cascade.nn.IntervDesign

class cascade.nn.IntervDesign(n_vars, k, design_scale_bias, mask, interv_scale, interv_bias, target_weight)[source]

Intervention design module

Parameters:

n_vars (int) – Number of variables
k (int) – Maximal combinatorial order to consider
design_scale_bias (bool) – Whether to optimize the intervention scale and bias
mask (BoolTensor) – Boolean mask that marks variables in the design candidate pool
interv_scale (Tensor) – Intervention scale tensor trained in the discover phase
interv_bias (Tensor) – Intervention bias tensor trained in the discover phase
target_weight (Tensor) – Variable weight when computing target deviation

Methods

`load`	Load design module from file
`loss`
`rsample`
`save`	Save the design module to file
`simplex2regime`

Attributes