layers

Custom layers that are needed for the various models.

bpreveal.layers.linearRegression(**kwargs)

A linear regression layer that’s compatible with shap.

Implements the following formula:

out = input * slope + offset

where slope and offset are two (scalar) parameters.

Note that although this is a function, it behaves like a normal Keras layer class and you create it like this:

regression = bpreveal.layers.LinearRegression(name="oct4_regression")(regressionInput)

Parameters:

kwargs – Passed to the keras Layer initializer.

Returns:

A function that you can call with a Keras layer as input, like a normal Keras Layer class.

Return type:

A function taking a tensor and returning a tensor.

class bpreveal.layers.CountsLogSumExp(*args, **kwargs)

A simple layer that wraps keras.ops.logaddexp.

Parameters:

• args (Any)

• kwargs (Any)

Return type:

Any

call(inp1, inp2)

Calculate logsumexp.

Parameters:

• inp1 (Tensor) – The first value

• inp2 (Tensor) – The second value

Returns:

log(exp(inp1) + exp(inp2)).

Return type:

Tensor

get_config()

Used to make this class deserializable.

Returns:

A configuration dictionary. There is no custom data in it.

Return type:

dict