Merge pull request #104 from DavidIkov/fix_for_inference

Fix for inference_evaluation and f_score

Author: DavidIkov

examples/mnist-learn/main.cpp

@@ -19,7 +19,7 @@
  * limitations under the License.
  */
 
-#include <knp/framework/inference_evaluation/classification.h>
+#include <knp/framework/inference_evaluation/classification/processor.h>
 
 #include <filesystem>
 #include <fstream>
@@ -80,7 +80,7 @@ int main(int argc, char** argv)
     std::cout << get_time_string() << ": inference finished  -- output spike count is " << spikes.size() << std::endl;
 
     // Evaluate results.
-    inference_evaluation::InferenceResultForClass::InferenceResultsProcessor inference_processor;
+    inference_evaluation::InferenceResultsProcessor inference_processor;
     inference_processor.process_inference_results(spikes, dataset);
 
     inference_processor.write_inference_results_to_stream_as_csv(std::cout);

knp/base-framework/CMakeLists.txt

@@ -73,8 +73,8 @@ knp_add_library("${PROJECT_NAME}-core"
     impl/sonata/types/additive_delta_synapse.cpp
     impl/data_processing/classification/dataset.cpp
     impl/data_processing/classification/image.cpp
-    impl/inference_evaluation/classification.cpp
     impl/inference_evaluation/perfomance_metrics.cpp
+    impl/inference_evaluation/classification/processor.cpp
     impl/observer.cpp
     ${${PROJECT_NAME}_headers}
     ALIAS KNP::BaseFramework::Core

…/inference_evaluation/classification.cpp …_evaluation/classification/processor.cppknp/base-framework/impl/inference_evaluation/classification.cpp renamed to knp/base-framework/impl/inference_evaluation/classification/processor.cpp knp/base-framework/impl/inference_evaluation/classification.cpp renamed to knp/base-framework/impl/inference_evaluation/classification/processor.cpp

@@ -20,7 +20,7 @@
  */
 
 #include <knp/core/messaging/messaging.h>
-#include <knp/framework/inference_evaluation/classification.h>
+#include <knp/framework/inference_evaluation/classification/processor.h>
 #include <knp/framework/inference_evaluation/perfomance_metrics.h>
 
 #include <algorithm>
@@ -30,14 +30,16 @@
 namespace knp::framework::inference_evaluation::classification
 {
 
-class InferenceResultForClass::InferenceResultsProcessor::EvaluationHelper
+class EvaluationHelper
 {
 public:
-    explicit EvaluationHelper(const knp::framework::data_processing::classification::Dataset &dataset);
+    explicit EvaluationHelper(
+        const knp::framework::data_processing::classification::Dataset &dataset,
+        std::vector<InferenceResult> &inference_results);
 
     void process_spikes(const knp::core::messaging::SpikeData &firing_neuron_indices, size_t step);
 
-    [[nodiscard]] std::vector<InferenceResultForClass> process_inference_predictions() const;
+    [[nodiscard]] std::vector<InferenceResult> process_inference_predictions() const;
 
 private:
     struct Prediction
@@ -53,18 +55,19 @@ class InferenceResultForClass::InferenceResultsProcessor::EvaluationHelper
     std::vector<size_t> class_votes_;
 
     const knp::framework::data_processing::classification::Dataset &dataset_;
+    std::vector<InferenceResult> &inference_results_;
 };
 
 
-InferenceResultForClass::InferenceResultsProcessor::EvaluationHelper::EvaluationHelper(
-    const knp::framework::data_processing::classification::Dataset &dataset)
-    : class_votes_(dataset.get_amount_of_classes(), 0), dataset_(dataset)
+EvaluationHelper::EvaluationHelper(
+    const knp::framework::data_processing::classification::Dataset &dataset,
+    std::vector<InferenceResult> &inference_results)
+    : class_votes_(dataset.get_amount_of_classes(), 0), dataset_(dataset), inference_results_(inference_results)
 {
 }
 
 
-void InferenceResultForClass::InferenceResultsProcessor::EvaluationHelper::process_spikes(
-    const knp::core::messaging::SpikeData &firing_neuron_indices, size_t step)
+void EvaluationHelper::process_spikes(const knp::core::messaging::SpikeData &firing_neuron_indices, size_t step)
 {
     for (auto i : firing_neuron_indices) ++class_votes_[i % dataset_.get_amount_of_classes()];
     if (!((step + 1) % dataset_.get_steps_per_frame()))
@@ -86,10 +89,9 @@ void InferenceResultForClass::InferenceResultsProcessor::EvaluationHelper::proce
 }
 
 
-std::vector<InferenceResultForClass>
-InferenceResultForClass::InferenceResultsProcessor::EvaluationHelper::process_inference_predictions() const
+std::vector<InferenceResult> EvaluationHelper::process_inference_predictions() const
 {
-    std::vector<InferenceResultForClass> prediction_results(dataset_.get_amount_of_classes());
+    std::vector<InferenceResult> prediction_results(dataset_.get_amount_of_classes());
     for (size_t i = 0; i < predictions_.size(); ++i)
     {
         auto const &prediction = predictions_[i];
@@ -113,11 +115,11 @@ InferenceResultForClass::InferenceResultsProcessor::EvaluationHelper::process_in
 }
 
 
-void InferenceResultForClass::InferenceResultsProcessor::process_inference_results(
+void InferenceResultsProcessor::process_inference_results(
     const std::vector<knp::core::messaging::SpikeMessage> &spikes,
     knp::framework::data_processing::classification::Dataset const &dataset)
 {
-    EvaluationHelper helper(dataset);
+    EvaluationHelper helper(dataset, inference_results_);
     knp::core::messaging::SpikeData firing_neuron_indices;
     auto spikes_iter = spikes.begin();
 
@@ -137,11 +139,10 @@ void InferenceResultForClass::InferenceResultsProcessor::process_inference_resul
 }
 
 
-void InferenceResultForClass::InferenceResultsProcessor::write_inference_results_to_stream_as_csv(
-    std::ostream &results_stream)
+void InferenceResultsProcessor::write_inference_results_to_stream_as_csv(std::ostream &results_stream)
 {
     results_stream << "CLASS,TOTAL_VOTES,TRUE_POSITIVES,FALSE_NEGATIVES,FALSE_POSITIVES,TRUE_NEGATIVES,PRECISION,"
-                      "RECALL,PREVALENCE,ACCURACY,F_MEASURE\n";
+                      "RECALL,PREVALENCE,ACCURACY,F_SCORE\n";
     for (size_t label = 0; label < inference_results_.size(); ++label)
     {
         auto const &prediction = inference_results_[label];
@@ -153,12 +154,12 @@ void InferenceResultForClass::InferenceResultsProcessor::write_inference_results
         const float accuracy = get_accuracy(
             prediction.true_positives_, prediction.false_negatives_, prediction.false_positives_,
             prediction.true_negatives_);
-        const float f_measure = get_f_measure(precision, recall);
+        const float f_score = get_f_score(precision, recall);
 
         results_stream << label << ',' << prediction.get_total_votes() << ',' << prediction.true_positives_ << ','
                        << prediction.false_negatives_ << ',' << prediction.false_positives_ << ','
                        << prediction.true_negatives_ << ',' << precision << ',' << recall << ',' << prevalence << ','
-                       << accuracy << ',' << f_measure << std::endl;
+                       << accuracy << ',' << f_score << std::endl;
     }
 }
 

knp/base-framework/impl/inference_evaluation/perfomance_metrics.cpp

@@ -55,7 +55,7 @@ float get_accuracy(size_t true_positives, size_t false_negatives, size_t false_p
 }
 
 
-float get_f_measure(float precision, float recall)
+float get_f_score(float precision, float recall)
 {
     if (precision * recall == 0) return 0.F;
     return 2.F * precision * recall / (precision + recall);

knp/base-framework/include/knp/framework/inference_evaluation/classification.h

@@ -0,0 +1,69 @@
+/**
+ * @file processor.h
+ * @brief Processing inference results.
+ * @kaspersky_support D. Postnikov
+ * @date 05.09.2025
+ * @license Apache 2.0
+ * @copyright © 2025 AO Kaspersky Lab
+ *
+ * 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.
+ */
+
+#pragma once
+
+#include <knp/core/messaging/messaging.h>
+#include <knp/framework/data_processing/classification/image.h>
+
+#include <vector>
+
+#include "result.h"
+
+
+namespace knp::framework::inference_evaluation::classification
+{
+
+/**
+ * @details A class to process inference results.
+ */
+class KNP_DECLSPEC InferenceResultsProcessor
+{
+public:
+    /**
+     * @brief Process inference results. Suited for classification models.
+     * @param spikes All spikes from inference.
+     * @param dataset Dataset.
+     */
+    void process_inference_results(
+        const std::vector<knp::core::messaging::SpikeMessage> &spikes,
+        const knp::framework::data_processing::classification::Dataset &dataset);
+
+    /**
+     * @brief Put inference results for each class to a stream in form of csv.
+     * @param results_stream stream for output.
+     */
+    void write_inference_results_to_stream_as_csv(std::ostream &results_stream);
+
+    /**
+     * @brief Get inference results.
+     * @return Inference results.
+     */
+    [[nodiscard]] const std::vector<InferenceResult> &get_inference_results() const { return inference_results_; }
+
+private:
+    /**
+     * @brief Processed inference results.
+     */
+    std::vector<InferenceResult> inference_results_;
+};
+
+}  // namespace knp::framework::inference_evaluation::classification

knp/base-framework/include/knp/framework/inference_evaluation/classification/processor.h

@@ -0,0 +1,61 @@
+/**
+ * @file result.h
+ * @brief Structure to hold inference results.
+ * @kaspersky_support D. Postnikov
+ * @date 16.07.2025
+ * @license Apache 2.0
+ * @copyright © 2025 AO Kaspersky Lab
+ *
+ * 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.
+ */
+
+#pragma once
+
+#include <knp/core/impexp.h>
+
+namespace knp::framework::inference_evaluation::classification
+{
+
+/**
+ * @brief Processed inference result for single class.
+ */
+struct KNP_DECLSPEC InferenceResult
+{
+    /**
+     * @brief Amount of times model, that is supposed  to predict dog, predicted dog when it is a dog.
+     */
+    size_t true_positives_ = 0;
+
+    /**
+     * @brief Amount of times model, that is supposed to predict dog, predicted not a dog when it is a dog.
+     */
+    size_t false_negatives_ = 0;
+
+    /**
+     * @brief Amount of times model, that is supposed to predict dog, predicted dog when it is not a dog.
+     */
+    size_t false_positives_ = 0;
+
+    /**
+     * @brief Amount of times model, that is supposed to predict dog, predicted not a dog when it is a not a dog.
+     */
+    size_t true_negatives_ = 0;
+
+    /**
+     * @brief Shortcut for getting total votes.
+     * @return Total votes.
+     */
+    [[nodiscard]] size_t get_total_votes() const { return true_positives_ + false_negatives_ + false_positives_; }
+};
+
+}  // namespace knp::framework::inference_evaluation::classification