def get_leaf_to_path(nodes):
node = nodes[0]
leaf_to_path = get_leaf_to_path(node.G)
wnid_to_node = {node.wnid: node for node in nodes}
leaf_to_path_nodes = {}
for leaf in leaf_to_path:
leaf_to_path_nodes[leaf] = [{
'node': wnid_to_node.get(wnid, None),
'name': wnid_to_name(wnid)
} for wnid in leaf_to_path[leaf]]
return leaf_to_path_nodes
def retprob(names, tree, tree_wnids, wnid_to_outputs, wnid_to_node, wnid,
sample_id):
node = wnid_to_node.get(wnid, None)
if node is not None:
outputs = wnid_to_outputs[node.wnid]
index_child = outputs['preds'][sample_id]
for i in range(2):
prob_child = float(outputs['probs'][sample_id][i])
name_child = wnid_to_name(node.children[i])
tree.append(prob_child)
names.append(name_child)
tree_wnids.append(node.children[i])
retprob(names, tree, tree_wnids, wnid_to_outputs, wnid_to_node,
node.children[i], sample_id)
def traverse_tree(cls, wnid_to_outputs, nodes, wnid_to_class, classes):
"""Convert node outputs to final prediction.
Note that the prediction output for this function can NOT be trained
on. The outputs have been detached from the computation graph.
"""
# move all to CPU, detach from computation graph
example = wnid_to_outputs[nodes[0].wnid]
n_samples = int(example['logits'].size(0))
device = example['logits'].device
for wnid in tuple(wnid_to_outputs.keys()):
outputs = wnid_to_outputs[wnid]
outputs['preds'] = list(map(int, outputs['preds'].cpu()))
outputs['probs'] = outputs['probs'].detach().cpu()
wnid_to_node = {node.wnid: node for node in nodes}
wnid_root = get_root(nodes[0].G)
node_root = wnid_to_node[wnid_root]
decisions = []
preds = []
for index in range(n_samples):
decision = [{'node': node_root, 'name': 'root', 'prob': 1}]
wnid, node = wnid_root, node_root
while node is not None:
if node.wnid not in wnid_to_outputs:
wnid = node = None
break
outputs = wnid_to_outputs[node.wnid]
index_child = outputs['preds'][index]
prob_child = float(outputs['probs'][index][index_child])
wnid = node.children[index_child]
node = wnid_to_node.get(wnid, None)
decision.append({
'node': node,
'name': wnid_to_name(wnid),
'prob': prob_child
})
cls = wnid_to_class.get(wnid, None)
pred = -1 if cls is None else classes.index(cls)
preds.append(pred)
decisions.append(decision)
return torch.Tensor(preds).long().to(device), decisions
def traverse_tree(cls, wnid_to_outputs, nodes, wnid_to_class, classes):
"""Convert node outputs to final prediction.
Note that the prediction output for this function can NOT be trained
on. The outputs have been detached from the computation graph.
"""
# move all to CPU, detach from computation graph
example = wnid_to_outputs[nodes[0].wnid]
n_samples = int(example['logits'].size(0))
for wnid in tuple(wnid_to_outputs.keys()):
outputs = wnid_to_outputs[wnid]
outputs['preds'] = list(map(int, outputs['preds'].cpu()))
outputs['probs'] = outputs['probs'].detach().cpu()
wnid_to_node = {node.wnid: node for node in nodes}
wnid_root = get_root(nodes[0].G)
node_root = wnid_to_node[wnid_root]
decisions = []
preds = []
names = []
tree = []
# retprob(names, tree, wnid_to_outputs, wnid_to_node, wnid_root)
path_inds = []
for index in range(n_samples):
decision = [{'node': node_root, 'name': 'root', 'prob': 1}]
wnid, node = wnid_root, node_root
tr = []
nm = []
tr_wn = []
path_wnids = []
retprob(nm, tr, tr_wn, wnid_to_outputs, wnid_to_node, wnid, index)
names.append(nm)
# print('---')
# print(tr)
# print('---')
tree.append(tr)
while node is not None:
if node.wnid not in wnid_to_outputs:
wnid = node = None
break
outputs = wnid_to_outputs[node.wnid]
index_child = outputs['preds'][index]
prob_child = float(outputs['probs'][index][index_child])
wnid = node.children[index_child]
node = wnid_to_node.get(wnid, None)
decision.append({
'node': node,
'name': wnid_to_name(wnid),
'prob': prob_child
})
path_wnids.append(wnid)
cls = wnid_to_class.get(wnid, None)
pred = -1 if cls is None else classes.index(cls)
preds.append(pred)
decisions.append(decision)
pth = []
for element in path_wnids:
pth.append(tr_wn.index(element))
path_inds.append(pth)
# print("Tree: ",tree)
# print("Names: ",names)
# print("Path Indices: ",path_inds)
return torch.Tensor(preds).long(), decisions, tree, names, path_inds