Compiles the model. The method prepares a neural network for estimation by setting the optimizer, loss function, and other parameters as needed.
DLMatchingModel
The method calls the
tf.keras.compile method of the tf.keras.Model class.
NSMatchingModel
Neural symbolic models are fitted using a custom training loop. The method sets the optimizer for model training. The loss (for hybrid neural-symbolic and deep learning models) is set to binary cross-entropy ( tf.keras.losses.BinaryCrossentropy)
RefutationModel
Refutation models, similar to neural-symbolic models (NSMatchingModel),
are fitted using a custom. The method sets the optimizer for model training.
Usage
compile(object, ...)
# S4 method for class 'neer_match.matching_model.DLMatchingModel'
compile(object, ...)
# S4 method for class 'neer_match.matching_model.NSMatchingModel'
compile(object, optimizer = tensorflow::tf$keras$optimizers$Adam())
# S4 method for class 'neer_match.reasoning.RefutationModel'
compile(object, optimizer = tensorflow::tf$keras$optimizers$Adam())Arguments
- object
A matching model object.
- ...
Additional arguments passed to tf.keras.compile.
- optimizer
A tf.keras.optimizers optimizer object.
Examples
smap <- SimilarityMap(
instructions = list(
`movie ~film` = list("jaro_winkler", "damerau_levenshtein"),
`studio` = list("jaccard", "levenshtein"),
`reviews~score` = list("euclidean")
)
)
model <- DLMatchingModel(smap)
compile(model, optimizer = tensorflow::tf$keras$optimizers$Adam())