05ff22f2cdff1f560975d60c7371cedcf4dca201,bindsnet/learning/__init__.py,,post_pre,#Any#,6

Before Change

        | :code:`conn` (:code:`bindsnet.network.topology.AbstractConnection`):
        An instance of class :code:`AbstractConnection`.
    
    if not "kernel_size" in conn.__dict__:
        x_source, x_target = conn.source.x.unsqueeze(-1), conn.target.x.unsqueeze(0)
        s_source, s_target = conn.source.s.float().unsqueeze(-1), conn.target.s.float().unsqueeze(0)
        
        // Post-synaptic.
        conn.w += conn.nu_post * x_source * s_target
        // Pre-synaptic.
        conn.w -= conn.nu_pre * s_source * x_target

        // Bound weights.
        conn.w = torch.clamp(conn.w, conn.wmin, conn.wmax)
    else:
        out_channels, _, kernel_height, kernel_width = conn.w.size()
        padding, stride = conn.padding, conn.stride
        
        x_source = im2col_indices(conn.source.x,
                                  kernel_height,
                                  kernel_width,
                                  padding=padding,
                                  stride=stride)

        x_target = conn.target.x.permute(1, 2, 3, 0).reshape(out_channels,
                                                             -1)
        s_source = im2col_indices(conn.source.s,
                                  kernel_height,
                                  kernel_width,
                                  padding=padding,
                                  stride=stride).float()

        s_target = conn.target.s.permute(1, 2, 3, 0).reshape(out_channels,
                                                             -1).float()
        
        // Post-synaptic.
        post = s_target @ x_source.t()
        conn.w += conn.nu_post * post.view(conn.w.size())
        
        // Pre-synaptic.
        pre = x_target @ s_source.t()
        conn.w -= conn.nu_pre * pre.view(conn.w.size())

        // Bound weights.
        conn.w = torch.clamp(conn.w, conn.wmin, conn.wmax)

def hebbian(conn, **kwargs):
    

After Change

        
        | :param conn: An ``AbstractConnection`` object whose weights are to be modified by the post-pre STDP rule.
    
    if isinstance(conn, Connection):
        // Unpack / reshape quantities of interest (spikes and spike_traces).
        s_source = conn.source.s.float().unsqueeze(-1)
        s_target = conn.target.s.float().unsqueeze(0)
        x_source = conn.source.x.unsqueeze(-1)
        x_target = conn.target.x.unsqueeze(0)

        conn.w += conn.nu_post * x_source * s_target  // Post-synaptic.
        conn.w -= conn.nu_pre * s_source * x_target  // Pre-synaptic.
        conn.w = torch.clamp(conn.w, conn.wmin, conn.wmax)  // Bound weights.

    elif isinstance(conn, Conv2dConnection):
        // Get convolutional layer parameters.
        out_channels, _, kernel_height, kernel_width = conn.w.size()
        padding, stride = conn.padding, conn.stride

        // Unpack / reshape quantities of interest (spikes and spike_traces).
        s_source = im2col_indices(conn.source.s, kernel_height, kernel_width, padding=padding, stride=stride).float()
        s_target = conn.target.s.permute(1, 2, 3, 0).reshape(out_channels, -1).float()
        x_source = im2col_indices(conn.source.x, kernel_height, kernel_width, padding=padding, stride=stride)
        x_target = conn.target.x.permute(1, 2, 3, 0).reshape(out_channels, -1)

        // Post-synaptic.
        post = s_target @ x_source.t()
        conn.w += conn.nu_post * post.view(conn.w.size())
        
        // Pre-synaptic.
        pre = x_target @ s_source.t()
        conn.w -= conn.nu_pre * pre.view(conn.w.size())

        // Bound weights.
        conn.w = torch.clamp(conn.w, conn.wmin, conn.wmax)

    else:
        raise NotImplementedError("This learning rule is not supported for this Connection type.")


def hebbian(conn: AbstractConnection, **kwargs):
    // language=rst

Instances