:
params = pickle.load(open('model.prms', 'rb'))
# self.model = gcn.GCN(*params)
self.model = gat.GAT(*params)
self.model.load_state_dict(torch.load('model.tch'))
self.model.eval()
device = torch.device("cpu")
structure = self.world.serialize()
train_dataset = socnav.SocNavDataset(structure, mode='train')
train_dataloader = DataLoader(train_dataset, batch_size=1, collate_fn=collate)
for batch, data in enumerate(train_dataloader):
subgraph, feats, labels = data
feats = feats.to(device)
self.model.g = subgraph
# for layer in self.model.layers:
for layer in self.model.gat_layers:
layer.g = subgraph
logits = self.model(feats.float())[0].detach().numpy()[0]
translate = 1. / (1. + math.exp(-logits))*100
if translate < 0: translate = 0
if translate > 100: translate = 100
self.ui.slider.setValue(int(translate))
def on_estimateButton_clicked(self):
self.ui.statusbar.showMessage("estimate")
if self.model is None:
self.device = torch.device('cpu')
self.params = pickle.load(open('../SNGNN_PARAMETERS.prms', 'rb'),
fix_imports=True)
self.GNNmodel = pg_rgcn_gat.PRGAT(self.params[5],
self.params[7],
self.params[8][0],
self.params[14],
self.params[14],
self.params[6],
int(self.params[4] / 2),
int(self.params[4] / 2),
self.params[10],
self.params[9],
self.params[12],
bias=True)
self.GNNmodel.load_state_dict(
torch.load('../SNGNN_MODEL.tch', map_location='cpu'))
self.GNNmodel.eval()
print(json.loads(self.current_line))
graph_type = 'relational'
train_dataset = socnav.SocNavDataset(json.loads(self.current_line),
mode='train',
alt=graph_type,
verbose=False)
train_dataloader = DataLoader(train_dataset,
batch_size=1,
collate_fn=collate)
for batch, data in enumerate(train_dataloader):
subgraph, feats, labels = data
feats = feats.to(self.device)
data = Data(
x=feats.float(),
edge_index=torch.stack(subgraph.edges()).to(self.device),
edge_type=subgraph.edata['rel_type'].squeeze().to(self.device))
logits = self.GNNmodel(data)[0].detach().numpy()[0]
score = logits * 100
if score > 100:
score = 100
elif score < 0:
score = 0
self.ui.slider.setValue(int(score))
def predict(self, sn_scenario):
jsonmodel = self.makeJson(sn_scenario)
graph_type = 'relational'
train_dataset = socnav.SocNavDataset(jsonmodel,
mode='train',
alt=graph_type)
train_dataloader = DataLoader(train_dataset,
batch_size=1,
collate_fn=collate)
for batch, data in enumerate(train_dataloader):
subgraph, feats, labels = data
feats = feats.to(self.device)
data = Data(
x=feats.float(),
edge_index=torch.stack(subgraph.edges()).to(self.device),
edge_type=subgraph.edata['rel_type'].squeeze().to(self.device))
logits = self.GNNmodel(data)[0].detach().numpy()[0]
score = logits * 100
if score > 100:
score = 100
elif score < 0:
score = 0
return score