probinet.evaluation.likelihood

probinet.evaluation.likelihood#

This module provides functions for computing the log-likelihood and pseudo log-likelihood of the models, as well as other related calculations.

Functions

PSloglikelihood(B, u, v, w, eta[, mask])

Compute the pseudo log-likelihood of the data.

calculate_opt_func(B, algo_obj[, mask, ...])

Compute the optimal value for the pseudo log-likelihood with the inferred parameters.

likelihood_conditional(M, beta, data, data_tm1)

Compute the log-likelihood of the data given the parameters.

log_likelihood_given_model(model, ...)

Calculate the log-likelihood of the model considering only the structural data.

loglikelihood(B, X, u, v, w, beta, gamma[, ...])

Return the log-likelihood of the model.

loglikelihood_attributes(X, u, v, beta[, mask])

Compute the log-likelihood for the attributes.

loglikelihood_network(B, u, v, w[, mask])

Compute the log-likelihood for the network structure.
probinet.evaluation.likelihood.PSloglikelihood(B: ndarray, u: ndarray, v: ndarray, w: ndarray, eta: float, mask: ndarray | None = None) → float[source]#

Compute the pseudo log-likelihood of the data.

Parameters:

  • B (np.ndarray) – Graph adjacency tensor.

  • u (np.ndarray) – Out-going membership matrix.

  • v (np.ndarray) – In-coming membership matrix.

  • w (np.ndarray) – Affinity tensor.

  • eta (float) – Reciprocity coefficient.

  • mask (Optional[np.ndarray]) – Mask for selecting the held out set in the adjacency tensor in case of cross-validation.

Returns:

Pseudo log-likelihood value.

Return type:

float

probinet.evaluation.likelihood.calculate_opt_func(B: ndarray, algo_obj, mask: ndarray | None = None, assortative: bool = False) → float[source]#

Compute the optimal value for the pseudo log-likelihood with the inferred parameters.

Parameters:

  • B (np.ndarray) – Graph adjacency tensor.

  • algo_obj (object) – The CRep object.

  • mask (Optional[np.ndarray]) – Mask for selecting a subset of the adjacency tensor.

  • assortative (bool) – Flag to use an assortative mode.

Returns:

Maximum pseudo log-likelihood value.

Return type:

float

probinet.evaluation.likelihood.likelihood_conditional(M: ndarray, beta: float, data: ndarray, data_tm1: ndarray, EPS: float | None = 1e-10) → float[source]#

Compute the log-likelihood of the data given the parameters.

Parameters:

  • M (np.ndarray) – Matrix of expected values.

  • beta (float) – Rate of edge removal.

  • data (np.ndarray) – Current adjacency tensor.

  • data_tm1 (np.ndarray) – Previous adjacency tensor.

  • EPS (float, optional) – Small constant to prevent division by zero and log of zero.

Returns:

Log-likelihood value.

Return type:

float

probinet.evaluation.likelihood.log_likelihood_given_model(model: object, adjacency_tensor: ndarray) → float[source]#

Calculate the log-likelihood of the model considering only the structural data.

Parameters:

  • model (object) – The model object containing the lambda0_ija and lambda0_nz attributes.

  • adjacency_tensor (np.ndarray) – The adjacency matrix.

Returns:

The log-likelihood value, rounded to three decimal places.

Return type:

float

probinet.evaluation.likelihood.loglikelihood(B: ndarray, X: DataFrame, u: ndarray, v: ndarray, w: ndarray, beta: ndarray, gamma: float, maskG: ndarray | None = None, maskX: ndarray | None = None) → float[source]#

Return the log-likelihood of the model.

Parameters:

  • B (np.ndarray) – Graph adjacency tensor.

  • X (pd.DataFrame) – Pandas DataFrame object representing the one-hot encoding version of the design matrix.

  • u (np.ndarray) – Membership matrix (out-degree).

  • v (np.ndarray) – Membership matrix (in-degree).

  • w (np.ndarray) – Affinity tensor.

  • beta (np.ndarray) – Beta parameter matrix.

  • gamma (float) – Scaling parameter gamma.

  • maskG (Optional[np.ndarray]) – Mask for selecting a subset in the adjacency tensor.

  • maskX (Optional[np.ndarray]) – Mask for selecting a subset in the design matrix.

Returns:

Log-likelihood value.

Return type:

float

probinet.evaluation.likelihood.loglikelihood_attributes(X: DataFrame, u: ndarray, v: ndarray, beta: ndarray, mask: ndarray | None = None) → float[source]#

Compute the log-likelihood for the attributes.

Parameters:

  • X (pd.DataFrame) – Pandas DataFrame object representing the one-hot encoding version of the design matrix.

  • u (np.ndarray) – Membership matrix (out-degree).

  • v (np.ndarray) – Membership matrix (in-degree).

  • beta (np.ndarray) – Beta parameter matrix.

  • mask (Optional[np.ndarray]) – Mask for selecting a subset in the design matrix.

Returns:

Log-likelihood value for the attributes.

Return type:

float

probinet.evaluation.likelihood.loglikelihood_network(B: ndarray, u: ndarray, v: ndarray, w: ndarray, mask: ndarray | None = None) → float[source]#

Compute the log-likelihood for the network structure.

Parameters:

  • B (np.ndarray) – Graph adjacency tensor.

  • u (np.ndarray) – Membership matrix (out-degree).

  • v (np.ndarray) – Membership matrix (in-degree).

  • w (np.ndarray) – Affinity tensor.

  • mask (Optional[np.ndarray]) – Mask for selecting a subset in the adjacency tensor.

Returns:

Log-likelihood value for the network structure.

Return type:

float