• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (C) 2018 The Android Open Source Project
3  *
4  * Licensed under the Apache License, Version 2.0 (the "License");
5  * you may not use this file except in compliance with the License.
6  * You may obtain a copy of the License at
7  *
8  *      http://www.apache.org/licenses/LICENSE-2.0
9  *
10  * Unless required by applicable law or agreed to in writing, software
11  * distributed under the License is distributed on an "AS IS" BASIS,
12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13  * See the License for the specific language governing permissions and
14  * limitations under the License.
15  */
16 
17 // Generic feature extractor for extracting features from objects. The feature
18 // extractor can be used for extracting features from any object. The feature
19 // extractor and feature function classes are template classes that have to
20 // be instantiated for extracting feature from a specific object type.
21 //
22 // A feature extractor consists of a hierarchy of feature functions. Each
23 // feature function extracts one or more feature type and value pairs from the
24 // object.
25 //
26 // The feature extractor has a modular design where new feature functions can be
27 // registered as components. The feature extractor is initialized from a
28 // descriptor represented by a protocol buffer. The feature extractor can also
29 // be initialized from a text-based source specification of the feature
30 // extractor. Feature specification parsers can be added as components. By
31 // default the feature extractor can be read from an ASCII protocol buffer or in
32 // a simple feature modeling language (fml).
33 
34 // A feature function is invoked with a focus. Nested feature function can be
35 // invoked with another focus determined by the parent feature function.
36 
37 #ifndef NLP_SAFT_COMPONENTS_COMMON_MOBILE_FEL_FEATURE_EXTRACTOR_H_
38 #define NLP_SAFT_COMPONENTS_COMMON_MOBILE_FEL_FEATURE_EXTRACTOR_H_
39 
40 #include <stddef.h>
41 
42 #include <string>
43 #include <vector>
44 
45 #include "lang_id/common/fel/feature-descriptors.h"
46 #include "lang_id/common/fel/feature-types.h"
47 #include "lang_id/common/fel/task-context.h"
48 #include "lang_id/common/fel/workspace.h"
49 #include "lang_id/common/lite_base/attributes.h"
50 #include "lang_id/common/lite_base/integral-types.h"
51 #include "lang_id/common/lite_base/logging.h"
52 #include "lang_id/common/lite_base/macros.h"
53 #include "lang_id/common/registry.h"
54 #include "lang_id/common/stl-util.h"
55 
56 namespace libtextclassifier3 {
57 namespace mobile {
58 
59 // TODO(djweiss) Clean this up as well.
60 // Use the same type for feature values as is used for predicated.
61 typedef int64 Predicate;
62 typedef Predicate FeatureValue;
63 
64 // A union used to represent discrete and continuous feature values.
65 union FloatFeatureValue {
66  public:
FloatFeatureValue(FeatureValue v)67   explicit FloatFeatureValue(FeatureValue v) : discrete_value(v) {}
FloatFeatureValue(uint32 i,float w)68   FloatFeatureValue(uint32 i, float w) : id(i), weight(w) {}
69   FeatureValue discrete_value;
70   struct {
71     uint32 id;
72     float weight;
73   };
74 };
75 
76 // A feature vector contains feature type and value pairs.
77 class FeatureVector {
78  public:
FeatureVector()79   FeatureVector() {}
80 
81   // Adds feature type and value pair to feature vector.
add(FeatureType * type,FeatureValue value)82   void add(FeatureType *type, FeatureValue value) {
83     features_.emplace_back(type, value);
84   }
85 
86   // Removes all elements from the feature vector.
clear()87   void clear() { features_.clear(); }
88 
89   // Returns the number of elements in the feature vector.
size()90   int size() const { return features_.size(); }
91 
92   // Reserves space in the underlying feature vector.
reserve(int n)93   void reserve(int n) { features_.reserve(n); }
94 
95   // Returns feature type for an element in the feature vector.
type(int index)96   FeatureType *type(int index) const { return features_[index].type; }
97 
98   // Returns feature value for an element in the feature vector.
value(int index)99   FeatureValue value(int index) const { return features_[index].value; }
100 
101  private:
102   // Structure for holding feature type and value pairs.
103   struct Element {
ElementElement104     Element() : type(nullptr), value(-1) {}
ElementElement105     Element(FeatureType *t, FeatureValue v) : type(t), value(v) {}
106 
107     FeatureType *type;
108     FeatureValue value;
109   };
110 
111   // Array for storing feature vector elements.
112   std::vector<Element> features_;
113 
114   SAFTM_DISALLOW_COPY_AND_ASSIGN(FeatureVector);
115 };
116 
117 // The generic feature extractor is the type-independent part of a feature
118 // extractor. This holds the descriptor for the feature extractor and the
119 // collection of feature types used in the feature extractor.  The feature
120 // types are not available until FeatureExtractor<>::Init() has been called.
121 class GenericFeatureExtractor {
122  public:
123   GenericFeatureExtractor();
124   virtual ~GenericFeatureExtractor();
125 
126   // Initializes the feature extractor from the FEL specification |source|.
127   //
128   // Returns true on success, false otherwise (e.g., FEL syntax error).
129   SAFTM_MUST_USE_RESULT bool Parse(const string &source);
130 
131   // Returns the feature extractor descriptor.
descriptor()132   const FeatureExtractorDescriptor &descriptor() const { return descriptor_; }
mutable_descriptor()133   FeatureExtractorDescriptor *mutable_descriptor() { return &descriptor_; }
134 
135   // Returns the number of feature types in the feature extractor.  Invalid
136   // before Init() has been called.
feature_types()137   int feature_types() const { return feature_types_.size(); }
138 
139  protected:
140   // Initializes the feature types used by the extractor.  Called from
141   // FeatureExtractor<>::Init().
142   //
143   // Returns true on success, false otherwise.
144   SAFTM_MUST_USE_RESULT bool InitializeFeatureTypes();
145 
146  private:
147   // Initializes the top-level feature functions.
148   //
149   // Returns true on success, false otherwise.
150   SAFTM_MUST_USE_RESULT virtual bool InitializeFeatureFunctions() = 0;
151 
152   // Returns all feature types used by the extractor. The feature types are
153   // added to the result array.
154   virtual void GetFeatureTypes(std::vector<FeatureType *> *types) const = 0;
155 
156   // Descriptor for the feature extractor. This is a protocol buffer that
157   // contains all the information about the feature extractor. The feature
158   // functions are initialized from the information in the descriptor.
159   FeatureExtractorDescriptor descriptor_;
160 
161   // All feature types used by the feature extractor. The collection of all the
162   // feature types describes the feature space of the feature set produced by
163   // the feature extractor.  Not owned.
164   std::vector<FeatureType *> feature_types_;
165 };
166 
167 // The generic feature function is the type-independent part of a feature
168 // function. Each feature function is associated with the descriptor that it is
169 // instantiated from.  The feature types associated with this feature function
170 // will be established by the time FeatureExtractor<>::Init() completes.
171 class GenericFeatureFunction {
172  public:
173   // A feature value that represents the absence of a value.
174   static constexpr FeatureValue kNone = -1;
175 
176   GenericFeatureFunction();
177   virtual ~GenericFeatureFunction();
178 
179   // Sets up the feature function. NB: FeatureTypes of nested functions are not
180   // guaranteed to be available until Init().
181   //
182   // Returns true on success, false otherwise.
Setup(TaskContext * context)183   SAFTM_MUST_USE_RESULT virtual bool Setup(TaskContext *context) {
184     return true;
185   }
186 
187   // Initializes the feature function. NB: The FeatureType of this function must
188   // be established when this method completes.
189   //
190   // Returns true on success, false otherwise.
Init(TaskContext * context)191   SAFTM_MUST_USE_RESULT virtual bool Init(TaskContext *context) { return true; }
192 
193   // Requests workspaces from a registry to obtain indices into a WorkspaceSet
194   // for any Workspace objects used by this feature function. NB: This will be
195   // called after Init(), so it can depend on resources and arguments.
RequestWorkspaces(WorkspaceRegistry * registry)196   virtual void RequestWorkspaces(WorkspaceRegistry *registry) {}
197 
198   // Appends the feature types produced by the feature function to types.  The
199   // default implementation appends feature_type(), if non-null.  Invalid
200   // before Init() has been called.
201   virtual void GetFeatureTypes(std::vector<FeatureType *> *types) const;
202 
203   // Returns the feature type for feature produced by this feature function. If
204   // the feature function produces features of different types this returns
205   // null.  Invalid before Init() has been called.
206   virtual FeatureType *GetFeatureType() const;
207 
208   // Returns value of parameter |name| from the feature function descriptor.
209   // If the parameter is not present, returns the indicated |default_value|.
210   string GetParameter(const string &name, const string &default_value) const;
211 
212   // Returns value of int parameter |name| from feature function descriptor.
213   // If the parameter is not present, or its value can't be parsed as an int,
214   // returns |default_value|.
215   int GetIntParameter(const string &name, int default_value) const;
216 
217   // Returns value of bool parameter |name| from feature function descriptor.
218   // If the parameter is not present, or its value is not "true" or "false",
219   // returns |default_value|.  NOTE: this method is case sensitive, it doesn't
220   // do any lower-casing.
221   bool GetBoolParameter(const string &name, bool default_value) const;
222 
223   // Returns the FEL function description for the feature function, i.e. the
224   // name and parameters without the nested features.
FunctionName()225   string FunctionName() const {
226     string output;
227     ToFELFunction(*descriptor_, &output);
228     return output;
229   }
230 
231   // Returns the prefix for nested feature functions. This is the prefix of this
232   // feature function concatenated with the feature function name.
SubPrefix()233   string SubPrefix() const {
234     return prefix_.empty() ? FunctionName() : prefix_ + "." + FunctionName();
235   }
236 
237   // Returns/sets the feature extractor this function belongs to.
extractor()238   const GenericFeatureExtractor *extractor() const { return extractor_; }
set_extractor(const GenericFeatureExtractor * extractor)239   void set_extractor(const GenericFeatureExtractor *extractor) {
240     extractor_ = extractor;
241   }
242 
243   // Returns/sets the feature function descriptor.
descriptor()244   const FeatureFunctionDescriptor *descriptor() const { return descriptor_; }
set_descriptor(const FeatureFunctionDescriptor * descriptor)245   void set_descriptor(const FeatureFunctionDescriptor *descriptor) {
246     descriptor_ = descriptor;
247   }
248 
249   // Returns a descriptive name for the feature function. The name is taken from
250   // the descriptor for the feature function. If the name is empty or the
251   // feature function is a variable the name is the FEL representation of the
252   // feature, including the prefix.
253   string name() const;
254 
255   // Returns the argument from the feature function descriptor. It defaults to
256   // 0 if the argument has not been specified.
argument()257   int argument() const {
258     return descriptor_->has_argument() ? descriptor_->argument() : 0;
259   }
260 
261   // Returns/sets/clears function name prefix.
prefix()262   const string &prefix() const { return prefix_; }
set_prefix(const string & prefix)263   void set_prefix(const string &prefix) { prefix_ = prefix; }
264 
265  protected:
266   // Returns the feature type for single-type feature functions.
feature_type()267   FeatureType *feature_type() const { return feature_type_; }
268 
269   // Sets the feature type for single-type feature functions.  This takes
270   // ownership of feature_type.  Can only be called once.
set_feature_type(FeatureType * feature_type)271   void set_feature_type(FeatureType *feature_type) {
272     SAFTM_CHECK_EQ(feature_type_, nullptr);
273     feature_type_ = feature_type;
274   }
275 
276  private:
277   // Feature extractor this feature function belongs to.  Not owned.  Set to a
278   // pointer != nullptr as soon as this object is created by Instantiate().
279   // Normal methods can safely assume this is != nullptr.
280   const GenericFeatureExtractor *extractor_ = nullptr;
281 
282   // Descriptor for feature function.  Not owned.  Set to a pointer != nullptr
283   // as soon as this object is created by Instantiate().  Normal methods can
284   // safely assume this is != nullptr.
285   const FeatureFunctionDescriptor *descriptor_ = nullptr;
286 
287   // Feature type for features produced by this feature function. If the
288   // feature function produces features of multiple feature types this is null
289   // and the feature function must return it's feature types in
290   // GetFeatureTypes().  Owned.
291   FeatureType *feature_type_ = nullptr;
292 
293   // Prefix used for sub-feature types of this function.
294   string prefix_;
295 };
296 
297 // Feature function that can extract features from an object.  Templated on
298 // two type arguments:
299 //
300 // OBJ:  The "object" from which features are extracted; e.g., a sentence.  This
301 //       should be a plain type, rather than a reference or pointer.
302 //
303 // ARGS: A set of 0 or more types that are used to "index" into some part of the
304 //       object that should be extracted, e.g. an int token index for a sentence
305 //       object.  This should not be a reference type.
306 template <class OBJ, class... ARGS>
307 class FeatureFunction
308     : public GenericFeatureFunction,
309       public RegisterableClass<FeatureFunction<OBJ, ARGS...> > {
310  public:
311   using Self = FeatureFunction<OBJ, ARGS...>;
312 
313   // Preprocesses the object.  This will be called prior to calling Evaluate()
314   // or Compute() on that object.
Preprocess(WorkspaceSet * workspaces,const OBJ * object)315   virtual void Preprocess(WorkspaceSet *workspaces, const OBJ *object) const {}
316 
317   // Appends features computed from the object and focus to the result.  The
318   // default implementation delegates to Compute(), adding a single value if
319   // available.  Multi-valued feature functions must override this method.
Evaluate(const WorkspaceSet & workspaces,const OBJ & object,ARGS...args,FeatureVector * result)320   virtual void Evaluate(const WorkspaceSet &workspaces, const OBJ &object,
321                         ARGS... args, FeatureVector *result) const {
322     FeatureValue value = Compute(workspaces, object, args...);
323     if (value != kNone) result->add(feature_type(), value);
324   }
325 
326   // Returns a feature value computed from the object and focus, or kNone if no
327   // value is computed.  Single-valued feature functions only need to override
328   // this method.
Compute(const WorkspaceSet & workspaces,const OBJ & object,ARGS...args)329   virtual FeatureValue Compute(const WorkspaceSet &workspaces,
330                                const OBJ &object, ARGS... args) const {
331     return kNone;
332   }
333 
334   // Instantiates a new feature function in a feature extractor from a feature
335   // descriptor.
336   //
337   // Returns a pointer to the newly-created object if everything goes well.
338   // Returns nullptr if the feature function could not be instantiated (e.g., if
339   // the function with that name is not registered; this usually happens because
340   // the relevant cc_library was not linked-in).
Instantiate(const GenericFeatureExtractor * extractor,const FeatureFunctionDescriptor * fd,const string & prefix)341   static Self *Instantiate(const GenericFeatureExtractor *extractor,
342                            const FeatureFunctionDescriptor *fd,
343                            const string &prefix) {
344     Self *f = Self::Create(fd->type());
345     if (f != nullptr) {
346       f->set_extractor(extractor);
347       f->set_descriptor(fd);
348       f->set_prefix(prefix);
349     }
350     return f;
351   }
352 
353  private:
354   // Special feature function class for resolving variable references. The type
355   // of the feature function is used for resolving the variable reference. When
356   // evaluated it will either get the feature value(s) from the variable portion
357   // of the feature vector, if present, or otherwise it will call the referenced
358   // feature extractor function directly to extract the feature(s).
359   class Reference;
360 };
361 
362 // Base class for features with nested feature functions. The nested functions
363 // are of type NES, which may be different from the type of the parent function.
364 // NB: NestedFeatureFunction will ensure that all initialization of nested
365 // functions takes place during Setup() and Init() -- after the nested features
366 // are initialized, the parent feature is initialized via SetupNested() and
367 // InitNested(). Alternatively, a derived classes that overrides Setup() and
368 // Init() directly should call Parent::Setup(), Parent::Init(), etc. first.
369 //
370 // Note: NestedFeatureFunction cannot know how to call Preprocess, Evaluate, or
371 // Compute, since the nested functions may be of a different type.
372 template <class NES, class OBJ, class... ARGS>
373 class NestedFeatureFunction : public FeatureFunction<OBJ, ARGS...> {
374  public:
375   using Parent = NestedFeatureFunction<NES, OBJ, ARGS...>;
376 
377   // Clean up nested functions.
~NestedFeatureFunction()378   ~NestedFeatureFunction() override { utils::STLDeleteElements(&nested_); }
379 
380   // By default, just appends the nested feature types.
GetFeatureTypes(std::vector<FeatureType * > * types)381   void GetFeatureTypes(std::vector<FeatureType *> *types) const override {
382     SAFTM_CHECK(!this->nested().empty())
383         << "Nested features require nested features to be defined.";
384     for (auto *function : nested_) function->GetFeatureTypes(types);
385   }
386 
387   // Sets up the nested features.
388   //
389   // Returns true on success, false otherwise.
Setup(TaskContext * context)390   SAFTM_MUST_USE_RESULT bool Setup(TaskContext *context) override {
391     bool success = CreateNested(this->extractor(), this->descriptor(), &nested_,
392                                 this->SubPrefix());
393     if (!success) return false;
394     for (auto *function : nested_) {
395       if (!function->Setup(context)) return false;
396     }
397     if (!SetupNested(context)) return false;
398     return true;
399   }
400 
401   // Sets up this NestedFeatureFunction specifically.
402   //
403   // Returns true on success, false otherwise.
SetupNested(TaskContext * context)404   SAFTM_MUST_USE_RESULT virtual bool SetupNested(TaskContext *context) {
405     return true;
406   }
407 
408   // Initializes the nested features.
409   //
410   // Returns true on success, false otherwise.
Init(TaskContext * context)411   SAFTM_MUST_USE_RESULT bool Init(TaskContext *context) override {
412     for (auto *function : nested_) {
413       if (!function->Init(context)) return false;
414     }
415     if (!InitNested(context)) return false;
416     return true;
417   }
418 
419   // Initializes this NestedFeatureFunction specifically.
420   //
421   // Returns true on success, false otherwise.
InitNested(TaskContext * context)422   SAFTM_MUST_USE_RESULT virtual bool InitNested(TaskContext *context) {
423     return true;
424   }
425 
426   // Gets all the workspaces needed for the nested functions.
RequestWorkspaces(WorkspaceRegistry * registry)427   void RequestWorkspaces(WorkspaceRegistry *registry) override {
428     for (auto *function : nested_) function->RequestWorkspaces(registry);
429   }
430 
431   // Returns the list of nested feature functions.
nested()432   const std::vector<NES *> &nested() const { return nested_; }
433 
434   // Instantiates nested feature functions for a feature function. Creates and
435   // initializes one feature function for each sub-descriptor in the feature
436   // descriptor.
437   //
438   // Returns true on success, false otherwise.
CreateNested(const GenericFeatureExtractor * extractor,const FeatureFunctionDescriptor * fd,std::vector<NES * > * functions,const string & prefix)439   SAFTM_MUST_USE_RESULT static bool CreateNested(
440       const GenericFeatureExtractor *extractor,
441       const FeatureFunctionDescriptor *fd, std::vector<NES *> *functions,
442       const string &prefix) {
443     for (int i = 0; i < fd->feature_size(); ++i) {
444       const FeatureFunctionDescriptor &sub = fd->feature(i);
445       NES *f = NES::Instantiate(extractor, &sub, prefix);
446       if (f == nullptr) return false;
447       functions->push_back(f);
448     }
449     return true;
450   }
451 
452  protected:
453   // The nested feature functions, if any, in order of declaration in the
454   // feature descriptor.  Owned.
455   std::vector<NES *> nested_;
456 };
457 
458 // Base class for a nested feature function that takes nested features with the
459 // same signature as these features, i.e. a meta feature. For this class, we can
460 // provide preprocessing of the nested features.
461 template <class OBJ, class... ARGS>
462 class MetaFeatureFunction
463     : public NestedFeatureFunction<FeatureFunction<OBJ, ARGS...>, OBJ,
464                                    ARGS...> {
465  public:
466   // Preprocesses using the nested features.
Preprocess(WorkspaceSet * workspaces,const OBJ * object)467   void Preprocess(WorkspaceSet *workspaces, const OBJ *object) const override {
468     for (auto *function : this->nested_) {
469       function->Preprocess(workspaces, object);
470     }
471   }
472 };
473 
474 // Template for a special type of locator: The locator of type
475 // FeatureFunction<OBJ, ARGS...> calls nested functions of type
476 // FeatureFunction<OBJ, IDX, ARGS...>, where the derived class DER is
477 // responsible for translating by providing the following:
478 //
479 // // Gets the new additional focus.
480 // IDX GetFocus(const WorkspaceSet &workspaces, const OBJ &object);
481 //
482 // This is useful to e.g. add a token focus to a parser state based on some
483 // desired property of that state.
484 template <class DER, class OBJ, class IDX, class... ARGS>
485 class FeatureAddFocusLocator
486     : public NestedFeatureFunction<FeatureFunction<OBJ, IDX, ARGS...>, OBJ,
487                                    ARGS...> {
488  public:
Preprocess(WorkspaceSet * workspaces,const OBJ * object)489   void Preprocess(WorkspaceSet *workspaces, const OBJ *object) const override {
490     for (auto *function : this->nested_) {
491       function->Preprocess(workspaces, object);
492     }
493   }
494 
Evaluate(const WorkspaceSet & workspaces,const OBJ & object,ARGS...args,FeatureVector * result)495   void Evaluate(const WorkspaceSet &workspaces, const OBJ &object, ARGS... args,
496                 FeatureVector *result) const override {
497     IDX focus =
498         static_cast<const DER *>(this)->GetFocus(workspaces, object, args...);
499     for (auto *function : this->nested()) {
500       function->Evaluate(workspaces, object, focus, args..., result);
501     }
502   }
503 
504   // Returns the first nested feature's computed value.
Compute(const WorkspaceSet & workspaces,const OBJ & object,ARGS...args)505   FeatureValue Compute(const WorkspaceSet &workspaces, const OBJ &object,
506                        ARGS... args) const override {
507     IDX focus =
508         static_cast<const DER *>(this)->GetFocus(workspaces, object, args...);
509     return this->nested()[0]->Compute(workspaces, object, focus, args...);
510   }
511 };
512 
513 // CRTP feature locator class. This is a meta feature that modifies ARGS and
514 // then calls the nested feature functions with the modified ARGS. Note that in
515 // order for this template to work correctly, all of ARGS must be types for
516 // which the reference operator & can be interpreted as a pointer to the
517 // argument. The derived class DER must implement the UpdateFocus method which
518 // takes pointers to the ARGS arguments:
519 //
520 // // Updates the current arguments.
521 // void UpdateArgs(const OBJ &object, ARGS *...args) const;
522 template <class DER, class OBJ, class... ARGS>
523 class FeatureLocator : public MetaFeatureFunction<OBJ, ARGS...> {
524  public:
525   // Feature locators have an additional check that there is no intrinsic type.
GetFeatureTypes(std::vector<FeatureType * > * types)526   void GetFeatureTypes(std::vector<FeatureType *> *types) const override {
527     SAFTM_CHECK_EQ(this->feature_type(), nullptr)
528         << "FeatureLocators should not have an intrinsic type.";
529     MetaFeatureFunction<OBJ, ARGS...>::GetFeatureTypes(types);
530   }
531 
532   // Evaluates the locator.
Evaluate(const WorkspaceSet & workspaces,const OBJ & object,ARGS...args,FeatureVector * result)533   void Evaluate(const WorkspaceSet &workspaces, const OBJ &object, ARGS... args,
534                 FeatureVector *result) const override {
535     static_cast<const DER *>(this)->UpdateArgs(workspaces, object, &args...);
536     for (auto *function : this->nested()) {
537       function->Evaluate(workspaces, object, args..., result);
538     }
539   }
540 
541   // Returns the first nested feature's computed value.
Compute(const WorkspaceSet & workspaces,const OBJ & object,ARGS...args)542   FeatureValue Compute(const WorkspaceSet &workspaces, const OBJ &object,
543                        ARGS... args) const override {
544     static_cast<const DER *>(this)->UpdateArgs(workspaces, object, &args...);
545     return this->nested()[0]->Compute(workspaces, object, args...);
546   }
547 };
548 
549 // Feature extractor for extracting features from objects of a certain class.
550 // Template type parameters are as defined for FeatureFunction.
551 template <class OBJ, class... ARGS>
552 class FeatureExtractor : public GenericFeatureExtractor {
553  public:
554   // Feature function type for top-level functions in the feature extractor.
555   typedef FeatureFunction<OBJ, ARGS...> Function;
556   typedef FeatureExtractor<OBJ, ARGS...> Self;
557 
558   // Feature locator type for the feature extractor.
559   template <class DER>
560   using Locator = FeatureLocator<DER, OBJ, ARGS...>;
561 
562   // Initializes feature extractor.
FeatureExtractor()563   FeatureExtractor() {}
564 
~FeatureExtractor()565   ~FeatureExtractor() override { utils::STLDeleteElements(&functions_); }
566 
567   // Sets up the feature extractor. Note that only top-level functions exist
568   // until Setup() is called. This does not take ownership over the context,
569   // which must outlive this.
570   //
571   // Returns true on success, false otherwise.
Setup(TaskContext * context)572   SAFTM_MUST_USE_RESULT bool Setup(TaskContext *context) {
573     for (Function *function : functions_) {
574       if (!function->Setup(context)) return false;
575     }
576     return true;
577   }
578 
579   // Initializes the feature extractor.  Must be called after Setup().  This
580   // does not take ownership over the context, which must outlive this.
581   //
582   // Returns true on success, false otherwise.
Init(TaskContext * context)583   SAFTM_MUST_USE_RESULT bool Init(TaskContext *context) {
584     for (Function *function : functions_) {
585       if (!function->Init(context)) return false;
586     }
587     if (!this->InitializeFeatureTypes()) return false;
588     return true;
589   }
590 
591   // Requests workspaces from the registry. Must be called after Init(), and
592   // before Preprocess(). Does not take ownership over registry. This should be
593   // the same registry used to initialize the WorkspaceSet used in Preprocess()
594   // and ExtractFeatures(). NB: This is a different ordering from that used in
595   // SentenceFeatureRepresentation style feature computation.
RequestWorkspaces(WorkspaceRegistry * registry)596   void RequestWorkspaces(WorkspaceRegistry *registry) {
597     for (auto *function : functions_) function->RequestWorkspaces(registry);
598   }
599 
600   // Preprocesses the object using feature functions for the phase.  Must be
601   // called before any calls to ExtractFeatures() on that object and phase.
Preprocess(WorkspaceSet * workspaces,const OBJ * object)602   void Preprocess(WorkspaceSet *workspaces, const OBJ *object) const {
603     for (Function *function : functions_) {
604       function->Preprocess(workspaces, object);
605     }
606   }
607 
608   // Extracts features from an object with a focus. This invokes all the
609   // top-level feature functions in the feature extractor. Only feature
610   // functions belonging to the specified phase are invoked.
ExtractFeatures(const WorkspaceSet & workspaces,const OBJ & object,ARGS...args,FeatureVector * result)611   void ExtractFeatures(const WorkspaceSet &workspaces, const OBJ &object,
612                        ARGS... args, FeatureVector *result) const {
613     result->reserve(this->feature_types());
614 
615     // Extract features.
616     for (int i = 0; i < functions_.size(); ++i) {
617       functions_[i]->Evaluate(workspaces, object, args..., result);
618     }
619   }
620 
621  private:
622   // Creates and initializes all feature functions in the feature extractor.
623   //
624   // Returns true on success, false otherwise.
InitializeFeatureFunctions()625   SAFTM_MUST_USE_RESULT bool InitializeFeatureFunctions() override {
626     // Create all top-level feature functions.
627     for (int i = 0; i < descriptor().feature_size(); ++i) {
628       const FeatureFunctionDescriptor &fd = descriptor().feature(i);
629       Function *function = Function::Instantiate(this, &fd, "");
630       if (function == nullptr) return false;
631       functions_.push_back(function);
632     }
633     return true;
634   }
635 
636   // Collect all feature types used in the feature extractor.
GetFeatureTypes(std::vector<FeatureType * > * types)637   void GetFeatureTypes(std::vector<FeatureType *> *types) const override {
638     for (int i = 0; i < functions_.size(); ++i) {
639       functions_[i]->GetFeatureTypes(types);
640     }
641   }
642 
643   // Top-level feature functions (and variables) in the feature extractor.
644   // Owned.
645   std::vector<Function *> functions_;
646 };
647 
648 }  // namespace mobile
649 }  // namespace nlp_saft
650 
651 #endif  // NLP_SAFT_COMPONENTS_COMMON_MOBILE_FEL_FEATURE_EXTRACTOR_H_
652