Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Adding PairNorm support #418

Open
wants to merge 1 commit into
base: master
Choose a base branch
from

Conversation

achiverram28
Copy link
Contributor

Addressing #405

Adding a new normalise.jl

  • Initially added the PairNorm support
  • Will add GraphNorm next

Signed-off-by: achiverram28 <[email protected]>
# Implementation of normalization layers for GraphNeuralNetworks

@doc raw"""
PairNorm(scale_value; [scale_individually])
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
PairNorm(scale_value; [scale_individually])
PairNorm(scale_value; scale_individually=false)

@doc raw"""
PairNorm(scale_value; [scale_individually])

PairNorm layer from paper [PairNorm: Tackling Oversmoothing in GNNs](https://arxiv.org/abs/1909.12223)
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
PairNorm layer from paper [PairNorm: Tackling Oversmoothing in GNNs](https://arxiv.org/abs/1909.12223)
PairNorm layer from paper [PairNorm: Tackling Oversmoothing in GNNs](https://arxiv.org/abs/1909.12223).


PairNorm layer from paper [PairNorm: Tackling Oversmoothing in GNNs](https://arxiv.org/abs/1909.12223)

Performs the operation(normalization)
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
Performs the operation(normalization)
Performs the operation

Comment on lines +10 to +15
\mathbf{x}_i^c &= \mathbf{x}_i - \frac{1}{n}
\sum_{i=1}^n \mathbf{x}_i \\

\mathbf{x}_i^{\prime} &= s \cdot
\frac{\mathbf{x}_i^c}{\sqrt{\frac{1}{n} \sum_{i=1}^n
{\| \mathbf{x}_i^c \|}^2_2}}
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
\mathbf{x}_i^c &= \mathbf{x}_i - \frac{1}{n}
\sum_{i=1}^n \mathbf{x}_i \\
\mathbf{x}_i^{\prime} &= s \cdot
\frac{\mathbf{x}_i^c}{\sqrt{\frac{1}{n} \sum_{i=1}^n
{\| \mathbf{x}_i^c \|}^2_2}}
\mathbf{x}_i^c &= \mathbf{x}_i - \frac{1}{n}
\sum_{i=1}^n \mathbf{x}_i \\
\mathbf{x}_i^{\prime} &= s \cdot
\frac{\mathbf{x}_i^c}{\sqrt{\frac{1}{n} \sum_{i=1}^n
{\| \mathbf{x}_i^c \|}^2_2}}

Comment on lines +18 to +19
The input to this layer is the output from GNN layers

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
The input to this layer is the output from GNN layers


# Arguments

- `scale_value`: Scaling factor `s` used in normalisation. Default `1.0`
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
- `scale_value`: Scaling factor `s` used in normalisation. Default `1.0`
- `scale_value`: Scaling factor `s` used in normalisation. Default `1.0`.

```
Default `false`

- `ϵ` : Small value added in the denominator for numerical stability. Default `1f-5`
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
- `ϵ` : Small value added in the denominator for numerical stability. Default `1f-5`
- `ϵ` : Small value added in the denominator for numerical stability. Default `1f-5`.

This should be mentioned in the first line fo the docstring.

@functor PairNorm

function PairNorm(scale_value::Real=1.0f0; scale_individually::Bool=false, eps::Real=1f-5, ϵ=nothing)
ε = _greek_ascii_depwarn(ϵ => eps, :BatchNorm, "ϵ" => "eps")
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
ε = _greek_ascii_depwarn(ϵ => eps, :BatchNorm, "ϵ" => "eps")


@functor PairNorm

function PairNorm(scale_value::Real=1.0f0; scale_individually::Bool=false, eps::Real=1f-5, ϵ=nothing)
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
function PairNorm(scale_value::Real=1.0f0; scale_individually::Bool=false, eps::Real=1f-5, ϵ=nothing)
function PairNorm(scale_value::Real=1.0f0; scale_individually::Bool=false, eps::Real=1f-5)

return PairNorm(scale_value, ε, scale_individually)
end

function (PN::PairNorm)(x::AbstractArray)
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
function (PN::PairNorm)(x::AbstractArray)
function (pn::PairNorm)(x::AbstractArray)
eps = ofeltype(x, pn.ϵ)
s = ofeltype(pn.scale_value)

end

function (PN::PairNorm)(x::AbstractArray)
xm = mean(x, dims=1)
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

all dimensions are wrong here and belowe. The node dimension is the secnd dimension, the feature dimension is the first

Suggested change
xm = mean(x, dims=1)
xm = mean(x, dims=2)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants