CausalEGM.causalEGM.VariationalCausalEGM

class CausalEGM.causalEGM.VariationalCausalEGM(params, timestamp=None, random_seed=None)[source]

Implementation of the variational CausalEGM model. Instead of distribution match with GAN, we use variational inference in the latent space.

Parameters:

params – Dict object denoting the hyperparameters for deployments and building the model architecture. See examples under the src/configs folder.
timestamp – Str object denoting the timestemp for specificing when the model is instanced. Default: None.
random_seed – Int object denoting the random seed for controling randomness. Default: None.

Examples

>>> from CausalEGM import VariationalCausalEGM, Sim_Hirano_Imbens_sampler
>>> import yaml
>>> params = yaml.safe_load(open('src/configs/Sim_Hirano_Imbens.yaml', 'r'))
>>> x,y,v = Sim_Hirano_Imbens_sampler(batch_size=32).load_all()
>>> model = VariationalCausalEGM(params=params,random_seed=12)
>>> model.train(data=[x,y,v],n_iter=30000,save_format='npy')

__init__(params, timestamp=None, random_seed=None)[source]

Methods

`__init__`(params[, timestamp, random_seed])
`encode`(v)	Encode process and get both mean and variance.
`evaluate`(data[, nb_intervals])	Internal evaluation in the training process of variational CausalEGM.
`get_config`()	Get the parameters CausalEGM model.
`initialize_nets`([print_summary])	Initialize all the networks in CausalEGM.
`log_normal_pdf`(sample, mean, logvar[, axis])	Log likelihood of a normal distribution
`reparameterize`(mean, logvar)	Reparameterization for sample latent features.
`sample`([eps])	Generate data by decoder.
`save`(fname, data)	Save the data to the specified path.
`train`([data, data_file, sep, header, ...])	Train a variational CausalEGM model given the input data.
`train_step`(data_z, data_v, data_x, data_y)	Training step in the Variational CausalEGM model.