Skip to content

Commit

Permalink
Merge remote-tracking branch 'origin/master'
Browse files Browse the repository at this point in the history
  • Loading branch information
@PatReis
PatReis committed Sep 28, 2023
1 parent cf30c10 commit aaf2a01
Show file tree
Hide file tree
Showing 15 changed files with 664 additions and 83 deletions.
2 changes: 1 addition & 1 deletion kgcnn/backend/_tensorflow.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -21,7 +21,7 @@ def scatter_reduce_max(indices, values, shape):
def scatter_reduce_mean(indices, values, shape):
indices = tf.expand_dims(indices, axis=-1)
counts = tf.scatter_nd(indices, tf.ones_like(values), shape)
return tf.scatter_nd(indices, values, shape)/counts
return tf.math.divide_no_nan(tf.scatter_nd(indices, values, shape), counts)


def scatter_reduce_softmax(indices, values, shape, normalize: bool = False):
Expand Down
2 changes: 1 addition & 1 deletion kgcnn/literature/GIN/_make.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -108,7 +108,7 @@ def make_model(inputs: list = None,

# Wrapping disjoint model.
out = model_disjoint(
inputs,
[n, disjoint_indices, batch_id_node, count_nodes],
use_node_embedding=len(inputs[0]['shape']) < 2, input_node_embedding=input_node_embedding,
depth=depth, gin_args=gin_args, gin_mlp=gin_mlp, last_mlp=last_mlp, dropout=dropout,
output_embedding=output_embedding, output_mlp=output_mlp
Expand Down
52 changes: 15 additions & 37 deletions kgcnn/literature/GraphSAGE/_make.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -30,14 +30,14 @@
model_default = {
'name': "GraphSAGE",
'inputs': [
{"shape": (None, ), "name": "node_attributes", "dtype": "float32"},
{"shape": (None, ), "name": "edge_attributes", "dtype": "float32"},
{"shape": (None,), "name": "node_attributes", "dtype": "float32"},
{"shape": (None,), "name": "edge_attributes", "dtype": "float32"},
{"shape": (None, 2), "name": "edge_indices", "dtype": "int64"},
{"shape": (), "name": "total_nodes", "dtype": "int64"},
{"shape": (), "name": "total_edges", "dtype": "int64"}
],
"cast_disjoint_kwargs": {},
"input_node_embedding": {"input_dim": 95, "output_dim": 64},
"input_node_embedding": {"input_dim": 95, "output_dim": 64},
"input_edge_embedding": {"input_dim": 5, "output_dim": 64},
'node_mlp_args': {"units": [100, 50], "use_bias": True, "activation": ['relu', "linear"]},
'edge_mlp_args': {"units": [100, 50], "use_bias": True, "activation": ['relu', "linear"]},
Expand Down Expand Up @@ -121,48 +121,25 @@ def make_model(inputs: list = None,
**cast_disjoint_kwargs)([batched_nodes, batched_indices, total_nodes, total_edges])
ed, _, _, _ = CastBatchedAttributesToDisjoint(**cast_disjoint_kwargs)([batched_edges, total_edges])

# Embedding, if no feature dimension
if len(inputs[0]['shape']) < 2:
n = Embedding(**input_node_embedding)(n)
if len(inputs[1]['shape']) < 2:
ed = Embedding(**input_edge_embedding)(ed)

for i in range(0, depth):

eu = GatherNodesOutgoing(**gather_args)([n, disjoint_indices])
if use_edge_features:
eu = Concatenate(**concat_args)([eu, ed])

eu = GraphMLP(**edge_mlp_args)([eu, batch_id_edge, count_edges])

# Pool message
if pooling_args['pooling_method'] in ["LSTM", "lstm"]:
nu = AggregateLocalEdgesLSTM(**pooling_args)([n, eu, disjoint_indices])
else:
nu = AggregateLocalEdges(**pooling_args)([n, eu, disjoint_indices]) # Summing for each node connection

nu = Concatenate(**concat_args)([n, nu]) # Concatenate node features with new edge updates

n = GraphMLP(**node_mlp_args)([nu, batch_id_node, count_nodes])

n = GraphLayerNormalization()([n, batch_id_node, count_nodes])
out = model_disjoint(
[n, ed, disjoint_indices, batch_id_node, batch_id_edge, count_nodes, count_edges],
use_node_embedding=len(inputs[0]['shape']) < 2, use_edge_embedding=len(inputs[1]['shape']) < 2,
input_node_embedding=input_node_embedding, input_edge_embedding=input_edge_embedding,
node_mlp_args=node_mlp_args, edge_mlp_args=edge_mlp_args, pooling_args=pooling_args,
pooling_nodes_args=pooling_nodes_args, gather_args=gather_args, concat_args=concat_args,
use_edge_features=use_edge_features, depth=depth, output_embedding=output_embedding,
output_mlp=output_mlp,
)

# Regression layer on output
if output_embedding == 'graph':
out = PoolingNodes(**pooling_nodes_args)([count_nodes, n, batch_id_node])
out = MLP(**output_mlp)(out)
out = CastDisjointToGraphState(**cast_disjoint_kwargs)(out)

elif output_embedding == 'node':
out = GraphMLP(**output_mlp)([n, batch_id_node, count_nodes])
if output_to_tensor:
out = CastDisjointToBatchedAttributes(**cast_disjoint_kwargs)([batched_nodes, out, batch_id_node, node_id])
else:
out = CastDisjointToGraphState(**cast_disjoint_kwargs)(out)

else:
raise ValueError("Unsupported output embedding for `GraphSAGE`")

if output_scaling is not None:
scaler = get_scaler(output_scaling["name"])(**output_scaling)
out = scaler(out)
Expand All @@ -173,6 +150,7 @@ def make_model(inputs: list = None,
if output_scaling is not None:
def set_scale(*args, **kwargs):
scaler.set_scale(*args, **kwargs)

setattr(model, "set_scale", set_scale)

return model
Expand All @@ -194,7 +172,7 @@ def model_disjoint(
depth: int = None,
output_embedding: str = None,
output_mlp: dict = None,
):
):
n, ed, disjoint_indices, batch_id_node, batch_id_edge, count_nodes, count_edges = inputs

# Embedding, if no feature dimension
Expand Down Expand Up @@ -231,4 +209,4 @@ def model_disjoint(
out = GraphMLP(**output_mlp)([n, batch_id_node, count_nodes])
else:
raise ValueError("Unsupported output embedding for `GraphSAGE`")
return out
return out
75 changes: 74 additions & 1 deletion training/hyper/hyper_clintox.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -41,7 +41,6 @@
"loss": "binary_crossentropy",
"metrics": ["binary_accuracy", {"class_name": "AUC", "config": {"name": "auc"}}]
},
"multi_target_indices": None
},
"dataset": {
"class_name": "ClinToxDataset",
Expand Down Expand Up @@ -393,4 +392,78 @@
"kgcnn_version": "4.0.0"
}
},
"DMPNN": {
"model": {
"class_name": "make_model",
"module_name": "kgcnn.literature.DMPNN",
"config": {
"name": "DMPNN",
"inputs": [
{"shape": (None, 41), "name": "node_attributes", "dtype": "float32"},
{"shape": (None, 11), "name": "edge_attributes", "dtype": "float32"},
{"shape": (None, 2), "name": "edge_indices", "dtype": "int64"},
{"shape": (None, 1), "name": "edge_indices_reverse", "dtype": "int64"},
{"shape": (), "name": "total_nodes", "dtype": "int64"},
{"shape": (), "name": "total_edges", "dtype": "int64"}
],
"cast_disjoint_kwargs": {},
"input_node_embedding": {"input_dim": 95, "output_dim": 64},
"input_edge_embedding": {"input_dim": 5, "output_dim": 64},
"input_graph_embedding": {"input_dim": 100, "output_dim": 64},
"pooling_args": {"pooling_method": "scatter_sum"},
"edge_initialize": {"units": 128, "use_bias": True, "activation": "relu"},
"edge_dense": {"units": 128, "use_bias": True, "activation": "linear"},
"edge_activation": {"activation": "relu"},
"node_dense": {"units": 128, "use_bias": True, "activation": "relu"},
"verbose": 10, "depth": 5,
"dropout": {"rate": 0.1},
"output_embedding": "graph",
"output_mlp": {
"use_bias": [True, True, False], "units": [64, 32, 1],
"activation": ["relu", "relu", "sigmoid"]
}
}
},
"training": {
"fit": {"batch_size": 32, "epochs": 50, "validation_freq": 1, "verbose": 2, "callbacks": []},
"compile": {
"optimizer": {
"class_name": "Adam",
"config": {
"learning_rate":
{"module": "keras_core.optimizers.schedules",
"class_name": "ExponentialDecay",
"config": {"initial_learning_rate": 0.001,
"decay_steps": 1600,
"decay_rate": 0.5, "staircase": False}}
}
},
# "loss": "kgcnn>BinaryCrossentropyNoNaN",
# "metrics": ["kgcnn>BinaryAccuracyNoNaN",
# {"class_name": "kgcnn>AUCNoNaN", "config": {"multi_label": True, "num_labels": 12}}],
# "metrics": ["kgcnn>BinaryAccuracyNoNaN", "kgcnn>AUCNoNaN"],
"loss": "binary_crossentropy",
"metrics": ["binary_accuracy", {"class_name": "AUC", "config": {"name": "auc"}}]
}
},
"dataset": {
"class_name": "ClinToxDataset",
"module_name": "kgcnn.data.datasets.ClinToxDataset",
"config": {},
"methods": [
{"set_attributes": {}},
{"set_train_test_indices_k_fold": {"n_splits": 5, "random_state": 42, "shuffle": True}},
{"map_list": {"method": "set_edge_indices_reverse"}},
{"map_list": {"method": "count_nodes_and_edges"}},
]
},
"data": {
"data_unit": "mol/L"
},
"info": {
"postfix": "",
"postfix_file": "",
"kgcnn_version": "4.0.0"
}
},
}
74 changes: 74 additions & 0 deletions training/hyper/hyper_cora_lu.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -328,4 +328,78 @@
"kgcnn_version": "4.0.0"
}
},
"DMPNN": {
"model": {
"class_name": "make_model",
"module_name": "kgcnn.literature.DMPNN",
"config": {
"name": "DMPNN",
"inputs": [
{"shape": [None, 1433], "name": "node_attributes", "dtype": "float32"},
{"shape": [None, 1], "name": "edge_weights", "dtype": "float32"},
{"shape": [None, 2], "name": "edge_indices", "dtype": "int64"},
{"shape": (None, 1), "name": "edge_indices_reverse", "dtype": "int64"},
{"shape": (), "name": "total_nodes", "dtype": "int64"},
{"shape": (), "name": "total_edges", "dtype": "int64"}
],
"cast_disjoint_kwargs": {},
"input_node_embedding": {"input_dim": 95, "output_dim": 64},
"input_edge_embedding": {"input_dim": 5, "output_dim": 64},
"input_graph_embedding": {"input_dim": 100, "output_dim": 64},
"pooling_args": {"pooling_method": "scatter_sum"},
"edge_initialize": {"units": 128, "use_bias": True, "activation": "relu"},
"edge_dense": {"units": 128, "use_bias": True, "activation": "linear"},
"edge_activation": {"activation": "relu"},
"node_dense": {"units": 128, "use_bias": True, "activation": "relu"},
"verbose": 10, "depth": 5,
"dropout": {"rate": 0.1},
"output_embedding": "node",
"output_mlp": {
"use_bias": [True, True, False], "units": [64, 32, 7],
"activation": ["relu", "relu", "softmax"]
}
}
},
"training": {
"cross_validation": {"class_name": "KFold",
"config": {"n_splits": 5, "random_state": 42, "shuffle": True}},
"multi_target_indices": None,
"fit": {"batch_size": 32, "epochs": 300, "validation_freq": 1, "verbose": 2, "callbacks": []},
"compile": {
"optimizer": {
"class_name": "Adam",
"config": {
"learning_rate":
{"module": "keras_core.optimizers.schedules",
"class_name": "ExponentialDecay",
"config": {"initial_learning_rate": 0.001,
"decay_steps": 1600,
"decay_rate": 0.5, "staircase": False}}
}
},
"loss": "categorical_crossentropy",
"weighted_metrics": ["categorical_accuracy", {"class_name": "AUC", "config": {"name": "auc"}}]
},
},
"dataset": {
"class_name": "CoraLuDataset",
"module_name": "kgcnn.data.datasets.CoraLuDataset",
"config": {},
"methods": [
{"map_list": {"method": "make_undirected_edges"}},
{"map_list": {"method": "add_edge_self_loops"}},
{"map_list": {"method": "normalize_edge_weights_sym"}},
{"map_list": {"method": "set_edge_indices_reverse"}},
{"map_list": {"method": "count_nodes_and_edges"}},
]
},
"data": {
"data_unit": ""
},
"info": {
"postfix": "",
"postfix_file": "",
"kgcnn_version": "4.0.0"
}
},
}
Loading

0 comments on commit aaf2a01

Please sign in to comment.