poisson.llk: Evaluate Poisson log-likelihood for counts matrix
In poismf: Factorization of Sparse Counts Matrices Through Poisson Likelihood

Description Usage Arguments Details Value See Also

Calculates Poisson log-likehood plus constant for new combinations of rows and columns of 'X'. Intended to use as a test metric or for monitoring a validation set.

By default, this Poisson log-likelihood is calculated only for the combinations of users (rows) and items (columns) provided in 'X_test' here, ignoring the missing entries. This is the usual use-case for evaluating a validation or test set, but can also be used for evaluating it on the training data with all missing entries included as zeros (see parameters for details).

Note that this calculates a sum rather than an average.

1	poisson.llk(model, X_test, full_llk = FALSE, include_missing = FALSE)

`model`	A Poisson factorization model object as returned by 'poismf'.
`X_test`	Input data on which to calculate log-likelihood, consisting of triplets. Can be passed as a 'data.frame' or as a sparse COO matrix (see documentation of poismf for details on the accepted data types). If the 'X' data passed to 'poismf' was a 'data.frame', should pass a 'data.frame' with entries corresponding to the same IDs, otherwise might pass either a 'data.frame' with the row and column indices (starting at 1), or a sparse COO matrix.
`full_llk`	Whether to add to the number a constant given by the data which doesn't depend on the fitted parameters. If passing 'False' (the default), there is some chance that the resulting log-likelihood will end up being positive - this is not an error, but is simply due to ommission of this constant. Passing 'TRUE' might result in numeric overflow and low numerical precision.
`include_missing`	If 'TRUE', will calculate the Poisson log-likelihood for all entries (i.e. all combinations of users/items, whole matrix 'X'), taking the missing ones as zero-valued. If passing 'FALSE', will calculate the Poisson log-likelihood only for the non-missing entries passed in 'X_test' - this is usually the desired behavior when evaluating a test dataset.