Merge pull request #181 from pyt-team/frantzen-typing

More concrete types throughout the project

topomodelx/base/aggregation.py

@@ -9,10 +9,10 @@ class Aggregation(torch.nn.Module):
 
     Parameters
     ----------
-    aggr_func : string
+    aggr_func : Literal["mean", "sum"], default="sum"
         Aggregation method.
         (Inter-neighborhood).
-    update_func : string
+    update_func : Literal["relu", "sigmoid", "tanh", None], default="sigmoid"
         Update method to apply to merged message.
     """
 

topomodelx/base/conv.py

@@ -1,4 +1,5 @@
 """Convolutional layer for message passing."""
+from typing import Literal
 
 import torch
 from torch.nn.parameter import Parameter
@@ -20,11 +21,11 @@ class Conv(MessagePassing):
         Dimension of output features.
     aggr_norm : bool, default=False
         Whether to normalize the aggregated message by the neighborhood size.
-    update_func : string, optional
+    update_func : Literal["relu", "sigmoid"], optional
         Update method to apply to message.
     att : bool, default=False
         Whether to use attention.
-    initialization : string
+    initialization : Literal["xavier_uniform", "xavier_normal"], default="xavier_uniform"
         Initialization method.
     with_linear_transform: bool
         Whether to apply a learnable linear transform.
@@ -36,9 +37,10 @@ def __init__(
         in_channels,
         out_channels,
         aggr_norm: bool = False,
-        update_func=None,
+        update_func: Literal["relu", "sigmoid"] | None = None,
         att: bool = False,
-        initialization: str = "xavier_uniform",
+        initialization: Literal["xavier_uniform", "xavier_normal"] = "xavier_uniform",
+        initialization_gain: float = 1.414,
         with_linear_transform: bool = True,
     ) -> None:
         super().__init__(

topomodelx/base/message_passing.py

@@ -1,5 +1,5 @@
 """Message passing module."""
-
+from typing import Literal
 
 import torch
 
@@ -23,11 +23,11 @@ class MessagePassing(torch.nn.Module):
 
     Parameters
     ----------
-    aggr_func : string
+    aggr_func : Literal["sum", "mean", "add"], default="sum"
         Aggregation function to use.
     att : bool, default=False
         Whether to use attention.
-    initialization : string
+    initialization : Literal["xavier_uniform", "xavier_normal"], default="xavier_uniform"
         Initialization method for the weights of the layer.
 
     References
@@ -43,9 +43,10 @@ class MessagePassing(torch.nn.Module):
 
     def __init__(
         self,
-        aggr_func: str = "sum",
+        aggr_func: Literal["sum", "mean", "add"] = "sum",
         att: bool = False,
-        initialization: str = "xavier_uniform",
+        initialization: Literal["xavier_uniform", "xavier_normal"] = "xavier_uniform",
+        initialization_gain: float = 1.414,
     ) -> None:
         super().__init__()
         self.aggr_func = aggr_func
@@ -113,7 +114,7 @@ def message(self, x_source, x_target=None):
     def attention(self, x_source, x_target=None):
         """Compute attention weights for messages.
 
-        This provides a default attention function to the message passing scheme.
+        This provides a default attention function to the message-passing scheme.
 
         Alternatively, users can subclass MessagePassing and overwrite
         the attention method in order to replace it with their own attention mechanism.

topomodelx/nn/cell/can_layer.py

@@ -1,6 +1,6 @@
 """Cell Attention Network layer."""
 
-from typing import Callable
+from typing import Callable, Literal
 
 import torch
 from torch import Tensor, nn, topk
@@ -396,9 +396,9 @@ class MultiHeadCellAttention(MessagePassing):
         Activation function to use for the attention weights.
     add_self_loops : bool, optional
         Whether to add self-loops to the adjacency matrix.
-    aggr_func : string, optional
-        Aggregation function to use. Options are "sum", "mean", "max".
-    initialization : string, optional
+    aggr_func : Literal["sum", "mean", "add"], default="sum"
+        Aggregation function to use.
+    initialization : Literal["xavier_uniform", "xavier_normal"], default="xavier_uniform"
         Initialization method for the weights of the layer.
 
     Notes
@@ -423,14 +423,10 @@ def __init__(
         concat: bool,
         att_activation: torch.nn.Module,
         add_self_loops: bool = False,
-        aggr_func: str = "sum",
-        initialization: str = "xavier_uniform",
+        aggr_func: Literal["sum", "mean", "add"] = "sum",
+        initialization: Literal["xavier_uniform", "xavier_normal"] = "xavier_uniform",
     ) -> None:
-        super().__init__(
-            att=True,
-            initialization=initialization,
-            aggr_func=aggr_func,
-        )
+        super().__init__(aggr_func=aggr_func, att=True, initialization=initialization)
 
         self.in_channels = in_channels
         self.out_channels = out_channels
@@ -587,9 +583,9 @@ class MultiHeadCellAttention_v2(MessagePassing):
         Activation function to use for the attention weights.
     add_self_loops : bool, optional
         Whether to add self-loops to the adjacency matrix.
-    aggr_func : string, optional
-        Aggregation function to use. Options are "sum", "mean", "max".
-    initialization : string, optional
+    aggr_func : Literal["sum", "mean", "add"], default="sum"
+        Aggregation function to use.
+    initialization : Literal["xavier_uniform", "xavier_normal"], default="xavier_uniform"
         Initialization method for the weights of the layer.
     share_weights : bool, optional
         Whether to share the weights between the attention heads.
@@ -616,14 +612,14 @@ def __init__(
         concat: bool,
         att_activation: torch.nn.Module,
         add_self_loops: bool = False,
-        aggr_func: str = "sum",
-        initialization: str = "xavier_uniform",
+        aggr_func: Literal["sum", "mean", "add"] = "sum",
+        initialization: Literal["xavier_uniform", "xavier_normal"] = "xavier_uniform",
         share_weights: bool = False,
     ) -> None:
         super().__init__(
+            aggr_func=aggr_func,
             att=True,
             initialization=initialization,
-            aggr_func=aggr_func,
         )
 
         self.in_channels = in_channels
@@ -797,28 +793,31 @@ class CANLayer(torch.nn.Module):
         Dimension of input features on n-cells.
     out_channels : int
         Dimension of output
-    heads : int, optional
-        Number of attention heads, by default 1
+    heads : int, default=1
+        Number of attention heads
     dropout : float, optional
-        Dropout probability of the normalized attention coefficients, by default 0.0
-    concat : bool, optional
-        If True, the output of each head is concatenated. Otherwise, the output of each head is averaged, by default True
-    skip_connection : bool, optional
-        If True, skip connection is added, by default True
+        Dropout probability of the normalized attention coefficients.
+    concat : bool, default=True
+        If True, the output of each head is concatenated. Otherwise, the output of each head is averaged.
+    skip_connection : bool, default=True
+        If True, skip connection is added.
     add_self_loops : bool, optional
-        If True, self-loops are added to the neighborhood matrix, by default False
-    att_activation : Callable, optional
-        Activation function applied to the attention coefficients, by default torch.nn.LeakyReLU()
-    aggr_func : str, optional
-        Between-neighborhood aggregation function applied to the messages, by default "sum"
-    update_func : str, optional
-        Update function applied to the messages, by default "relu"
-    version : str, optional
+        If True, self-loops are added to the neighborhood matrix.
+    att_activation : Callable, default=torch.nn.LeakyReLU()
+        Activation function applied to the attention coefficients.
+    aggr_func : Literal["mean", "sum"], default="sum"
+        Between-neighborhood aggregation function applied to the messages.
+    update_func : Literal["relu", "sigmoid", "tanh", None], default="relu"
+        Update function applied to the messages.
+    version : Literal["v1", "v2"], default="v1"
         Version of the layer, by default "v1" which is the same as the original CAN layer. While "v2" has the same attetion mechanism as the GATv2 layer.
-    share_weights : bool, optional
+    share_weights : bool, default=False
         This option is valid only for "v2". If True, the weights of the linear transformation applied to the source and target features are shared, by default False
     """
 
+    lower_att: MultiHeadCellAttention | MultiHeadCellAttention_v2
+    upper_att: MultiHeadCellAttention | MultiHeadCellAttention_v2
+
     def __init__(
         self,
         in_channels: int,
@@ -829,9 +828,9 @@ def __init__(
         skip_connection: bool = True,
         att_activation: torch.nn.Module = torch.nn.LeakyReLU(),
         add_self_loops: bool = False,
-        aggr_func: str = "sum",
-        update_func: str = "relu",
-        version: str = "v1",
+        aggr_func: Literal["mean", "sum"] = "sum",
+        update_func: Literal["relu", "sigmoid", "tanh"] | None = "relu",
+        version: Literal["v1", "v2"] = "v1",
         share_weights: bool = False,
         **kwargs,
     ) -> None:
@@ -841,7 +840,7 @@ def __init__(
         assert out_channels > 0, ValueError("Number of output channels must be > 0")
         assert heads > 0, ValueError("Number of heads must be > 0")
         assert dropout >= 0.0 and dropout <= 1.0, ValueError("Dropout must be in [0,1]")
-        assert version in ["v1", "v2"], ValueError("Version must be 'v1' or 'v2'")
+
         # assert that shared weight is True only if version is v2
         assert share_weights is False or version == "v2", ValueError(
             "Shared weights is valid only for v2"

topomodelx/nn/cell/can_layer_bis.py

@@ -1,4 +1,6 @@
 """Cellular Attention Network Layer."""
+from typing import Literal
+
 import torch
 from torch.nn.parameter import Parameter
 
@@ -50,23 +52,24 @@ class CANLayer(torch.nn.Module):
     ----------
     channels : int
         Dimension of input features on edges (1-cells).
-    activation : string
-        Activation function to apply to merged message
+    activation : Literal["relu", "sigmoid", "tanh", None], default="sigmoid"
+        Activation function to apply to merged message.
     att : bool
         Whether to use attention.
     eps : float
         Epsilon used in the attention mechanism.
-    initialization : string
+    initialization : Literal["xavier_uniform", "xavier_normal"], default="xavier_uniform"
         Initialization method.
     """
 
     def __init__(
         self,
         channels,
-        activation: str = "sigmoid",
+        activation: Literal["relu", "sigmoid", "tanh"] | None = "sigmoid",
         att: bool = True,
         eps: float = 1e-5,
-        initialization: str = "xavier_uniform",
+        initialization: Literal["xavier_uniform", "xavier_normal"] = "xavier_uniform",
+        initialization_gain: float = 1.414,
     ) -> None:
         super().__init__()
         # Do I need upper and lower convolution layers? Since I think they will have different parameters

topomodelx/nn/hypergraph/allset_layer.py

@@ -135,6 +135,9 @@ class AllSetBlock(nn.Module):
         Type of layer normalization in the MLP.
     """
 
+    encoder: nn.Module
+    decoder: nn.Module
+
     def __init__(
         self,
         in_channels,
@@ -241,7 +244,7 @@ def __init__(
         bias: bool = False,
     ) -> None:
         params = {} if inplace is None else {"inplace": inplace}
-        layers = []
+        layers: list[nn.Module] = []
         in_dim = in_channels
         for hidden_dim in hidden_channels[:-1]:
             layers.append(nn.Linear(in_dim, hidden_dim, bias=bias))

topomodelx/nn/hypergraph/allset_transformer_layer.py

@@ -1,4 +1,6 @@
 """AllSetTransformer Layer Module."""
+from typing import Literal
+
 import torch
 import torch.nn.functional as F
 from torch import nn
@@ -273,7 +275,7 @@ class MultiHeadAttention(MessagePassing):
         Number of attention heads.
     number_queries : int, default=1
         Number of queries.
-    initialization : str, default="xavier_uniform"
+    initialization : Literal["xavier_uniform", "xavier_normal"], default="xavier_uniform"
         Initialization method.
     """
 
@@ -285,7 +287,8 @@ def __init__(
         update_func=None,
         heads: int = 4,
         number_queries: int = 1,
-        initialization: str = "xavier_uniform",
+        initialization: Literal["xavier_uniform", "xavier_normal"] = "xavier_uniform",
+        initialization_gain: float = 1.414,
     ) -> None:
         super().__init__(
             att=True,
@@ -429,7 +432,7 @@ def __init__(
         bias: bool = False,
     ) -> None:
         params = {} if inplace is None else {"inplace": inplace}
-        layers = []
+        layers: list[nn.Module] = []
         in_dim = in_channels
         for hidden_dim in hidden_channels[:-1]:
             layers.append(nn.Linear(in_dim, hidden_dim, bias=bias))

topomodelx/nn/hypergraph/hmpnn_layer.py

@@ -1,4 +1,6 @@
 """HMPNN (Hypergraph Message Passing Neural Network) Layer introduced in Heydari et Livi 2022."""
+from typing import Literal
+
 import torch
 from torch import nn
 from torch.nn import functional as F
@@ -21,7 +23,10 @@ def apply_dropout(self, neighborhood, dropout_rate: float):
 
 class _NodeToHyperedgeMessenger(MessagePassing, _AdjacencyDropoutMixin):
     def __init__(
-        self, messaging_func, adjacency_dropout: float = 0.7, aggr_func: str = "sum"
+        self,
+        messaging_func,
+        adjacency_dropout: float = 0.7,
+        aggr_func: Literal["sum", "mean", "add"] = "sum",
     ) -> None:
         super().__init__(aggr_func)
         self.messaging_func = messaging_func
@@ -46,7 +51,7 @@ def __init__(
         self,
         messaging_func,
         adjacency_dropout: float = 0.7,
-        aggr_func: str = "sum",
+        aggr_func: Literal["sum", "mean", "add"] = "sum",
     ) -> None:
         super().__init__(aggr_func)
         self.messaging_func = messaging_func
@@ -116,8 +121,8 @@ class HMPNNLayer(nn.Module):
         to the paper.
     adjacency_dropout: 0.7
         Adjacency dropout rate.
-    aggr_func: "sum"
-        Message aggregation function. A value among "sum", "mean" and "add".
+    aggr_func: Literal["sum", "mean", "add"], default="sum"
+        Message aggregation function.
     updating_dropout: 0.5
         Regular dropout rate applied to node and hyperedge features.
     updating_func: None
@@ -132,7 +137,7 @@ def __init__(
         node_to_hyperedge_messaging_func=None,
         hyperedge_to_node_messaging_func=None,
         adjacency_dropout: float = 0.7,
-        aggr_func: str = "sum",
+        aggr_func: Literal["sum", "mean", "add"] = "sum",
         updating_dropout: float = 0.5,
         updating_func=None,
     ) -> None:

topomodelx/nn/hypergraph/hnhn_layer.py

@@ -1,4 +1,6 @@
 """Template Layer with two conv passing steps."""
+from typing import Literal
+
 import torch
 from torch.nn.parameter import Parameter
 
@@ -48,7 +50,7 @@ class HNHNLayer(torch.nn.Module):
         Scalar controlling the importance of node cardinality.
     bias_gain : float
         Gain for the bias initialization.
-    bias_init : string ["xavier_uniform"|"xavier_normal"]
+    bias_init : Literal["xavier_uniform", "xavier_normal"], default="xavier_uniform"
         Controls the bias initialization method.
     """
 
@@ -62,7 +64,7 @@ def __init__(
         alpha: float = -1.5,
         beta: float = -0.5,
         bias_gain: float = 1.414,
-        bias_init: str = "xavier_uniform",
+        bias_init: Literal["xavier_uniform", "xavier_normal"] = "xavier_uniform",
     ) -> None:
         super().__init__()
         self.use_bias = use_bias