update for keras 3.0

changelog.md

@@ -1,9 +1,19 @@
 v4.0.0
 
-* Reworked training scripts to have a single ``train_graph.py`` script. Command line arguments are now optional and just used for verification, but `category` to select a model/hyperparameter combination from hyper file.
+Completely reworked version of kgcnn for Keras 3.0 and multi-backend support. A lot of fundamental changes have been made.
+However, we tried to keep as much of the API from kgcnn 3.0 so that models in literature can be used with minimal changes.
+Mainly, the ``"input_tensor_type"="ragged"`` model parameter has to be added if ragged tensors are used as input in tensorflow.
+The scope of models has been reduced for initial release but will be extended in upcoming versions.
+
+* Reworked training scripts to have a single ``train_graph.py`` script. Command line arguments are now optional and just used for verification, all but `category` which has to select a model/hyperparameter combination from hyper file.
 Since the hyperparameter file already contains all necessary information.
 * Train test indices can now also be set and loaded from the dataset directly.
 * Scaler behaviour has changed with regard to `transform_dataset`. Key names of properties to transform has been moved to the constructor!
+Also be sure to check ``StandardLabelScaler`` if you want to scale regression targets, since target properties are default here.
+* Literature Models have an optional output scaler from new `kgcnn.layers.scale` layer controlled by `output_scaling` model argument.
+* Input embedding in literature models is now controlled with separate ``input_node_embedding`` or ``input_edge_embedding`` arguments which can be set to `None` for no embedding.
+Also embedding input tokens must be of dtype int now. No autocasting from float anymore.
+*  
 
 
 

kgcnn/literature/AttentiveFP/_make.py

@@ -36,15 +36,16 @@
     ],
     "input_tensor_type": "padded",
     "cast_disjoint_kwargs": {},
-    "input_embedding": None,
+    "input_embedding": None,  # deprecated
     "input_node_embedding": {"input_dim": 95, "output_dim": 64},
     "input_edge_embedding": {"input_dim": 5, "output_dim": 64},
     "attention_args": {"units": 32},
     "depthmol": 2,
     "depthato": 2,
     "dropout": 0.1,
     "verbose": 10,
-    "output_embedding": "graph", "output_to_tensor": True,
+    "output_embedding": "graph",
+    "output_to_tensor": True,  # deprecated
     "output_tensor_type": "padded",
     "output_mlp": {"use_bias": [True, True, False], "units": [25, 10, 1],
                    "activation": ["relu", "relu", "sigmoid"]}
@@ -57,36 +58,37 @@ def make_model(inputs: list = None,
                input_tensor_type: str = None,
                input_node_embedding: dict = None,
                input_edge_embedding: dict = None,
-               input_embedding: dict = None,
+               input_embedding: dict = None,  # noqa
                depthmol: int = None,
                depthato: int = None,
                dropout: float = None,
                attention_args: dict = None,
                name: str = None,
-               verbose: int = None,
+               verbose: int = None,  # noqa
                output_embedding: str = None,
-               output_to_tensor: bool = None,
+               output_to_tensor: bool = None,  # noqa
                output_tensor_type: str = None,
                output_scaling: dict = None,
                output_mlp: dict = None
                ):
-    r"""Make `AttentiveFP <https://doi.org/10.1021/acs.jmedchem.9b00959>`_ graph network via functional API.
+    r"""Make `AttentiveFP <https://doi.org/10.1021/acs.jmedchem.9b00959>`__ graph network via functional API.
     Default parameters can be found in :obj:`kgcnn.literature.AttentiveFP.model_default`.
 
     Model inputs:
     Model uses the list template of inputs and standard output template.
     The supported inputs are  :obj:`[nodes, edges, edge_indices, ...]`
     with '...' indicating mask or id tensors following the template below:
+
     %s
 
     Model outputs:
     The standard output template:
-    %s
 
+    %s
 
     Args:
-        inputs (list): List of dictionaries unpacked in :obj:`tf.keras.layers.Input`. Order must match model definition.
-        cast_disjoint_kwargs (dict): Dictionary of arguments for :obj:`CastBatchedIndicesToDisjoint` .
+        inputs (list): List of dictionaries unpacked in :obj:`Input`. Order must match model definition.
+        cast_disjoint_kwargs (dict): Dictionary of arguments for casting layers if used.
         input_tensor_type (str): Input type of graph tensor. Default is "padded".
         input_embedding (dict): Deprecated in favour of `input_node_embedding` etc.
         input_node_embedding (dict): Dictionary of arguments for nodes unpacked in :obj:`Embedding` layers.
@@ -119,8 +121,8 @@ def make_model(inputs: list = None,
     # Wrapping disjoint model.
     out = model_disjoint(
         [n, ed, disjoint_indices, batch_id_node, count_nodes],
-        use_node_embedding=len(inputs[0]['shape']) < 2,
-        use_edge_embedding=len(inputs[1]['shape']) < 2,
+        use_node_embedding=("int" in inputs[0]['dtype']) if input_node_embedding is not None else False,
+        use_edge_embedding=("int" in inputs[1]['dtype']) if input_edge_embedding is not None else False,
         input_node_embedding=input_node_embedding,
         input_edge_embedding=input_edge_embedding,
         depthmol=depthmol,
@@ -154,4 +156,4 @@ def set_scale(*args, **kwargs):
     return model
 
 
-make_model.__doc__ = make_model.__doc__ % (template_cast_list_input.__doc__, template_cast_output.__doc__)
+make_model.__doc__ = make_model.__doc__ % (template_cast_list_input.__doc__, template_cast_output.__doc__)

kgcnn/literature/DMPNN/_make.py

@@ -27,8 +27,8 @@
 model_default = {
     "name": "DMPNN",
     "inputs": [
-        {"shape": (None,), "name": "node_attributes", "dtype": "float32"},
-        {"shape": (None,), "name": "edge_attributes", "dtype": "float32"},
+        {"shape": (None,), "name": "node_number", "dtype": "int64"},
+        {"shape": (None,), "name": "edge_number", "dtype": "int64"},
         {"shape": (None, 2), "name": "edge_indices", "dtype": "int64"},
         {"shape": (None, 1), "name": "edge_indices_reverse", "dtype": "int64"},
         {"shape": (), "name": "total_nodes", "dtype": "int64"},
@@ -48,7 +48,8 @@
     "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_to_tensor": True,
+    "output_embedding": "graph",
+    "output_to_tensor": None,  # deprecated
     "output_tensor_type": "padded",
     "output_mlp": {"use_bias": [True, True, False], "units": [64, 32, 1],
                    "activation": ["relu", "relu", "linear"]},
@@ -61,7 +62,7 @@ def make_model(name: str = None,
                inputs: list = None,
                input_tensor_type: str = None,
                cast_disjoint_kwargs: dict = None,
-               input_embedding: dict = None,
+               input_embedding: dict = None,  # noqa
                input_node_embedding: dict = None,
                input_edge_embedding: dict = None,
                input_graph_embedding: dict = None,
@@ -72,41 +73,36 @@ def make_model(name: str = None,
                node_dense: dict = None,
                dropout: dict = None,
                depth: int = None,
-               verbose: int = None,
+               verbose: int = None,  # noqa
                use_graph_state: bool = False,
                output_embedding: str = None,
-               output_to_tensor: bool = None,
+               output_to_tensor: bool = None,  # noqa
                output_tensor_type: str = None,
                output_mlp: dict = None,
                output_scaling: dict = None
                ):
     r"""Make `DMPNN <https://pubs.acs.org/doi/full/10.1021/acs.jcim.9b00237>`__ graph network via functional API.
     Default parameters can be found in :obj:`kgcnn.literature.DMPNN.model_default`.
 
-    Inputs:
-        list: `[node_attributes, edge_attributes, edge_indices, edge_pairs, total_nodes, total_edges]` or
-        `[node_attributes, edge_attributes, edge_indices, edge_pairs, total_nodes, total_edges, state_attributes]`
-        if `use_graph_state=True` .
-
-            - node_attributes (Tensor): Node attributes of shape `(batch, None, F)` or `(batch, None)`
-              using an embedding layer.
-            - edge_attributes (Tensor): Edge attributes of shape `(batch, None, F)` or `(batch, None)`
-              using an embedding layer.
-            - edge_indices (Tensor): Index list for edges of shape `(batch, None, 2)`.
-            - edge_pairs (Tensor): Pair mappings for reverse edge for each edge `(batch, None, 1)`.
-            - state_attributes (Tensor): Environment or graph state attributes of shape `(batch, F)` or `(batch,)`
-              using an embedding layer.
-            - total_nodes(Tensor): Number of Nodes in graph of shape `(batch, )` .
-            - total_edges(Tensor): Number of Edges in graph of shape `(batch, )` .
-
-    Outputs:
-        Tensor: Graph embeddings of shape `(batch, L)` if :obj:`output_embedding="graph"`.
+    Model inputs:
+    Model uses the list template of inputs and standard output template.
+    The supported inputs are  :obj:`[nodes, edges, edge_indices, reverse_indices, (graph_state), ...]`
+    with '...' indicating mask or id tensors following the template below.
+    Here, reverse indices are in place of angle indices and refer to edges. The graph state is optional and controlled
+    by `use_graph_state` parameter.
+
+    %s
+
+    Model outputs:
+    The standard output template:
+
+    %s
 
     Args:
         name (str): Name of the model. Should be "DMPNN".
-        inputs (list): List of dictionaries unpacked in :obj:`keras.layers.Input`. Order must match model definition.
+        inputs (list): List of dictionaries unpacked in :obj:`Input`. Order must match model definition.
         input_tensor_type (str): Input type of graph tensor. Default is "padded".
-        cast_disjoint_kwargs (dict): Dictionary of arguments for :obj:`CastBatchedIndicesToDisjoint` .
+        cast_disjoint_kwargs (dict): Dictionary of arguments for casting layers if used.
         input_embedding (dict): Deprecated in favour of input_node_embedding etc.
         input_node_embedding (dict): Dictionary of arguments for nodes unpacked in :obj:`Embedding` layers.
         input_edge_embedding (dict): Dictionary of arguments for edge unpacked in :obj:`Embedding` layers.
@@ -122,7 +118,7 @@ def make_model(name: str = None,
         verbose (int): Level for print information.
         use_graph_state (bool): Whether to use graph state information. Default is False.
         output_embedding (str): Main embedding task for graph network. Either "node", "edge" or "graph".
-        output_to_tensor (bool): Whether to cast model output to :obj:`Tensor`.
+        output_to_tensor (bool): WDeprecated in favour of `output_tensor_type` .
         output_tensor_type (str): Output type of graph tensors such as nodes or edges. Default is "padded".
         output_mlp (dict): Dictionary of layer arguments unpacked in the final classification :obj:`MLP` layer block.
             Defines number of model outputs and activation.
@@ -150,10 +146,12 @@ def make_model(name: str = None,
     # Wrapping disjoint model.
     out = model_disjoint(
         [n, ed, edi, batch_id_node, e_pairs, count_nodes, gs],
-        use_node_embedding=len(inputs[0]['shape']) < 2, use_edge_embedding=len(inputs[1]['shape']) < 2,
-        use_graph_embedding=len(inputs[7]["shape"]) < 1 if use_graph_state else False,
+        use_node_embedding=("int" in inputs[0]['dtype']) if input_node_embedding is not None else False,
+        use_edge_embedding=("int" in inputs[1]['dtype']) if input_edge_embedding is not None else False,
+        use_graph_embedding=("int" in inputs[4]['dtype']) if input_graph_embedding is not None else False,
         input_node_embedding=input_node_embedding,
-        input_edge_embedding=input_edge_embedding, input_graph_embedding=input_graph_embedding,
+        input_edge_embedding=input_edge_embedding,
+        input_graph_embedding=input_graph_embedding,
         pooling_args=pooling_args, edge_initialize=edge_initialize, edge_activation=edge_activation,
         node_dense=node_dense, dropout=dropout, depth=depth, use_graph_state=use_graph_state,
         output_embedding=output_embedding, output_mlp=output_mlp, edge_dense=edge_dense
@@ -180,3 +178,6 @@ def set_scale(*args, **kwargs):
         setattr(model, "set_scale", set_scale)
 
     return model
+
+
+make_model.__doc__ = make_model.__doc__ % (template_cast_list_input.__doc__, template_cast_output.__doc__)

kgcnn/literature/GAT/_make.py

@@ -24,8 +24,8 @@
 model_default = {
     "name": "GAT",
     "inputs": [
-        {"shape": (None,), "name": "node_attributes", "dtype": "float32"},
-        {"shape": (None,), "name": "edge_attributes", "dtype": "float32"},
+        {"shape": (None,), "name": "node_number", "dtype": "int64"},
+        {"shape": (None,), "name": "edge_number", "dtype": "int64"},
         {"shape": (None, 2), "name": "edge_indices", "dtype": "int64"},
         {"shape": (), "name": "total_nodes", "dtype": "int64"},
         {"shape": (), "name": "total_edges", "dtype": "int64"}
@@ -34,15 +34,16 @@
     "cast_disjoint_kwargs": {},
     "input_embedding": None,  # deprecated
     "input_node_embedding": {"input_dim": 95, "output_dim": 64},
-    "input_edge_embedding": {"input_dim": 5, "output_dim": 64},
+    "input_edge_embedding": {"input_dim": 10, "output_dim": 64},
     "attention_args": {"units": 32, "use_final_activation": False, "use_edge_features": True,
                        "has_self_loops": True, "activation": "kgcnn>leaky_relu", "use_bias": True},
     "pooling_nodes_args": {"pooling_method": "scatter_mean"},
-    "depth": 3, "attention_heads_num": 5,
+    "depth": 3,
+    "attention_heads_num": 5,
     "attention_heads_concat": False,
     "verbose": 10,
     "output_embedding": "graph",
-    "output_to_tensor": True,
+    "output_to_tensor": None,  # deprecated
     "output_tensor_type": "padded",
     "output_mlp": {"use_bias": [True, True, False], "units": [25, 10, 1],
                    "activation": ["relu", "relu", "sigmoid"]},
@@ -54,7 +55,7 @@
 def make_model(inputs: list = None,
                input_tensor_type: str = None,
                cast_disjoint_kwargs: dict = None,
-               input_embedding: dict = None,
+               input_embedding: dict = None,  # noqa
                input_node_embedding: dict = None,
                input_edge_embedding: dict = None,
                attention_args: dict = None,
@@ -63,14 +64,14 @@ def make_model(inputs: list = None,
                attention_heads_num: int = None,
                attention_heads_concat: bool = None,
                name: str = None,
-               verbose: int = None,
+               verbose: int = None,  # noqa
                output_embedding: str = None,
-               output_to_tensor: bool = None,
+               output_to_tensor: bool = None,  # noqa
                output_mlp: dict = None,
                output_scaling: dict = None,
                output_tensor_type: str = None,
                ):
-    r"""Make `GAT <https://arxiv.org/abs/1710.10903>`_ graph network via functional API.
+    r"""Make `GAT <https://arxiv.org/abs/1710.10903>`__ graph network via functional API.
     Default parameters can be found in :obj:`kgcnn.literature.GAT.model_default`.
 
     Model inputs:
@@ -85,10 +86,9 @@ def make_model(inputs: list = None,
 
     %s
 
-
     Args:
-        inputs (list): List of dictionaries unpacked in :obj:`tf.keras.layers.Input`. Order must match model definition.
-        cast_disjoint_kwargs (dict): Dictionary of arguments for :obj:`CastBatchedIndicesToDisjoint` .
+        inputs (list): List of dictionaries unpacked in :obj:`Input`. Order must match model definition.
+        cast_disjoint_kwargs (dict): Dictionary of arguments for casting layers if used.
         input_tensor_type (str): Input type of graph tensor. Default is "padded".
         input_embedding (dict): Deprecated in favour of input_node_embedding etc.
         input_node_embedding (dict): Dictionary of arguments for nodes unpacked in :obj:`Embedding` layers.
@@ -108,7 +108,7 @@ def make_model(inputs: list = None,
         output_scaling (dict): Dictionary of layer arguments unpacked in scaling layers.
 
     Returns:
-        :obj:`ks.models.Model`
+        :obj:`keras.models.Model`
     """
     # Make input
     model_inputs = [Input(**x) for x in inputs]
@@ -121,8 +121,10 @@ def make_model(inputs: list = None,
     # Wrapping disjoint model.
     out = model_disjoint(
         [n, ed, disjoint_indices, batch_id_node, count_nodes],
-        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,
+        use_node_embedding=("int" in inputs[0]['dtype']) if input_node_embedding is not None else False,
+        use_edge_embedding=("int" in inputs[1]['dtype']) if input_edge_embedding is not None else False,
+        input_node_embedding=input_node_embedding,
+        input_edge_embedding=input_edge_embedding,
         attention_args=attention_args, pooling_nodes_args=pooling_nodes_args, depth=depth,
         attention_heads_num=attention_heads_num, attention_heads_concat=attention_heads_concat,
         output_embedding=output_embedding, output_mlp=output_mlp
@@ -151,4 +153,4 @@ def set_scale(*args, **kwargs):
     return model
 
 
-make_model.__doc__ = make_model.__doc__ % (template_cast_list_input.__doc__, template_cast_output.__doc__)
+make_model.__doc__ = make_model.__doc__ % (template_cast_list_input.__doc__, template_cast_output.__doc__)