Connectivity datasets
The connectivity datasets are collections of graphs that represent the connectivity of a set of networks.
There are two ways to work with the connectivity datasets. The first is to use the ConnectivityDataset class.
This class allows us to easily load weight matrices from a directory that contains a set of torch or numpy files,
and to convert them to a PyTorch Geometric dataset. This is useful if we have some predesigned connectivity matrices
that we want to use in our experiments.
The second way to work with the connectivity datasets is to use one of the ConnectivityGenerator classes
that are provided in the package. These classes can be used to generate random connectivity matrices from a couple common
distributions, save them to disk and load them into a PyTorch Geometric dataset. It is also possible to generate them
on the fly, without saving them, by using the generate() method. This is useful if we want to generate
some example connectivity matrices to use in our experiments.
Common for both ways is that the connectivity matrices are represented as a list of PyTorch Geometric graphs. Each graph
represents the connectivity of a single network. This includes a edge_index tensor of shape [2, num_edges]
that contains the indices of the connected nodes, and a W0 tensor of shape [num_edges,] that contains
the weights of the connections.
One of the main benefits of the list of torch_geometric.data.Data format is that it is easy to convert them
to a torch_geometric.data.Batch object by using the data loader torch_geometric.loader.DataLoader.
This is useful since it is often much faster in total to simulate a batch of networks than to simulate them one by one.
Dataset classes
A dataset of connectivity matrices for networks of neurons. |
Dataset generators
Generates a dataset of connectivity matrices W0 with an equal number of excitatory and inhibitory neurons with a normal distribution of weights. |
|
A dataset of connectivity matrices for networks of neurons, where each matrix is generated from a uniform distribution over the range [low, high] and then sparsified to the specified sparsity. |
|
A dataset of connectivity matrices for ring-networks of neurons, with weights generated from a Mexican hat distribution. |