MedLinearModel.h

#ifndef __LINEAR_MODEL_H__
#define __LINEAR_MODEL_H__
#include <SerializableObject/SerializableObject/SerializableObject.h>
#include <MedAlgo/MedAlgo/PredictiveModel.h>
#include <MedAlgo/MedAlgo/MedAlgo.h>
 
using namespace std;
 
class MedLinearModel : public MedPredictor, public PredictiveModel
{
public:
    MedLinearModel();
    int init(map<string, string>& mapper) { return set_params(mapper); }
    int set_params(map<string, string>& mapper);
 
    subGradientFunction getSubGradients(); 
    subGradientFunction  getSubGradientsAUC(); 
    subGradientFunction  getSubGradientsSvm(); 
    double predict(const vector<float> &input) const;
    void predict(const vector<vector<float>> &inputs, vector<double> &preds) const;
    PredictiveModel *clone() const;
 
    void print(const vector<string> &signalNames) const;
    void set_normalization(const vector<float> &meanShift, const vector<float> &factors); 
    void apply_normalization(vector<vector<float>> &input) const; 
    void get_normalization(vector<float> &meanShift, vector<float> &factors) const;
    
    //Set Loss Fucntions to learn:
    double(*loss_function)(const vector<double> &got, const vector<float> &y, const vector<float> *weights);
    double(*loss_function_step)(const vector<double> &, const vector<float> &, const vector<double> &, const vector<float> *);
    int sample_count; 
    int tot_steps; 
    double learning_rate; 
    float block_num; 
    bool norm_l1; 
    int print_steps; 
    bool print_model; 
    int poly_degree; 
    int min_cat; 
    bool print_auc = false; 
    bool normalize = true; 
 
    //MedPredictor Api:
    int Learn(float *x, float *y, const float *w, int nsamples, int nftrs);
    int Predict(float *x, float *&preds, int nsamples, int nftrs) const;
 
    void calc_feature_importance(vector<float> &features_importance_scores,
        const string &general_params, const MedFeatures *features);
 
    ADD_CLASS_NAME(MedLinearModel)
    ADD_SERIALIZATION_FUNCS(classifier_type, model_params, _meanShift, _factor, model_features, features_count, normalize, poly_degree)
 
private:
    vector<float> _meanShift;
    vector<float> _factor;
    bool mark_learn_finish;
};
 
 
double _linear_loss_target_auc(const vector<double> &preds, const vector<float> &y, const vector<float> *weights);
double _linear_loss_step_auc(const vector<double> &preds, const vector<float> &y, const vector<double> &params, const vector<float> *weights);
double _linear_loss_step_auc_fast(const vector<double> &preds, const vector<float> &y, const vector<float> *weights);
double _linear_loss_target_work_point(const vector<double> &preds, const vector<float> &y, const vector<float> *weights);
double _linear_loss_step_work_point(const vector<double> &preds, const vector<float> &y, const vector<double> &model_params, const vector<float> *weights);
double _linear_loss_target_rmse(const vector<double> &preds, const vector<float> &y, const vector<float> *weights);
double _linear_loss_step_rmse(const vector<double> &preds, const vector<float> &y, const vector<double> &model_params, const vector<float> *weights);
double _linear_loss_target_svm(const vector<double> &preds, const vector<float> &y, const vector<float> *weights);
double _linear_loss_step_svm(const vector<double> &preds, const vector<float> &y, const vector<double> &model_params, const vector<float> *weights);
 
MEDSERIALIZE_SUPPORT(MedLinearModel)
 
#endif