@@ -38,32 +38,18 @@ void Classifier::initialize(const Real alpha)
3838{
3939 NTA_CHECK (alpha > 0 .0f );
4040 alpha_ = alpha;
41- dimensions_. clear () ;
41+ dimensions_ = 0 ;
4242 numCategories_ = 0u ;
4343 weights_.clear ();
4444}
4545
4646
47- PDF Classifier::infer (const SDR & pattern)
48- {
49- // Check input dimensions, or if this is the first time the Classifier has
50- // been used then initialize it with the given SDR's dimensions.
51- if ( dimensions_.empty () ) {
52- dimensions_ = pattern.dimensions ;
53- while ( weights_.size () < pattern.size ) {
54- weights_.push_back ( vector<Real>( numCategories_, 0 .0f ));
55- }
56- } else if ( pattern.dimensions != dimensions_ ) {
57- stringstream err_msg;
58- err_msg << " Classifier input SDR.dimensions mismatch: previously given SDR with dimensions ( "
59- for ( auto dim : dimensions_ )
60- { err_msg << dim << " "
61- err_msg << " ), now given SDR with dimensions ( "
62- for ( auto dim : pattern.dimensions )
63- { err_msg << dim << " "
64- err_msg << " )."
65- NTA_THROW << err_msg.str ();
66- }
47+ PDF Classifier::infer (const SDR & pattern) const {
48+ // Check input dimensions, or if this is the first time the Classifier is used and dimensions
49+ // are unset, return zeroes.
50+ NTA_CHECK ( dimensions_ != 0 )
51+ << " Classifier: must call `learn` before `infer`."
52+ NTA_ASSERT (pattern.size == dimensions_) << " Input SDR does not match previously seen size!"
6753
6854 // Accumulate feed forward input.
6955 PDF probabilities ( numCategories_, 0 .0f );
@@ -81,8 +67,19 @@ PDF Classifier::infer(const SDR & pattern)
8167
8268void Classifier::learn (const SDR &pattern, const vector<UInt> &categoryIdxList)
8369{
70+ // If this is the first time the Classifier is being used, weights are empty,
71+ // so we set the dimensions to that of the input `pattern`
72+ if ( dimensions_ == 0 ) {
73+ dimensions_ = pattern.size ;
74+ while ( weights_.size () < pattern.size ) {
75+ const auto initialEmptyWeights = PDF ( numCategories_, 0 .0f );
76+ weights_.push_back ( initialEmptyWeights );
77+ }
78+ }
79+ NTA_ASSERT (pattern.size == dimensions_) << " Input SDR does not match previously seen size!"
80+
8481 // Check if this is a new category & resize the weights table to hold it.
85- const auto maxCategoryIdx = *max_element (categoryIdxList.begin (), categoryIdxList.end ());
82+ const auto maxCategoryIdx = *max_element (categoryIdxList.cbegin (), categoryIdxList.cend ());
8683 if ( maxCategoryIdx >= numCategories_ ) {
8784 numCategories_ = maxCategoryIdx + 1 ;
8885 for ( auto & vec : weights_ ) {
@@ -93,7 +90,7 @@ void Classifier::learn(const SDR &pattern, const vector<UInt> &categoryIdxList)
9390 }
9491
9592 // Compute errors and update weights.
96- const vector<Real> error = calculateError_ (categoryIdxList, pattern);
93+ const auto & error = calculateError_ (categoryIdxList, pattern);
9794 for ( const auto & bit : pattern.getSparse () ) {
9895 for (size_t i = 0u ; i < numCategories_; i++) {
9996 weights_[bit][i] += alpha_ * error[i];
@@ -103,9 +100,8 @@ void Classifier::learn(const SDR &pattern, const vector<UInt> &categoryIdxList)
103100
104101
105102// Helper function to compute the error signal in learning.
106- std::vector<Real> Classifier::calculateError_ (
107- const std::vector<UInt> &categoryIdxList, const SDR &pattern)
108- {
103+ std::vector<Real64> Classifier::calculateError_ (const std::vector<UInt> &categoryIdxList,
104+ const SDR &pattern) const {
109105 // compute predicted likelihoods
110106 auto likelihoods = infer (pattern);
111107
@@ -165,56 +161,49 @@ void Predictor::reset() {
165161}
166162
167163
168- Predictions Predictor::infer (const UInt recordNum, const SDR &pattern)
169- {
170- updateHistory_ ( recordNum, pattern );
171-
164+ Predictions Predictor::infer (const SDR &pattern) const {
172165 Predictions result;
173166 for ( const auto step : steps_ ) {
174- result[ step] = classifiers_[ step] .infer ( pattern );
167+ result. insert ({ step, classifiers_. at ( step) .infer ( pattern )} );
175168 }
176169 return result;
177170}
178171
179172
180- void Predictor::learn (const UInt recordNum, const SDR &pattern,
173+ void Predictor::learn (const UInt recordNum, // TODO make recordNum optional, autoincrement as steps
174+ const SDR &pattern,
181175 const std::vector<UInt> &bucketIdxList)
182176{
183- updateHistory_ ( recordNum, pattern );
177+ checkMonotonic_ (recordNum);
178+
179+ // Update pattern history if this is a new record.
180+ const UInt lastRecordNum = recordNumHistory_.empty () ? 0 : recordNumHistory_.back ();
181+ if (recordNumHistory_.size () == 0u || recordNum > lastRecordNum) {
182+ patternHistory_.emplace_back ( pattern );
183+ recordNumHistory_.push_back (recordNum);
184+ if (patternHistory_.size () > steps_.back () + 1u ) { // steps_ are sorted, so steps_.back() is the "oldest/deepest" N-th step (ie 10 of [1,2,10])
185+ patternHistory_.pop_front ();
186+ recordNumHistory_.pop_front ();
187+ }
188+ }
184189
185190 // Iterate through all recently given inputs, starting from the furthest in the past.
186191 auto pastPattern = patternHistory_.begin ();
187192 auto pastRecordNum = recordNumHistory_.begin ();
188- for ( ; pastRecordNum != recordNumHistory_.end (); pastPattern++, pastRecordNum++ )
193+ for ( ; pastRecordNum != recordNumHistory_.cend (); pastPattern++, pastRecordNum++ )
189194 {
190195 const UInt nSteps = recordNum - *pastRecordNum;
191196
192197 // Update weights.
193198 if ( binary_search ( steps_.begin (), steps_.end (), nSteps )) {
194- classifiers_[ nSteps] .learn ( *pastPattern, bucketIdxList );
199+ classifiers_. at ( nSteps) .learn ( *pastPattern, bucketIdxList );
195200 }
196201 }
197202}
198203
199204
200- void Predictor::updateHistory_ (const UInt recordNum, const SDR & pattern)
201- {
205+ void Predictor::checkMonotonic_ (const UInt recordNum) const {
202206 // Ensure that recordNum increases monotonically.
203- UInt lastRecordNum = -1 ;
204- if ( not recordNumHistory_.empty () ) {
205- lastRecordNum = recordNumHistory_.back ();
206- if (recordNum < lastRecordNum) {
207- NTA_THROW << " The record number must increase monotonically."
208- }
209- }
210-
211- // Update pattern history if this is a new record.
212- if (recordNumHistory_.size () == 0u || recordNum > lastRecordNum) {
213- patternHistory_.emplace_back ( pattern );
214- recordNumHistory_.push_back (recordNum);
215- if (patternHistory_.size () > steps_.back () + 1u ) {
216- patternHistory_.pop_front ();
217- recordNumHistory_.pop_front ();
218- }
219- }
207+ const UInt lastRecordNum = recordNumHistory_.empty () ? 0 : recordNumHistory_.back ();
208+ NTA_CHECK (recordNum >= lastRecordNum) << " The record number must increase monotonically."
220209}
0 commit comments