1NGRAMSHRINK(1) User Commands NGRAMSHRINK(1)
2
3
4
6 ngramshrink - manual page for ngramshrink 1.3.4
7
9 Shrink ngram model from input model file.
10
11 Usage: ngramshrink [--options] [in.fst [out.fst]]
12
13 PROGRAM FLAGS:
14
15 --backoff_label: type = int64, default = 0
16
17 Backoff label --check_consistency: type = bool, default = false
18 Check model consistency --context_pattern: type = string, de‐
19 fault = "" Pattern of contexts to prune --count_pattern: type =
20 string, default = "" Pattern of counts to prune --method: type =
21 string, default = "seymore" One of: "context_prune",
22 "count_prune", "relative_entropy", "seymore" --norm_eps: type =
23 double, default = 0.001 Normalization check epsilon
24 --shrink_opt: type = int32, default = 0 Optimization level:
25 Range 0 (fastest) to 2 (most accurate) --target_num‐
26 ber_of_ngrams: type = int64, default = -1 Maximum number of
27 ngrams to leave in model after pruning. Value less than zero
28 means no target number, just use theta. --theta: type = double,
29 default = 0 Pruning threshold theta --total_unigram_count: type
30 = double, default = -1 Total unigram count
31
32 LIBRARY FLAGS:
33
34 Flags from: flags.cc
35
36 --help: type = bool, default = false
37
38 show usage information --helpshort: type = bool, default = false
39 show brief usage information --tmpdir: type = string, default =
40 "/tmp" temporary directory --v: type = int32, default = 0 ver‐
41 bosity level
42
43 Flags from: fst.cc
44
45 --fst_align: type = bool, default = false
46
47 Write FST data aligned where appropriate --fst_default_cache_gc:
48 type = bool, default = true Enable garbage collection of cache
49 --fst_default_cache_gc_limit: type = int64, default = 1048576
50 Cache byte size that triggers garbage collection
51 --fst_read_mode: type = string, default = "read" Default file
52 reading mode for mappable files --fst_verify_properties: type =
53 bool, default = false Verify FST properties queried by TestProp‐
54 erties --save_relabel_ipairs: type = string, default = "" Save
55 input relabel pairs to file --save_relabel_opairs: type =
56 string, default = "" Save output relabel pairs to file
57
58 Flags from: ngram-output.cc
59
60 --end_symbol: type = string, default = "</s>"
61
62 Class label for sentence end --start_symbol: type = string, de‐
63 fault = "<s>" Class label for sentence start
64
65 Flags from: symbol-table.cc
66
67 --fst_compat_symbols: type = bool, default = true
68
69 Require symbol tables to match when appropriate --fst_field_sep‐
70 arator: type = string, default = " " Set of characters used
71 as a separator between printed fields
72
73 Flags from: util.cc
74
75 --fst_error_fatal: type = bool, default = true
76
77 FST errors are fatal; o.w. return objects flagged as bad: e.g.,
78 FSTs: kError property set, FST weights: not a Member()
79 --ngram_error_fatal: type = bool, default = true NGram errors
80 are fatal if true; otherwise returns objects flagged as bad:
81 e.g., NGramModel::Error() is true
82
83 Flags from: weight.cc
84
85 --fst_weight_parentheses: type = string, default = ""
86
87 Characters enclosing the first weight of a printed composite
88 weight (e.g., pair weight, tuple weight and derived classes) to
89 ensure proper I/O of nested composite weights; must have size 0
90 (none) or 2 (open and close parenthesis) --fst_weight_separator:
91 type = string, default = "," Character separator between printed
92 composite weights; must be a single character
93
94
95
96ngramshrink 1.3.4 January 2018 NGRAMSHRINK(1)