model.fbs

//
// Copyright (C) 2017 The Android Open Source Project
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//      http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
//

file_identifier "TC2 ";

// The possible model modes, represents a bit field.
namespace libtextclassifier2;
enum ModeFlag : int {
  NONE = 0,
  ANNOTATION = 1,
  CLASSIFICATION = 2,
  ANNOTATION_AND_CLASSIFICATION = 3,
  SELECTION = 4,
  ANNOTATION_AND_SELECTION = 5,
  CLASSIFICATION_AND_SELECTION = 6,
  ALL = 7,
}

namespace libtextclassifier2;
enum DatetimeExtractorType : int {
  UNKNOWN_DATETIME_EXTRACTOR_TYPE = 0,
  AM = 1,
  PM = 2,
  JANUARY = 3,
  FEBRUARY = 4,
  MARCH = 5,
  APRIL = 6,
  MAY = 7,
  JUNE = 8,
  JULY = 9,
  AUGUST = 10,
  SEPTEMBER = 11,
  OCTOBER = 12,
  NOVEMBER = 13,
  DECEMBER = 14,
  NEXT = 15,
  NEXT_OR_SAME = 16,
  LAST = 17,
  NOW = 18,
  TOMORROW = 19,
  YESTERDAY = 20,
  PAST = 21,
  FUTURE = 22,
  DAY = 23,
  WEEK = 24,
  MONTH = 25,
  YEAR = 26,
  MONDAY = 27,
  TUESDAY = 28,
  WEDNESDAY = 29,
  THURSDAY = 30,
  FRIDAY = 31,
  SATURDAY = 32,
  SUNDAY = 33,
  DAYS = 34,
  WEEKS = 35,
  MONTHS = 36,
  HOURS = 37,
  MINUTES = 38,
  SECONDS = 39,
  YEARS = 40,
  DIGITS = 41,
  SIGNEDDIGITS = 42,
  ZERO = 43,
  ONE = 44,
  TWO = 45,
  THREE = 46,
  FOUR = 47,
  FIVE = 48,
  SIX = 49,
  SEVEN = 50,
  EIGHT = 51,
  NINE = 52,
  TEN = 53,
  ELEVEN = 54,
  TWELVE = 55,
  THIRTEEN = 56,
  FOURTEEN = 57,
  FIFTEEN = 58,
  SIXTEEN = 59,
  SEVENTEEN = 60,
  EIGHTEEN = 61,
  NINETEEN = 62,
  TWENTY = 63,
  THIRTY = 64,
  FORTY = 65,
  FIFTY = 66,
  SIXTY = 67,
  SEVENTY = 68,
  EIGHTY = 69,
  NINETY = 70,
  HUNDRED = 71,
  THOUSAND = 72,
}

namespace libtextclassifier2;
enum DatetimeGroupType : int {
  GROUP_UNKNOWN = 0,
  GROUP_UNUSED = 1,
  GROUP_YEAR = 2,
  GROUP_MONTH = 3,
  GROUP_DAY = 4,
  GROUP_HOUR = 5,
  GROUP_MINUTE = 6,
  GROUP_SECOND = 7,
  GROUP_AMPM = 8,
  GROUP_RELATIONDISTANCE = 9,
  GROUP_RELATION = 10,
  GROUP_RELATIONTYPE = 11,

  // Dummy groups serve just as an inflator of the selection. E.g. we might want
  // to select more text than was contained in an envelope of all extractor
  // spans.
  GROUP_DUMMY1 = 12,

  GROUP_DUMMY2 = 13,
}

namespace libtextclassifier2;
table CompressedBuffer {
  buffer:[ubyte];
  uncompressed_size:int;
}

// Options for the model that predicts text selection.
namespace libtextclassifier2;
table SelectionModelOptions {
  // If true, before the selection is returned, the unpaired brackets contained
  // in the predicted selection are stripped from the both selection ends.
  // The bracket codepoints are defined in the Unicode standard:
  // http://www.unicode.org/Public/UNIDATA/BidiBrackets.txt
  strip_unpaired_brackets:bool = 1;

  // Number of hypothetical click positions on either side of the actual click
  // to consider in order to enforce symmetry.
  symmetry_context_size:int;

  // Number of examples to bundle in one batch for inference.
  batch_size:int = 1024;

  // Whether to always classify a suggested selection or only on demand.
  always_classify_suggested_selection:bool = 0;
}

// Options for the model that classifies a text selection.
namespace libtextclassifier2;
table ClassificationModelOptions {
  // Limits for phone numbers.
  phone_min_num_digits:int = 7;

  phone_max_num_digits:int = 15;

  // Limits for addresses.
  address_min_num_tokens:int;

  // Maximum number of tokens to attempt a classification (-1 is unlimited).
  max_num_tokens:int = -1;
}

// List of regular expression matchers to check.
namespace libtextclassifier2.RegexModel_;
table Pattern {
  // The name of the collection of a match.
  collection_name:string;

  // The pattern to check.
  // Can specify a single capturing group used as match boundaries.
  pattern:string;

  // The modes for which to apply the patterns.
  enabled_modes:libtextclassifier2.ModeFlag = ALL;

  // The final score to assign to the results of this pattern.
  target_classification_score:float = 1;

  // Priority score used for conflict resolution with the other models.
  priority_score:float = 0;

  // If true, will use an approximate matching implementation implemented
  // using Find() instead of the true Match(). This approximate matching will
  // use the first Find() result and then check that it spans the whole input.
  use_approximate_matching:bool = 0;

  compressed_pattern:libtextclassifier2.CompressedBuffer;
}

namespace libtextclassifier2;
table RegexModel {
  patterns:[libtextclassifier2.RegexModel_.Pattern];
}

// List of regex patterns.
namespace libtextclassifier2.DatetimeModelPattern_;
table Regex {
  pattern:string;

  // The ith entry specifies the type of the ith capturing group.
  // This is used to decide how the matched content has to be parsed.
  groups:[libtextclassifier2.DatetimeGroupType];

  compressed_pattern:libtextclassifier2.CompressedBuffer;
}

namespace libtextclassifier2;
table DatetimeModelPattern {
  regexes:[libtextclassifier2.DatetimeModelPattern_.Regex];

  // List of locale indices in DatetimeModel that represent the locales that
  // these patterns should be used for. If empty, can be used for all locales.
  locales:[int];

  // The final score to assign to the results of this pattern.
  target_classification_score:float = 1;

  // Priority score used for conflict resulution with the other models.
  priority_score:float = 0;

  // The modes for which to apply the patterns.
  enabled_modes:libtextclassifier2.ModeFlag = ALL;
}

namespace libtextclassifier2;
table DatetimeModelExtractor {
  extractor:libtextclassifier2.DatetimeExtractorType;
  pattern:string;
  locales:[int];
  compressed_pattern:libtextclassifier2.CompressedBuffer;
}

namespace libtextclassifier2;
table DatetimeModel {
  // List of BCP 47 locale strings representing all locales supported by the
  // model. The individual patterns refer back to them using an index.
  locales:[string];

  patterns:[libtextclassifier2.DatetimeModelPattern];
  extractors:[libtextclassifier2.DatetimeModelExtractor];

  // If true, will use the extractors for determining the match location as
  // opposed to using the location where the global pattern matched.
  use_extractors_for_locating:bool = 1;

  // List of locale ids, rules of whose are always run, after the requested
  // ones.
  default_locales:[int];
}

namespace libtextclassifier2.DatetimeModelLibrary_;
table Item {
  key:string;
  value:libtextclassifier2.DatetimeModel;
}

// A set of named DateTime models.
namespace libtextclassifier2;
table DatetimeModelLibrary {
  models:[libtextclassifier2.DatetimeModelLibrary_.Item];
}

// Options controlling the output of the Tensorflow Lite models.
namespace libtextclassifier2;
table ModelTriggeringOptions {
  // Lower bound threshold for filtering annotation model outputs.
  min_annotate_confidence:float = 0;

  // The modes for which to enable the models.
  enabled_modes:libtextclassifier2.ModeFlag = ALL;
}

// Options controlling the output of the classifier.
namespace libtextclassifier2;
table OutputOptions {
  // Lists of collection names that will be filtered out at the output:
  // - For annotation, the spans of given collection are simply dropped.
  // - For classification, the result is mapped to the class "other".
  // - For selection, the spans of given class are returned as
  // single-selection.
  filtered_collections_annotation:[string];

  filtered_collections_classification:[string];
  filtered_collections_selection:[string];
}

namespace libtextclassifier2;
table Model {
  // Comma-separated list of locales supported by the model as BCP 47 tags.
  locales:string;

  version:int;

  // A name for the model that can be used for e.g. logging.
  name:string;

  selection_feature_options:libtextclassifier2.FeatureProcessorOptions;
  classification_feature_options:libtextclassifier2.FeatureProcessorOptions;

  // Tensorflow Lite models.
  selection_model:[ubyte] (force_align: 16);

  classification_model:[ubyte] (force_align: 16);
  embedding_model:[ubyte] (force_align: 16);

  // Options for the different models.
  selection_options:libtextclassifier2.SelectionModelOptions;

  classification_options:libtextclassifier2.ClassificationModelOptions;
  regex_model:libtextclassifier2.RegexModel;
  datetime_model:libtextclassifier2.DatetimeModel;

  // Options controlling the output of the models.
  triggering_options:libtextclassifier2.ModelTriggeringOptions;

  // Global switch that controls if SuggestSelection(), ClassifyText() and
  // Annotate() will run. If a mode is disabled it returns empty/no-op results.
  enabled_modes:libtextclassifier2.ModeFlag = ALL;

  // If true, will snap the selections that consist only of whitespaces to the
  // containing suggested span. Otherwise, no suggestion is proposed, since the
  // selections are not part of any token.
  snap_whitespace_selections:bool = 1;

  // Global configuration for the output of SuggestSelection(), ClassifyText()
  // and Annotate().
  output_options:libtextclassifier2.OutputOptions;
}

// Role of the codepoints in the range.
namespace libtextclassifier2.TokenizationCodepointRange_;
enum Role : int {
  // Concatenates the codepoint to the current run of codepoints.
  DEFAULT_ROLE = 0,

  // Splits a run of codepoints before the current codepoint.
  SPLIT_BEFORE = 1,

  // Splits a run of codepoints after the current codepoint.
  SPLIT_AFTER = 2,

  // Each codepoint will be a separate token. Good e.g. for Chinese
  // characters.
  TOKEN_SEPARATOR = 3,

  // Discards the codepoint.
  DISCARD_CODEPOINT = 4,

  // Common values:
  // Splits on the characters and discards them. Good e.g. for the space
  // character.
  WHITESPACE_SEPARATOR = 7,
}

// Represents a codepoint range [start, end) with its role for tokenization.
namespace libtextclassifier2;
table TokenizationCodepointRange {
  start:int;
  end:int;
  role:libtextclassifier2.TokenizationCodepointRange_.Role;

  // Integer identifier of the script this range denotes. Negative values are
  // reserved for Tokenizer's internal use.
  script_id:int;
}

// Method for selecting the center token.
namespace libtextclassifier2.FeatureProcessorOptions_;
enum CenterTokenSelectionMethod : int {
  DEFAULT_CENTER_TOKEN_METHOD = 0,

  // Use click indices to determine the center token.
  CENTER_TOKEN_FROM_CLICK = 1,

  // Use selection indices to get a token range, and select the middle of it
  // as the center token.
  CENTER_TOKEN_MIDDLE_OF_SELECTION = 2,
}

// Controls the type of tokenization the model will use for the input text.
namespace libtextclassifier2.FeatureProcessorOptions_;
enum TokenizationType : int {
  INVALID_TOKENIZATION_TYPE = 0,

  // Use the internal tokenizer for tokenization.
  INTERNAL_TOKENIZER = 1,

  // Use ICU for tokenization.
  ICU = 2,

  // First apply ICU tokenization. Then identify stretches of tokens
  // consisting only of codepoints in internal_tokenizer_codepoint_ranges
  // and re-tokenize them using the internal tokenizer.
  MIXED = 3,
}

// Range of codepoints start - end, where end is exclusive.
namespace libtextclassifier2.FeatureProcessorOptions_;
table CodepointRange {
  start:int;
  end:int;
}

// Bounds-sensitive feature extraction configuration.
namespace libtextclassifier2.FeatureProcessorOptions_;
table BoundsSensitiveFeatures {
  // Enables the extraction of bounds-sensitive features, instead of the click
  // context features.
  enabled:bool;

  // The numbers of tokens to extract in specific locations relative to the
  // bounds.
  // Immediately before the span.
  num_tokens_before:int;

  // Inside the span, aligned with the beginning.
  num_tokens_inside_left:int;

  // Inside the span, aligned with the end.
  num_tokens_inside_right:int;

  // Immediately after the span.
  num_tokens_after:int;

  // If true, also extracts the tokens of the entire span and adds up their
  // features forming one "token" to include in the extracted features.
  include_inside_bag:bool;

  // If true, includes the selection length (in the number of tokens) as a
  // feature.
  include_inside_length:bool;

  // If true, for selection, single token spans are not run through the model
  // and their score is assumed to be zero.
  score_single_token_spans_as_zero:bool;
}

namespace libtextclassifier2.FeatureProcessorOptions_;
table AlternativeCollectionMapEntry {
  key:string;
  value:string;
}

namespace libtextclassifier2;
table FeatureProcessorOptions {
  // Number of buckets used for hashing charactergrams.
  num_buckets:int = -1;

  // Size of the embedding.
  embedding_size:int = -1;

  // Number of bits for quantization for embeddings.
  embedding_quantization_bits:int = 8;

  // Context size defines the number of words to the left and to the right of
  // the selected word to be used as context. For example, if context size is
  // N, then we take N words to the left and N words to the right of the
  // selected word as its context.
  context_size:int = -1;

  // Maximum number of words of the context to select in total.
  max_selection_span:int = -1;

  // Orders of charactergrams to extract. E.g., 2 means character bigrams, 3
  // character trigrams etc.
  chargram_orders:[int];

  // Maximum length of a word, in codepoints.
  max_word_length:int = 20;

  // If true, will use the unicode-aware functionality for extracting features.
  unicode_aware_features:bool = 0;

  // Whether to extract the token case feature.
  extract_case_feature:bool = 0;

  // Whether to extract the selection mask feature.
  extract_selection_mask_feature:bool = 0;

  // List of regexps to run over each token. For each regexp, if there is a
  // match, a dense feature of 1.0 is emitted. Otherwise -1.0 is used.
  regexp_feature:[string];

  // Whether to remap all digits to a single number.
  remap_digits:bool = 0;

  // Whether to lower-case each token before generating hashgrams.
  lowercase_tokens:bool;

  // If true, the selection classifier output will contain only the selections
  // that are feasible (e.g., those that are shorter than max_selection_span),
  // if false, the output will be a complete cross-product of possible
  // selections to the left and possible selections to the right, including the
  // infeasible ones.
  // NOTE: Exists mainly for compatibility with older models that were trained
  // with the non-reduced output space.
  selection_reduced_output_space:bool = 1;

  // Collection names.
  collections:[string];

  // An index of collection in collections to be used if a collection name can't
  // be mapped to an id.
  default_collection:int = -1;

  // If true, will split the input by lines, and only use the line that contains
  // the clicked token.
  only_use_line_with_click:bool = 0;

  // If true, will split tokens that contain the selection boundary, at the
  // position of the boundary.
  // E.g. "foo{bar}@google.com" -> "foo", "bar", "@google.com"
  split_tokens_on_selection_boundaries:bool = 0;

  // Codepoint ranges that determine how different codepoints are tokenized.
  // The ranges must not overlap.
  tokenization_codepoint_config:[libtextclassifier2.TokenizationCodepointRange];

  center_token_selection_method:libtextclassifier2.FeatureProcessorOptions_.CenterTokenSelectionMethod;

  // If true, span boundaries will be snapped to containing tokens and not
  // required to exactly match token boundaries.
  snap_label_span_boundaries_to_containing_tokens:bool;

  // A set of codepoint ranges supported by the model.
  supported_codepoint_ranges:[libtextclassifier2.FeatureProcessorOptions_.CodepointRange];

  // A set of codepoint ranges to use in the mixed tokenization mode to identify
  // stretches of tokens to re-tokenize using the internal tokenizer.
  internal_tokenizer_codepoint_ranges:[libtextclassifier2.FeatureProcessorOptions_.CodepointRange];

  // Minimum ratio of supported codepoints in the input context. If the ratio
  // is lower than this, the feature computation will fail.
  min_supported_codepoint_ratio:float = 0;

  // Used for versioning the format of features the model expects.
  // - feature_version == 0:
  // For each token the features consist of:
  // - chargram embeddings
  // - dense features
  // Chargram embeddings for tokens are concatenated first together,
  // and at the end, the dense features for the tokens are concatenated
  // to it. So the resulting feature vector has two regions.
  feature_version:int = 0;

  tokenization_type:libtextclassifier2.FeatureProcessorOptions_.TokenizationType = INTERNAL_TOKENIZER;
  icu_preserve_whitespace_tokens:bool = 0;

  // List of codepoints that will be stripped from beginning and end of
  // predicted spans.
  ignored_span_boundary_codepoints:[int];

  bounds_sensitive_features:libtextclassifier2.FeatureProcessorOptions_.BoundsSensitiveFeatures;

  // List of allowed charactergrams. The extracted charactergrams are filtered
  // using this list, and charactergrams that are not present are interpreted as
  // out-of-vocabulary.
  // If no allowed_chargrams are specified, all charactergrams are allowed.
  // The field is typed as bytes type to allow non-UTF8 chargrams.
  allowed_chargrams:[string];

  // If true, tokens will be also split when the codepoint's script_id changes
  // as defined in TokenizationCodepointRange.
  tokenize_on_script_change:bool = 0;
}

root_type libtextclassifier2.Model;