Previous  Up  Next

Module Pll

module Pll: sig .. end
The Pll module performs efficient computation of pseudo-likelihood and its gradient, given a Markov network represented as a set of weighted conjunctive features. It includes implementations in both C and OCaml, with and without a cache of feature statistics.
type satisfaction =
| Sat
| Unsat of (int * bool array)
| NeverSat
Keep track of how many violations of a conjunctive feature are present -- zero, one, or more than one.
type wexample_t = Data.wexample_t 
type example_t = Data.example_t 
type schema_t = Data.schema_t 
type valgrad_callback = float array -> float array -> float 
type pll_cache_t 
val build_pll_cache : Mn.network -> wexample_t array -> pll_cache_t
Construct key statistics for PLL value and gradient computations
val pll_val_and_grad_cached : pll_cache_t -> valgrad_callback
Compute value and gradient of PLL using cache
val build_pll_minicache : Mn.network -> wexample_t array -> pll_cache_t
Construct small cache data structure, not precomputing feature counts
val pll_val_and_grad : pll_cache_t -> valgrad_callback
Compute value and gradient of PLL without using full cache
val pll_cache_to_mn : pll_cache_t -> float array -> Mn.network
Construct an MN from the PLL cache data and the weight vector.
val fa_pll : schema_t -> Mn.Factor.feature array -> example_t -> float
Compute PLL of a single example from a feature array.

Computing PLL and its gradient with external C implementation

type mn_t 
val create_mn : int array -> Mn.Factor.feature array -> mn_t
Build MN structure for use with external C PLL implementation.
val pll_mn : mn_t -> int array -> float
Compute PLL of MN. External C implementation.
type minicache_c_t 
val build_pll_minicache_c : Mn.network -> wexample_t array -> minicache_c_t
Construct small cache data structure, not precomputing feature counts, for external C implementation.
val pll_val_and_grad_c : minicache_c_t -> valgrad_callback
Compute value and gradient PLL without using full cache. External C implementation.
type cache_c_t 
val build_pll_cache_c : Mn.network -> wexample_t array -> cache_c_t
Construct key statistics for PLL value and gradient computations, for use with external C implementation.
val pll_val_and_grad_cached_c : cache_c_t -> valgrad_callback
Compute PLL and its gradient using cache. External C implementation.