Refactor TGB Mock in Tests

[Jacob-Chmura]

Purpose

The purpose of this PR is to refactor the TGB mock dataset objects in our DGData.from_tgb unit tests.

Key Changes

• I did catch a bug in the from_tgb when working with NodeProp datasets, which occurs if there are no node events within a dataset split (e.g. no node features in validation set):

I guarded this accordingly:

tgm/tgm/data.py

Lines 415 to 438 in 968ed03

	if len(node_label_dict):
	# Node events could be missing from the current data split (e.g. validation)
	num_node_events = 0
	node_label_dim = 0
	for t in node_label_dict:
	for node_id, label in node_label_dict[t].items():
	num_node_events += 1
	node_label_dim = label.shape[0]
	temp_node_timestamps = np.zeros(num_node_events, dtype=np.int64)
	temp_node_ids = np.zeros(num_node_events, dtype=np.int64)
	temp_dynamic_node_feats = np.zeros(
	(num_node_events, node_label_dim), dtype=np.float32
	)
	idx = 0
	for t in node_label_dict:
	for node_id, label in node_label_dict[t].items():
	temp_node_timestamps[idx] = t
	temp_node_ids[idx] = node_id
	temp_dynamic_node_feats[idx] = label
	idx += 1
	node_timestamps = torch.from_numpy(temp_node_timestamps).long()
	node_ids = torch.from_numpy(temp_node_ids).long()
	dynamic_node_feats = torch.from_numpy(temp_dynamic_node_feats).float()

Relevant Prs

Close #70

[shenyangHuang]

nice and clean

[shenyangHuang]

nice