SciCodePile/SciCode-Domain-Code

1"keyword","repo_name","file_path","file_extension","file_size","line_count","content","language"
2"Biochemistry","Bin-Cao/TCLRmodel","Template/Execution template/template.py",".py","5138","104","#coding=utf-83from TCLR import TCLRalgorithm as model4 5 6""""""7    :param correlation : {'PearsonR(+)','PearsonR(-)',''MIC','R2'}，default PearsonR(+).8            Methods:9            * PearsonR: (+)(-). for linear relationship.10            * MIC for no-linear relationship.11            * R2 for no-linear relationship.12 13    :param tolerance_list: constraints imposed on features, default is null14            list shape in two dimensions, viz., [[constraint_1,tol_1],[constraint_2,tol_2]...]15            constraint_1, constraint_2 （string） are the feature name ; 16            tol_1, tol_2 （float）are feature's tolerance ratios;17            relative variation range of features must be within the tolerance;18            example: tolerance_list = [['feature_name1',0.2],['feature_name2',0.1]].19    20    :param gpl_dummyfea: dummy features in gpleran regression, default is null21            list shape in one dimension, viz., ['feature_name1','feature_name2',...]22            dummy features : 'feature_name1','feature_name2',... are not used anymore in gpleran regression     23            24    :param minsize : a int number (default=3), minimum unique values for linear features of data on each leaf.25    26    :param threshold : a float (default=0.9), less than or equal to 1, default 0.95 for PearsonR.27            In the process of dividing the dataset, the smallest relevant index allowed in the you research.28            To avoid overfitting, threshold = 0.5 is suggested for MIC 0.5.29    30    :param mininc : Minimum expected gain of objective function (default=0.01)31    32    :param split_tol : a float (default=0.8), constrained features value shound be narrowed in a minmimu ratio of split_tol on split path33 34    :param gplearn : Whether to call the embedded gplearn package of TCLR to regress formula (default=False).35    36    :param population_size : integer, optional (default=500), the number of programs in each generation.37    38    :param generations : integer, optional (default=100),the number of generations to evolve.39 40    :param verbose : int, optional (default=0). Controls the verbosity of the evolution building process.41    42    :param metric : str, optional (default='mean absolute error')43            The name of the raw fitness metric. Available options include:44            - 'mean absolute error'.45            - 'mse' for mean squared error.46            - 'rmse' for root mean squared error.47            - 'pearson', for Pearson's product-moment correlation coefficient.48            - 'spearman' for Spearman's rank-order correlation coefficient.49    50    :param function_set : iterable, optional (default=['add', 'sub', 'mul', 'div', 'log', 'sqrt', 51                                               'abs', 'neg','inv','sin','cos','tan', 'max', 'min'])52            The functions to use when building and evolving programs. This iterable can include strings 53            to indicate either individual functions as outlined below.54            Available individual functions are:55            - 'add' : addition, arity=2.56            - 'sub' : subtraction, arity=2.57            - 'mul' : multiplication, arity=2.58            - 'div' : protected division where a denominator near-zero returns 1.,59                arity=2.60            - 'sqrt' : protected square root where the absolute value of the61                argument is used, arity=1.62            - 'log' : protected log where the absolute value of the argument is63                used and a near-zero argument returns 0., arity=1.64            - 'abs' : absolute value, arity=1.65            - 'neg' : negative, arity=1.66            - 'inv' : protected inverse where a near-zero argument returns 0.,67                arity=1.68            - 'max' : maximum, arity=2.69            - 'min' : minimum, arity=2.70            - 'sin' : sine (radians), arity=1.71            - 'cos' : cosine (radians), arity=1.72            - 'tan' : tangent (radians), arity=1.73 74    Algorithm Patent No. : 2021SR1951267, China75    Reference : Domain knowledge guided interpretive machine learning ——  Formula discovery for the oxidation behavior of Ferritic-Martensitic steels in supercritical water. Bin Cao et al., 2022, JMI, journal paper.76    DOI : 10.20517/jmi.2022.0477""""""78 79 80dataSet = ""testdata.csv""81correlation = 'PearsonR(+)'82tolerance_list = [83    ['E_Cr_split_feature_1',0.001],84]85 86gpl_dummyfea = ['ln(t)_split_feature_4',]87minsize = 388threshold = 0.989mininc = 0.0190split_tol = 0.891gplearn = True92population_size = 50093generations = 10094verbose = 1 95metric = 'mean absolute error'96function_set = ['add', 'sub', 'mul', 'div', 'log', 'sqrt', 'abs', 'neg','inv','sin','cos','tan', 'max', 'min']97 98 99model.start(filePath = dataSet, correlation = correlation, tolerance_list = tolerance_list, gpl_dummyfea = gpl_dummyfea, minsize = minsize, threshold = threshold,100            mininc = mininc ,split_tol = split_tol, gplearn = gplearn,  population_size = population_size,101            generations = generations,verbose = verbose, metric =metric, function_set =function_set)102 103 104 105","Python"
106"Biochemistry","Bin-Cao/TCLRmodel","TCLR/TCLRalgorithm.py",".py","38792","861","""""""107    Tree Classifier for Linear Regression (TCLR) 108    Author : Bin CAO (binjacobcao@gmail.com) 109 110    TCLR is a new tree model proposed by Professor T-Y Zhang and Mr. Bin Cao et al. for capturing the functional relationships 111    between features and target variables. The model partitions the feature space into a set of rectangles, with each partition112    embodying a specific function. This approach is conceptually simple, yet powerful for distinguishing mechanisms. The entire113    feature space is divided into disjointed unit intervals by hyperplanes parallel to the coordinate axes. Within each partition,114    the target variable y is modeled as a linear function of a feature xj (j = 1,⋯,m), which is the linear function used in our studied problem.115    116    Patent No. : 2021SR1951267, China117    Reference : Domain knowledge guided interpretive machine learning ——  Formula discovery for the oxidation behavior of Ferritic-Martensitic steels in supercritical water. Bin Cao et al., 2022, JMI, journal paper.118    DOI : 10.20517/jmi.2022.04119""""""120 121import math122import re123from tabnanny import check124from textwrap import indent125import time126import copy127import os128import warnings129import random130from typing import List131import numpy as np132import pandas as pd133from graphviz import Digraph134from scipy import stats135from gplearn import genetic136from minepy import MINE137 138 139# Define the basic structure of a Tree Model - Node140class Node:141    def __init__(self, data):142        self.data = data143        self.lc = None144        self.rc = None145        self.slope = None146        self.intercept = None147        self.size = data.shape[0]148        self.R = 0149        self.bestFeature = 0150        self.bestValue = 0151        self.leaf_no = -1152 153 154def start(filePath, correlation='PearsonR(+)', minsize=3, threshold=0.95, mininc=0.01, split_tol = 0.8, epochs = 5,random_seed=42, Generate_Features = True, tolerance_list = None , weight=True,155         gplearn = False, gpl_dummyfea = None, population_size = 500, generations = 100, verbose = 1, 156         metric = 'mean absolute error',157         function_set = ['add', 'sub', 'mul', 'div', 'log', 'sqrt', 'abs', 'neg','inv','sin','cos','tan',]):158    159    """"""160    Tree Classifier for Linear Regression (TCLR) 161 162    TCLR is a new tree model proposed by Professor T-Y Zhang and Mr. Bin Cao et al. for capturing the functional relationships 163    between features and target variables. The model partitions the feature space into a set of rectangles, with each partition164    embodying a specific function. This approach is conceptually simple, yet powerful for distinguishing mechanisms. The entire165    feature space is divided into disjointed unit intervals by hyperplanes parallel to the coordinate axes. Within each partition,166    the target variable y is modeled as a linear function of a feature xj (j = 1,⋯,m), which is the linear function used in our studied problem.167    168    Patent No. : 2021SR1951267, China169    Reference : Domain knowledge guided interpretive machine learning ——  Formula discovery for the oxidation behavior of Ferritic-Martensitic steels in supercritical water. Bin Cao et al., 2022, JMI, journal paper.170    DOI : 10.20517/jmi.2022.04171 172    :param correlation : {'PearsonR(+)','PearsonR(-)',''MIC','R2'}, default PearsonR(+).173            Methods:174            * PearsonR: (+)(-). for linear relationship.175            * MIC for no-linear relationship.176            * R2 for no-linear relationship.177 178            The evaluation factor for capture the functional relationship between feature and response179            1>180            PearsonR:181            Pearson correlation coefficient, also known as Pearson's r, the Pearson product-moment correlation coefficient.182            PearsonR is a measure of linear correlation between two sets of data. 183            PearsonR = Cov(X,Y) / (sigmaX * sigmaY)184        185            2>186            MIC:187            The maximal information coefficient (MIC). MIC captures a wide range of associations both functional and not, 188            and for functional relationships provides a score that roughly equals the coefficient of determination (R2) of 189            the data relative to the regression function.  MIC belongs to a larger class of maximal information-based 190            nonparametric exploration (MINE) statistics for identifying and classifying relationship.  191            Reference : Reshef, D. N., Reshef, Y. A., Finucane, H. K., Grossman, S. R., McVean, G., Turnbaugh, P. J., ... 192            and Sabeti, P. C. (2011). Detecting novel associations in large data sets. science, 334(6062), 1518-1524.193            194            3>195            R2:196            In statistics, the coefficient of determination, denoted R2 or r2 and pronounced ""R squared"", 197            is the proportion of the variation in the dependent variable that is predictable from the independent variable(s).198            t is a statistic used in the context of statistical models whose main purpose is either the prediction of future 199            outcomes or the testing of hypotheses, on the basis of other related information. It provides a measure of how well200            observed outcomes are replicated by the model, based on the proportion of total variation of outcomes explained by the model.201            Definition from Wikipedia : https://en.wikipedia.org/wiki/Coefficient_of_determination202            R2 = 1 - SSres / SStot. Its value may be a negative one for poor correlation.203 204  205    :param minsize : 206            a int number (default=3), minimum unique values for linear features of data on each leaf.207    208    :param threshold : 209            a float (default=0.9), less than or equal to 1, default 0.95 for PearsonR.210            In the process of dividing the dataset, the smallest relevant index allowed in the you research.211            To avoid overfitting, threshold = 0.5 is suggested for MIC 0.5.212    213    :param mininc : Minimum expected gain of objective function (default=0.01)214 215    :param split_tol : a float (default=0.8), constrained features value shound be narrowed in a minmimu ratio of split_tol on split path216 217    :param epochs : an integer (default=5), see parameter Generate_Features (below)218 219    :param random_seed :  an integer (default=42), see parameter Generate_Features (below)220 221    :param Generate_Features : boole (default=True). When Generate_Features = True, TCLR will generate new features by operating  222        the ['+','-','*'] on original features. Iterating [param : epoachs] times, and each time generating 3 new features.  [param : random_seed]223        is used to control the randomness.224        When Generate_Features = False, TCLR will apply the original features225 226    :param tolerance_list: 227            constraints imposed on features, default is null228            list shape in two dimensions, viz., [['feature_name1',tol_1],['feature_name2',tol_2]...]229            'feature_name1', 'feature_name2' (string) are names of input features;230            tol_1, tol_2 (float, between 0 to 1) are feature's tolerance ratios;231            the variations of feature values on each leaf must be in the tolerance;232            if tol_1 = 0, the value of feature 'feature_name1' must be a constant on each leaf,233            if tol_1 = 1, there is no constraints on value of feature 'feature_name1';234            example: tolerance_list = [['feature_name1',0.2],['feature_name2',0.1]].235 236    :param weight:237            The weight of the gain function, default is True.238            When weight is True: linear_gain = R(father node) - ( W_l * R(left child node) + W_r * R(right child node)) / 2239            Where W_l is the ratio of the number of samples in the left child node to the total number of samples ;240                  W_r is the ratio of the number of samples in the right child node to the total number of samples.241            When weight is False: linear_gain = R(father node) - ( R(left child node) + R(right child node)) / 2242 243    :param gplearn : Whether to call the embedded gplearn package of TCLR to regress formula (default=False).244    245    :param gpl_dummyfea: 246            dummy features in gpleran regression, default is null247            list shape in one dimension, viz., ['feature_name1','feature_name2',...]248            dummy features : 'feature_name1','feature_name2',... are not used anymore in gpleran regression 249 250    :param population_size : integer, optional (default=500), the number of programs in each generation.251    252    :param generations : integer, optional (default=100),the number of generations to evolve.253 254    :param verbose : int, optional (default=0). Controls the verbosity of the evolution building process.255    256    :param metric : 257            str, optional (default='mean absolute error')258            The name of the raw fitness metric. Available options include:259            - 'mean absolute error'.260            - 'mse' for mean squared error.261            - 'rmse' for root mean squared error.262            - 'pearson', for Pearson's product-moment correlation coefficient.263            - 'spearman' for Spearman's rank-order correlation coefficient.264    265    :param function_set : 266            iterable, optional (default=['add', 'sub', 'mul', 'div', 'log', 'sqrt', 267                                               'abs', 'neg','inv','sin','cos','tan', 'max', 'min'])268            The functions to use when building and evolving programs. This iterable can include strings 269            to indicate either individual functions as outlined below.270            Available individual functions are:271            - 'add' : addition, arity=2.272            - 'sub' : subtraction, arity=2.273            - 'mul' : multiplication, arity=2.274            - 'div' : protected division where a denominator near-zero returns 1.,275                arity=2.276            - 'sqrt' : protected square root where the absolute value of the277                argument is used, arity=1.278            - 'log' : protected log where the absolute value of the argument is279                used and a near-zero argument returns 0., arity=1.280            - 'abs' : absolute value, arity=1.281            - 'neg' : negative, arity=1.282            - 'inv' : protected inverse where a near-zero argument returns 0.,283                arity=1.284            - 'max' : maximum, arity=2.285            - 'min' : minimum, arity=2.286            - 'sin' : sine (radians), arity=1.287            - 'cos' : cosine (radians), arity=1.288            - 'tan' : tangent (radians), arity=1.289    290    Exampel :291    #coding=utf-8292    from TCLR import TCLRalgorithm as model293 294    dataSet = ""testdata.csv""295    correlation = 'PearsonR(+)'296    minsize = 3297    threshold = 0.9298    mininc = 0.01299    split_tol = 0.8300 301    model.start(filePath = dataSet, correlation = correlation, minsize = minsize, threshold = threshold,302                mininc = mininc ,split_tol = split_tol,)303    """"""304 305    os.makedirs('Segmented', exist_ok=True)306    # global var. for statisticaling  results307    global record308    record = 0309    timename = time.localtime(time.time())310    namey, nameM, named, nameh, namem = timename.tm_year, timename.tm_mon, timename.tm_mday, timename.tm_hour, timename.tm_min311 312    read_csvData = pd.read_csv(filePath)313 314    input_csvData = read_csvData.iloc[:,:-2]315    if Generate_Features == True:316        # cal an appropriate value of batch317        if len(input_csvData) - 1 <= 3:318            batch = 1319        else:320            batch = 3321        # generate new dataset322        for epoch in range(epochs):323            # for increasing the randomness324            random_seed += 1325            input_csvData = generate_random_features(input_csvData,[column for column in input_csvData],batch,random_seed)326           327        input_csvData = input_csvData.assign(linear_X=read_csvData.iloc[:,-2])328        csvData = input_csvData.assign(linear_Y=read_csvData.iloc[:,-1])329 330    else:331        csvData = read_csvData332    333    334    copy_csvData = copy.deepcopy(csvData)335    copy_csvData['slope'] = None336    copy_csvData['intercept'] = None337    copy_csvData[correlation] = None338    copy_csvData.to_csv('Segmented/all_dataset.csv', index=False)339 340    feats = [column for column in csvData]341    csvData = np.array(csvData)342    root, _ = createTree(csvData, csvData, feats, 0, correlation,tolerance_list, minsize, threshold, mininc, split_tol,weight)343    344    print('All non-image results have been successfully saved!')345    print('#'*80,'\n')346   347    # excute gplearn 348    if gplearn == True :349        if correlation == 'MIC' or correlation == 'R2':350            print('{name} is a non-linear correlation metrics'.format(name = correlation ))351            print('This is illegal, linear slopes are only allowed to generate when PearsonR is chosen')352        elif correlation == 'PearsonR(+)' or correlation == 'PearsonR(-)':353            sr_data = pd.read_csv('Segmented/all_dataset.csv')354            sr_featurname = sr_data.columns355            sr_data = np.array(sr_data)356 357            if gpl_dummyfea == None: 358 359                gpmodel = genetic.SymbolicRegressor(360                    population_size = population_size, generations = generations, 361                    verbose = verbose,feature_names = sr_featurname[:-4],function_set = function_set,362                    metric = metric363                    )364                formula = gpmodel.fit(sr_data[:,:-4], sr_data[:,-3])365                score = gpmodel.score(sr_data[:,:-4], sr_data[:,-3])366                print( 'slope = ' + str(formula))367 368            else:369                # fea_num --> fea_loc370                dummyfea = []371                for i in range(len(gpl_dummyfea)):372                    index = feats.index(gpl_dummyfea[i])373                    dummyfea.append(index)374                # remove fea_loc375                index_array = [i for i in range(len(sr_featurname)-4)]376                for i in range(len(gpl_dummyfea)):377                    index_array.remove(dummyfea[i])378                379                gpmodel = genetic.SymbolicRegressor(380                    population_size = population_size, generations = generations, 381                    verbose = verbose,feature_names = sr_featurname[index_array],function_set = function_set,382                    metric = metric383                    )384                formula = gpmodel.fit(sr_data[:,index_array], sr_data[:,-3])385                score = gpmodel.score(sr_data[:,index_array], sr_data[:,-3])386                print( 'slope = ' + str(formula))387 388          389            with open(os.path.join('Segmented', 'A_formula derived by gplearn.txt'), 'w') as wfid:390                    print('Formula : ', file=wfid)391                    print(str(formula), file=wfid)392                    print('Fitness : ', file=wfid)393                    print(str(metric) + ' = ' + str(score), file=wfid)394                    print('\n', file=wfid)395                    print('#'*80, file=wfid)396                    print('Symbols annotation:', file=wfid)397                    print('- add : addition, arity=2.', file=wfid)398                    print('- sub : subtraction, arity=2.', file=wfid) 399                    print('- mul : multiplication, arity=2.', file=wfid) 400                    print('- div : protected division where a denominator near-zero returns 1.', file=wfid) 401                    print('- sqrt : protected square root where the absolute value of the argument is used.', file=wfid) 402                    print('- log : protected log where the absolute value of the argument is used.', file=wfid) 403                    print('- abs : absolute value, arity=1.', file=wfid) 404                    print('- neg : negative, arity=1.', file=wfid) 405                    print('- inv : protected inverse where a near-zero argument returns 0.', file=wfid)  406                    print('- max : maximum, arity=2.', file=wfid) 407                    print('- sin : sine (radians), arity=1.', file=wfid) 408                    print('- cos : cosine (radians), arity=1.', file=wfid)409                    print('- tan : tangent (radians), arity=1.', file=wfid)  410                 411                        412            413    elif gplearn == False:414        pass415    416 417    try:418        # generate figure in pdf419        warnings.filterwarnings('ignore')420        dot = Digraph(comment='Result of TCLR')421        render('A', root, dot, feats)422        dot.render(423            'Result of TCLR {year}.{month}.{day}-{hour}.{minute}'.format(year=namey, month=nameM, day=named, hour=nameh,424                                                                        minute=namem))425        return True426    except :427        print('Can not generate the Tree plot !')428        print('Please ensure that the executable files of Graphviz are present on your system.')429        print('See : https://github.com/Bin-Cao/TCLRmodel/tree/main/User%20Guide')430        return True 431  432 433 434# Capture the functional relationships between features and target435# Partitions the feature space into a set of rectangles,436def createTree(dataSet, ori_dataset,feats, leaf_no, correlation,tolerance_list, minsize, threshold, mininc,split_tol,weight):437    # It is a  positive linear relationship438    if correlation == 'PearsonR(+)':439        node = Node(dataSet)440        # Initial R0441        bestR = PearsonR(dataSet[:, -2], dataSet[:, -1])442        node.R = bestR443        __slope = stats.linregress(dataSet[:, -2], dataSet[:, -1])[0]444        node.slope = __slope445        node.intercept = stats.linregress(dataSet[:, -2], dataSet[:, -1])[1]446        if bestR >= threshold and fea_tol(dataSet,ori_dataset,feats,tolerance_list) == True:447            node.leaf_no = leaf_no448            leaf_no += 1449            write_csv(node, feats, True, correlation)450            return node, leaf_no451        # Leave the last two columns of DataSet, a feature of interest and a response452        numFeatures = len(dataSet[0]) - 2453        splitSuccess = False454        bestFeature = -1455        bestValue = 0456 457        check_valve = False458        for i in range(numFeatures):459            featList = [example[i] for example in dataSet]460            uniqueVals = sorted(list(set(featList)))461            for value in range(len(uniqueVals) - 1):462                # constraints imposed on features  (greater tolerance in split process)463                if not fea_tol_split(dataSet,ori_dataset,feats,tolerance_list,split_tol):464                    continue465                subDataSetA, subDataSetB = splitDataSet(dataSet, i, uniqueVals[value])466                467                if np.unique(subDataSetA[:, -2]).size <= minsize - 1 or np.unique(468                        subDataSetB[:, -2]).size <= minsize - 1:469                    continue470                471                R = weight_gain(subDataSetA,subDataSetB,weight,0)472 473                if R - bestR >= mininc:474                    check_valve = True475                    splitSuccess = True476                    bestR = R477                    lc = subDataSetA478                    rc = subDataSetB479                    bestFeature = i480                    bestValue = uniqueVals[value]481                482        if check_valve == False:483            for i in range(numFeatures):484                featList = [example[i] for example in dataSet]485                uniqueVals = sorted(list(set(featList)))486                for value in range(len(uniqueVals) - 1):487                    subDataSetA, subDataSetB = splitDataSet(dataSet, i, uniqueVals[value])488                    489                    if np.unique(subDataSetA[:, -2]).size <= minsize - 1 or np.unique(490                            subDataSetB[:, -2]).size <= minsize - 1:491                        continue492 493                    R = weight_gain(subDataSetA,subDataSetB,weight,0)494 495                    if R - bestR >= mininc:496                        splitSuccess = True497                        bestR = R498                        lc = subDataSetA499                        rc = subDataSetB500                        bestFeature = i501                        bestValue = uniqueVals[value]502            else:503                pass504                        505 506        # The recursive boundary is unable to find a division node that can increase factor(R, MIC, R2) by mininc or more.507        if splitSuccess:508            node.lc, leaf_no = createTree(lc, ori_dataset,feats, leaf_no, correlation, tolerance_list,minsize, threshold, mininc,split_tol,weight)509            node.rc, leaf_no = createTree(rc, ori_dataset,feats, leaf_no, correlation,tolerance_list, minsize, threshold, mininc,split_tol,weight)510            node.bestFeature, node.bestValue = bestFeature, bestValue511 512        # This node is leaf513        if node.lc is None:514            node.leaf_no = leaf_no515            leaf_no += 1516            # determine if this node is to save in all_dataset.csv517            save_in_all = False518            if node.R >= threshold and fea_tol(node.data,ori_dataset,feats,tolerance_list) == True:519                save_in_all = True 520            write_csv(node, feats, save_in_all, correlation)521 522        return node, leaf_no523 524    # It is a negative linear relationship525    elif correlation == 'PearsonR(-)':526        node = Node(dataSet)527        bestR = PearsonR(dataSet[:, -2], dataSet[:, -1])528        node.R = bestR529        __slope = stats.linregress(dataSet[:, -2], dataSet[:, -1])[0]530        node.slope = __slope531        node.intercept = stats.linregress(dataSet[:, -2], dataSet[:, -1])[1]532        if bestR <= -threshold and fea_tol(dataSet,ori_dataset,feats,tolerance_list) == True:533            node.leaf_no = leaf_no534            leaf_no += 1535            write_csv(node, feats, True, correlation)536            return node, leaf_no537 538        numFeatures = len(dataSet[0]) - 2539        splitSuccess = False540        bestFeature = -1541        bestValue = 0542 543        check_valve = False544        for i in range(numFeatures):545            featList = [example[i] for example in dataSet]546            uniqueVals = sorted(list(set(featList)))547            for value in range(len(uniqueVals) - 1):548                # constraints imposed on features  (greater tolerance in split process)549                if not fea_tol_split(dataSet,ori_dataset,feats,tolerance_list,split_tol):550                    continue551                subDataSetA, subDataSetB = splitDataSet(dataSet, i, uniqueVals[value])552 553                if np.unique(subDataSetA[:, -2]).size <= minsize - 1 or np.unique(554                        subDataSetB[:, -2]).size <= minsize - 1:555                    continue556 557                558                R = weight_gain(subDataSetA,subDataSetB,weight,0)559 560                if R - bestR <= - mininc:561                    check_valve = True562                    splitSuccess = True563                    bestR = R564                    lc = subDataSetA565                    rc = subDataSetB566                    bestFeature = i567                    bestValue = uniqueVals[value]568 569        if check_valve == False:570            for i in range(numFeatures):571                featList = [example[i] for example in dataSet]572                uniqueVals = sorted(list(set(featList)))573                for value in range(len(uniqueVals) - 1):574                    subDataSetA, subDataSetB = splitDataSet(dataSet, i, uniqueVals[value])575 576                    if np.unique(subDataSetA[:, -2]).size <= minsize - 1 or np.unique(577                            subDataSetB[:, -2]).size <= minsize - 1:578                        continue579                    580                    R = weight_gain(subDataSetA,subDataSetB,weight,0)581 582                    if R - bestR <= - mininc:583                        splitSuccess = True584                        bestR = R585                        lc = subDataSetA586                        rc = subDataSetB587                        bestFeature = i588                        bestValue = uniqueVals[value]589        else: pass590 591        if splitSuccess:592            node.lc, leaf_no = createTree(lc, ori_dataset,feats, leaf_no, correlation,tolerance_list, minsize, threshold, mininc,split_tol,weight)593            node.rc, leaf_no = createTree(rc, ori_dataset,feats, leaf_no, correlation,tolerance_list, minsize, threshold, mininc,split_tol,weight)594            node.bestFeature, node.bestValue = bestFeature, bestValue595 596        if node.lc is None:597            node.leaf_no = leaf_no598            leaf_no += 1599            # determine if this node is to save in all_dataset.csv600            save_in_all = False601            if node.R >= threshold and fea_tol(node.data,ori_dataset,feats,tolerance_list) == True:602                save_in_all = True 603            write_csv(node, feats, save_in_all, correlation)604 605        return node, leaf_no606 607    elif correlation == 'MIC':608        node = Node(dataSet)609        bestR = MIC(dataSet[:, -2], dataSet[:, -1])610        node.R = bestR611        node.slope == None612        if bestR >= threshold and fea_tol(dataSet,ori_dataset,feats,tolerance_list) == True:613            node.leaf_no = leaf_no614            leaf_no += 1615            write_csv(node, feats, True, correlation)616            return node, leaf_no617 618        numFeatures = len(dataSet[0]) - 2619        splitSuccess = False620        bestFeature = -1621        bestValue = 0622 623        check_valve = False624        for i in range(numFeatures):625            featList = [example[i] for example in dataSet]626            uniqueVals = sorted(list(set(featList)))627            for value in range(len(uniqueVals) - 1):628                # constraints imposed on features  (greater tolerance in split process)629                if not fea_tol_split(dataSet,ori_dataset,feats,tolerance_list,split_tol):630                    continue631                subDataSetA, subDataSetB = splitDataSet(dataSet, i, uniqueVals[value])632                if np.unique(subDataSetA[:, -2]).size <= minsize - 1 or np.unique(633                        subDataSetB[:, -2]).size <= minsize - 1:634                    continue635                636                R = weight_gain(subDataSetA,subDataSetB,weight,1)637 638                if R - bestR >= mininc:639                    check_valve = True640                    splitSuccess = True641                    bestR = R642                    lc = subDataSetA643                    rc = subDataSetB644                    bestFeature = i645                    bestValue = uniqueVals[value]646 647        if check_valve == False:648            for i in range(numFeatures):649                featList = [example[i] for example in dataSet]650                uniqueVals = sorted(list(set(featList)))651                for value in range(len(uniqueVals) - 1):652                    subDataSetA, subDataSetB = splitDataSet(dataSet, i, uniqueVals[value])653                    if np.unique(subDataSetA[:, -2]).size <= minsize - 1 or np.unique(654                            subDataSetB[:, -2]).size <= minsize - 1:655                        continue656    657                    R = weight_gain(subDataSetA,subDataSetB,weight,1)658 659                    if R - bestR >= mininc:660                        splitSuccess = True661                        bestR = R662                        lc = subDataSetA663                        rc = subDataSetB664                        bestFeature = i665                        bestValue = uniqueVals[value]666        else: pass667 668        if splitSuccess:669            node.lc, leaf_no = createTree(lc, ori_dataset,feats, leaf_no, correlation,tolerance_list, minsize, threshold, mininc,split_tol,weight)670            node.rc, leaf_no = createTree(rc,ori_dataset, feats, leaf_no, correlation,tolerance_list, minsize, threshold, mininc,split_tol,weight)671            node.bestFeature, node.bestValue = bestFeature, bestValue672 673        if node.lc is None:674            node.leaf_no = leaf_no675            leaf_no += 1676            # determine if this node is to save in all_dataset.csv677            save_in_all = False678            if node.R >= threshold and fea_tol(node.data,ori_dataset,feats,tolerance_list) == True:679                save_in_all = True 680            write_csv(node, feats, save_in_all, correlation)681 682        return node, leaf_no683 684    elif correlation == 'R2':685        node = Node(dataSet)686        bestR = R2(dataSet[:, -2], dataSet[:, -1])687        node.R = bestR688        node.slope == None689        if bestR >= threshold and fea_tol(dataSet,ori_dataset,feats,tolerance_list) == True:690            node.leaf_no = leaf_no691            leaf_no += 1692            write_csv(node, feats, True, correlation)693            return node, leaf_no694 695        numFeatures = len(dataSet[0]) - 2696        splitSuccess = False697        bestFeature = -1698        bestValue = 0699 700        check_valve = False701        for i in range(numFeatures):702            featList = [example[i] for example in dataSet]703            uniqueVals = sorted(list(set(featList)))704            for value in range(len(uniqueVals) - 1):705                # constraints imposed on features  (greater tolerance in split process)706                if not fea_tol_split(dataSet,ori_dataset,feats,tolerance_list,split_tol):707                    continue708 709                subDataSetA, subDataSetB = splitDataSet(dataSet, i, uniqueVals[value])710 711                if np.unique(subDataSetA[:, -2]).size <= minsize - 1 or np.unique(712                        subDataSetB[:, -2]).size <= minsize - 1:713                    continue714 715                R = weight_gain(subDataSetA,subDataSetB,weight,2)716 717                if R - bestR >= mininc:718                    check_valve = True719                    splitSuccess = True720                    bestR = R721                    lc = subDataSetA722                    rc = subDataSetB723                    bestFeature = i724                    bestValue = uniqueVals[value]725 726        if check_valve == False:727            for i in range(numFeatures):728                featList = [example[i] for example in dataSet]729                uniqueVals = sorted(list(set(featList)))730 731                for value in range(len(uniqueVals) - 1):732                    if np.unique(subDataSetA[:, -2]).size <= minsize - 1 or np.unique(733                            subDataSetB[:, -2]).size <= minsize - 1:734                        continue735 736                    R = weight_gain(subDataSetA,subDataSetB,weight,2)737 738                    if R - bestR >= mininc:739                        splitSuccess = True740                        bestR = R741                        lc = subDataSetA742                        rc = subDataSetB743                        bestFeature = i744                        bestValue = uniqueVals[value]745        else: pass746        747        if splitSuccess:748            node.lc, leaf_no = createTree(lc, ori_dataset,feats, leaf_no, correlation, tolerance_list,minsize, threshold, mininc,split_tol)749            node.rc, leaf_no = createTree(rc, ori_dataset,feats, leaf_no, correlation,tolerance_list, minsize, threshold, mininc,split_tol)750            node.bestFeature, node.bestValue = bestFeature, bestValue751 752        if node.lc is None:753            node.leaf_no = leaf_no754            leaf_no += 1755            # determine if this node is to save in all_dataset.csv756            save_in_all = False757            if node.R >= threshold and fea_tol(node.data,feats,tolerance_list) == True:758                save_in_all = True 759            write_csv(node, feats, save_in_all, correlation)760 761        return node, leaf_no762 763 764def PearsonR(X, Y):765    xBar = np.mean(X)766    yBar = np.mean(Y)767    SSR = 0768    varX = 0769    varY = 0770    if len(X) > 1:771        for i in range(0, len(X)):772            diffXXBar = X[i] - xBar773            diffYYBar = Y[i] - yBar774            SSR += (diffXXBar * diffYYBar)775            varX += diffXXBar ** 2776            varY += diffYYBar ** 2777        SST = math.sqrt(varX * varY)778    else:779        SST = 1780        SSR = 0781    if SST == 0:782        return 0783    return SSR / SST784 785 786def MIC(X, Y):787    if len(X) > 0:788        mine = MINE(alpha=0.6, c=15)789        mine.compute_score(X, Y)790        return mine.mic()791    else:792        MICs = 0793        return MICs794 795 796def R2(X, Y):797    X = np.array(X)798    Y = np.array(Y)799    if len(X) > 0:800        a = (X - np.mean(Y)) ** 2801        SStot = np.sum(a)802        b = (X - Y) ** 2803        SSres = np.sum(b)804        r2 = 1 - SSres / SStot805        return r2806    else:807        r2 = -10808        return r2809 810 811# Split the DataSet in a specific node812def splitDataSet(dataSet, axis, value):813    retDataSetA = []814    retDataSetB = []815    for featVec in dataSet:816        if featVec[axis] <= value:817            retDataSetA.append(featVec)818        else:819            retDataSetB.append(featVec)820    return np.array(retDataSetA), np.array(retDataSetB)821 822def fea_tol(dataSet,ori_dataSet,feats,tolerance_list):823    if tolerance_list == None: return True824    else:825        record = 0826        for i in range(len(tolerance_list)):827            __feaname = tolerance_list[i][0]828            __tolratio = float(tolerance_list[i][1])829            index = feats.index(__feaname)830            if (dataSet[:,index].max() - dataSet[:,index].min()) / (ori_dataSet[:,index].max()- ori_dataSet[:,index].min()) <= __tolratio:831                record += 1832        if record == len(tolerance_list):833            return True834 835def fea_tol_split(dataSet,ori_dataSet,feats,tolerance_list,split_tol):836    if tolerance_list == None: return True837    else:838        record = 0839        for i in range(len(tolerance_list)):840            __feaname = tolerance_list[i][0]841            __tolratio = float(tolerance_list[i][1])842            criter = max(split_tol,__tolratio)843            index = feats.index(__feaname)844            if (dataSet[:,index].max() - dataSet[:,index].min()) / (ori_dataSet[:,index].max()- ori_dataSet[:,index].min()) <= criter:845                record += 1846        if record > int(0.5*len(tolerance_list)):847            return True848 849 850# Use graphviz to visualize the TCLR851def render(label, node, dot, feats):852    mark = ''853    if node.slope == None:854        mark = ""#="" + str(node.size) + "" , ρ="" + str(round(node.R, 3))855    else:856        mark = ""#="" + str(node.size) + "" , ρ="" + str(round(node.R, 3)) + ' , slope=' + str(857            round(node.slope, 3)) + ' , intercept=' + str(round(node.intercept, 3))858 859    if node.lc is None:860        mark = 'No_{}, '.format(node.leaf_no) + mark861    dot.node(label, mark)862 863    if node.lc is not None:864        render(label + 'A', node.lc, dot, feats)865        render(label + 'B', node.rc, dot, feats)866        dot.edge(label, label + 'A', feats[node.bestFeature] + ""≤"" + str(node.bestValue))867        dot.edge(label, label + 'B', feats[node.bestFeature] + "">"" + str(node.bestValue))868 869 870def write_csv(node, feats, save_in_all, correlation):871    global record872 873    frame = {}874    for i in range(len(feats)):875        frame[feats[i]] = node.data[:, i]876    frame = pd.DataFrame(frame)877 878    if node.slope == None:879        frame['slope'] = None880        frame['intercept'] = None881        frame[correlation] = np.repeat(node.R, node.size)882        frame.to_csv('Segmented/subdataset_{}.csv'.format(str(node.leaf_no)))883    else:884        frame['slope'] = node.slope885        frame['intercept'] = node.intercept886        frame[correlation] = np.repeat(node.R, node.size)887        frame.to_csv('Segmented/subdataset_{}.csv'.format(str(node.leaf_no)))888 889    frame.to_csv('Segmented/subdataset_{}.csv'.format(str(node.leaf_no)))890 891    if not save_in_all:  # do not save in the all_dataset.csv892        _all_dataset = pd.read_csv('./Segmented/all_dataset.csv')893        _all_dataset.drop(index=range(record, record+node.size), axis=0, inplace=True)894        _all_dataset.to_csv('Segmented/all_dataset.csv', index=False)895        return896 897    _all_dataset = pd.read_csv('./Segmented/all_dataset.csv')898    for item in range(len(frame.iloc[:, 0])):899        _all_dataset.iloc[item + record, :] = frame.iloc[item, :]900    record += len(frame.iloc[:, 0])901    _all_dataset.to_csv('Segmented/all_dataset.csv', index=False)902 903def weight_gain(subDataSetA,subDataSetB,weight,matrix):904    if matrix == 0:905        newRa = PearsonR(subDataSetA[:, -2], subDataSetA[:, -1])906        newRb = PearsonR(subDataSetB[:, -2], subDataSetB[:, -1])907    elif matrix == 1:908        newRa = MIC(subDataSetA[:, -2], subDataSetA[:, -1])909        newRb = MIC(subDataSetB[:, -2], subDataSetB[:, -1])910    elif matrix == 2:911        newRa = R2(subDataSetA[:, -2], subDataSetA[:, -1])912        newRb = R2(subDataSetB[:, -2], subDataSetB[:, -1])913  914 915    if weight == False:916        R = (newRa + newRb) / 2917        return R918    elif weight == True:919        weightRa = len(subDataSetA[:, -1]) / (len(subDataSetA[:, -1]) + len(subDataSetB[:, -1]))920        weightRb = len(subDataSetB[:, -1]) / (len(subDataSetA[:, -1]) + len(subDataSetB[:, -1]))921        R = weightRa * newRa + weightRb * newRb922        return R923    else:924        print('Parameter error | weight')925 926 927# code on 2023 May 9, Bin Cao928def generate_random_features(df: pd.DataFrame, 929                            feature_list: List[str],930                            num_combinations: int,931                             random_seed:int) -> pd.DataFrame:932    """"""933    randomly generates new feature combinations.934 935    :param df: DataFrame containing the original features.936    :param feature_list: List of original features.937    :param num_combinations: Number of combination features to generate.938    :return: DataFrame containing the new features.939    """"""940    new_features = []941    random.seed(random_seed)942    # randomly generates combination features943    for i in range(num_combinations):944        # randomly chooses two features945        f1 = random.choice(feature_list)946        f2 = random.choice(feature_list)947        948        # choose a operator949        op = random.choice(['+', '-', '*',])950 951        self_op1 = random.choice(['*1', '*2', '*3','*4','**2','**3'])952        self_op2 = random.choice(['*1', '*2', '*3','*4','**2','**3'])953        954        new_f1 = f'{f1} {self_op1}'955        new_f2 = f'{f2} {self_op2}'956 957        # new feature name958        new_feature = f'({new_f1} {op} {new_f2})'959        960        new_features.append(new_feature)961        962        # cal new features963        df[new_feature] = eval(f'(df[""{f1}""] {self_op1}) {op} (df[""{f2}""] {self_op2})')964    965    # reture DataFrame 966    return df","Python"
967"Biochemistry","Bin-Cao/TCLRmodel","Researches/Note.md",".md","38","2","# Relevant researches applied of TCLR968","Markdown"
969