features.py

from deapdata import EEG_CHANELS, PHYS_DATA_CHANELS, FREQ, STATISTICS, BANDS, ASYM_BANDS, ASYM_ELECTRODE_PAIRS
from metadata_data import ParticipantRatings
from phys_data import ParticipantSignalsFeatures, ClearedEEGParticipantFeatures
from eventlogger import EventLogger

from datetime import datetime
import numpy as np
import matplotlib.pyplot as plt
from matplotlib import cm
import os
import pandas as pd
import random
from scipy.stats import skew

class ParticipantFeatureVectors:
    def __init__(self, nParticipant, from_file = False):
        self.nParticipant = nParticipant
        self.events = EventLogger()
        self.ratings = ParticipantRatings(self.nParticipant)
        self.featureVectors = {}  #  {1:{'Fp1_theta': 2453476, 'Fp1_slow_alpha': 482418, ... , 'avgSkinRes': 69'}}
        self.featureDF = pd.DataFrame  #  use special function
        self.Y = {}
        for trial in range(1,41):
            self.featureVectors[trial] = {}
            self.Y[trial] = 0
        #  seed and variables for splitting data
        self.randomSeed = random.randint(1, 1000000)
        self.X_train = {}
        self.X_test = {}
        self.X_validation = {}
        self.Y_train = {}
        self.Y_test = {}
        self.Y_validation = {}
        #  if we not use precomputed feature from file - compute it from raw signals
        if not from_file:
            self.ratings = ParticipantRatings(self.nParticipant)
            self.physSignalsFeatures = ParticipantSignalsFeatures(self.nParticipant)
            self.physSignalsFeatures.computeFeatures(range(1, 41), range(1, 33), BANDS, FREQ, 0, 8063,
                                                     ASYM_ELECTRODE_PAIRS, ASYM_BANDS)
        else:
            self.loadFeatureVectorsFromCSV()
            self.convertFeatureVectorsToDataFrame()

    def fillFeatureVectors(self):
        self.addEEGSpectralToFeatureVector()
        self.addEEGAsymetryToFeatureVector()
        self.addGSRToFeatureVector()

    def createYVector(self, yType = 'f'):
        self.Y = {}
        if yType == 'f':
            if self.ratings.getFamiliarity() != []:
                for trial in self.featureVectors.keys():
                    self.Y[trial] = self.ratings.familiarity.reset_index(drop=True).iloc[trial-1]
        elif yType == 'a':
            self.ratings.getArousal()
            for trial in self.featureVectors.keys():
                self.Y[trial] = self.ratings.arousal.reset_index(drop=True).iloc[trial - 1]
        elif yType == 'v':
            self.ratings.getValence()
            for trial in self.featureVectors.keys():
                self.Y[trial] = self.ratings.valence.reset_index(drop=True).iloc[trial - 1]
        elif yType == 'l':
            self.ratings.getLiking()
            for trial in self.featureVectors.keys():
                self.Y[trial] = self.ratings.liking.reset_index(drop=True).iloc[trial - 1]
        elif yType == 'd':
            self.ratings.getDominance()
            for trial in self.featureVectors.keys():
                self.Y[trial] = self.ratings.dominance.reset_index(drop=True).iloc[trial - 1]
        elif yType == 'save':
            for type in ['f', 'a', 'v', 'd', 'l']:
                self.createYVector(yType=type)
                self.saveYVectorToCSV(yType=type)
        else:
            print('No such Y vector type')

    def convertFeatureVectorsToDataFrame(self):
        self.featureDF = pd.DataFrame.from_dict(self.featureVectors, orient='index')

    def addEEGSpectralToFeatureVector(self):
        for trial in self.featureVectors.keys():
            for electrode in range(len(EEG_CHANELS)):
                for band in BANDS.keys():
                    feature_name = EEG_CHANELS[electrode] + '_' + band
                    self.featureVectors[trial][feature_name] = \
                    self.physSignalsFeatures.spectralEEGFeatures[trial][electrode+1][band]

    def addEEGAsymetryToFeatureVector(self):
        for trial in self.featureVectors.keys():
            for electrodePair in ASYM_ELECTRODE_PAIRS:
                leftE, rightE = electrodePair
                for band in ASYM_BANDS.keys():
                    feature_name = EEG_CHANELS[leftE-1] + '-' + EEG_CHANELS[rightE-1] + '_' + band
                    self.featureVectors[trial][feature_name] = \
                    self.physSignalsFeatures.spectralEEGAsymetry[trial][leftE][band]

    def addGSRToFeatureVector(self):
        for trial in self.featureVectors.keys():
            self.featureVectors[trial]['avgSkinRes'] = self.physSignalsFeatures.averageSkinResistance[trial]

    def randomSplitSetForTraining(self, train=70, test=30, validation=0, seed=None):
        '''
        Split self.featureVectors and self.Y in random train, test and validation parts in a given proportions
        :param train: proportion of train part, default 70
        :param test: proportion of test part, default 30
        :param validation: proportion of validation part, default 0
        :param seed: seed for random, if None - there will be self.randomSeed used
        :return: self.X_train, self.X_test, self.X_validation, self.Y_train, self.Y_test, self.Y_validation - feature
        and target variable set divided in a given proportion
        '''
        #  init and fill proportion variables
        self.trainPart = train
        self.testPart = test
        self.validationPart = validation

        #  get random sample from feature vector index of test proportion length
        if seed is None:
            seed = self.randomSeed
        random.seed(seed)
        train_index = self.featureVectors.keys()
        test_index = random.sample(train_index,
                                   round(len(self.featureVectors.keys()) * test / (train + test + validation)))
        test_index.sort() #  to have ordered index
        train_index = [item for item in train_index if item not in test_index]

        #  Not all model requires validation set, so we could skip it creation in such case
        if validation != 0:
            validation_index = random.sample(train_index, round(
                len(self.featureVectors.keys()) * validation / (train + test + validation)))
            validation_index.sort()
            train_index = [item for item in train_index if item not in validation_index]

        #  create dict by created index
        self.X_train = {key: self.featureVectors[key] for key in train_index}
        try:
            self.Y_train = {key: self.Y[key] for key in train_index}
        except KeyError:
            errorMsg = 'Participant {} self.Y is empty, so no data for {}'.format(str(self.nParticipant),
                                                                                  'self.Y_train')
            self.events.addEvent(204, errorMsg)
            print(errorMsg)
        self.X_test = {key: self.featureVectors[key] for key in test_index}
        try:
            self.Y_test = {key: self.Y[key] for key in test_index}
        except KeyError:
            errorMsg = 'Participant {} self.Y is empty, so no data for {}'.format(str(self.nParticipant), 'self.Y_text')
            self.events.addEvent(204, errorMsg)
            print(errorMsg)
        if validation != 0:
            self.X_validation = {key: self.featureVectors[key] for key in validation_index}
            try:
                self.Y_validation = {key: self.Y[key] for key in validation_index}
            except KeyError:
                errorMsg = 'Participant {} self.Y is empty, so no data for {}'.format(str(self.nParticipant),
                                                                                      'self.Y_validation')
                self.events.addEvent(204, errorMsg)
                print(errorMsg)
        return self.X_train, self.Y_train, self.X_test, self.Y_test, self.X_validation, self.Y_validation

    def saveSplitedSetToCSV(self, seed=None):
        if seed is None:
            seed = self.randomSeed
        names = ['X_train', 'X_test', 'X_validation', 'Y_train', 'Y_test', 'Y_validation']
        sets = [self.X_train, self.X_test, self.X_validation, self.Y_train, self.Y_test, self.Y_validation]
        pathname = 'training_data/seed={}&train={}&test={}&val={}/'.format(str(seed), str(self.trainPart),
                                                                           str(self.testPart), str(self.validationPart))
        if not os.path.isdir(pathname):
            os.makedirs(pathname)
        for name, set in zip(names, sets):
            if set != {}:
                file_name = '{1}_{2}_{0}.csv'.format(str(seed), str(self.nParticipant), name)
                pd.DataFrame.from_dict(set, orient='index').to_csv(pathname+file_name)

    def saveFeatureVectorToCSV(self):
        filename = 'feature_vectors/FV{}.csv'.format(str(self.nParticipant))
        pd.DataFrame.from_dict(self.featureVectors, orient='index').to_csv(filename)

    def saveYVectorToCSV(self, yType):
        filename = 'feature_vectors/YV'+str(self.nParticipant)+yType+'.csv'
        pd.DataFrame.from_dict(self.Y, orient='index').to_csv(filename)

    #  in most cases we no need to recalculate features from data, so it's necessary to load the previously computed
    #  (and saved to *.csv) featureVectors
    def loadFeatureVectorsFromCSV(self):
        filename = 'feature_vectors/FV{}.csv'.format(str(self.nParticipant))
        self.featureVectors = pd.DataFrame.from_csv(filename).to_dict(orient='index')


def createHugeFeatureVector(seed, from_file=False):
    if from_file:
        if not os.path.isfile('feature_vectors/huge/'+str(seed)+'X_train.npy'):
            saveHugeFeatureVector(seed)
        X_huge_train = np.load('feature_vectors/huge/' + str(seed) + 'X_train.npy')
        Y_huge_train = np.load('feature_vectors/huge/' + str(seed) + 'Y_train.npy')
        X_huge_test = np.load('feature_vectors/huge/' + str(seed) + 'X_test.npy')
        Y_huge_test = np.load('feature_vectors/huge/' + str(seed) + 'Y_test.npy')
    else:
        X_huge_train = []
        Y_huge_train = []
        X_huge_test = []
        Y_huge_test = []
        for part in range(1, 33):
            #create participant object, split feature vectors,
            p = ParticipantFeatureVectors(part, from_file=True)
            p.createYVector()
            X_train, Y_train, X_test, Y_test, X_validation, Y_validation = p.randomSplitSetForTraining(70, 30, seed=seed)
            p.saveSplitedSetToCSV(seed=seed)
            X_train = np.array([list(X_train[trial].values()) for trial in X_train.keys()])
            Y_train = np.array([Y_train[trial] for trial in Y_train.keys()])
            X_test = np.array([list(X_test[trial].values()) for trial in X_test.keys()])
            Y_test = np.array([Y_test[trial] for trial in Y_test.keys()])
            # use only Participant with necessary label vector
            if Y_train != []:
                X_huge_train.extend(X_train)
                Y_huge_train.extend(Y_train)
                X_huge_test.extend(X_test)
                Y_huge_test.extend(Y_test)
            else:
                print('Y_train for {} is empty'.format(part))
    return X_huge_train, Y_huge_train, X_huge_test, Y_huge_test


class Features:
    def __init__(self, dataframe):
        self.dataframe = dataframe
        self.normDF = pd.DataFrame
        self.StandardScaler = pd.DataFrame
        self.RobustScaler = pd.DataFrame
        self.MinMaxScaler = pd.DataFrame
        self.normCoef = {}
        self.features = self.dataframe.columns
        self.statistics = {}
        self.statisticsDF = pd.DataFrame

    def getFeatures(self):
        return self.features

    def setStatistics(self):
        for feature in self.features:
            featureData = self.dataframe[feature]
            zipper = zip(STATISTICS, [np.min(featureData), np.max(featureData), np.round(np.mean(featureData)),
                                      np.round(np.median(featureData)), np.round(np.std(featureData)),
                                      np.round(skew(featureData), 3)])
            self.statistics[feature] = {statKey: statValue for (statKey, statValue) in zipper}
        self.statisticsDF = pd.DataFrame.from_dict(self.statistics, orient='index', columns=STATISTICS)

    def getStatistics(self, feature):
        return self.statistics[feature]

    def getFeatureVector(self, feature):
        return self.dataframe[feature]

    def normalizeData(self):
        self.normDF = self.dataframe
        self.setStatistics()
        for feature in self.features:
            range = self.statistics[feature]['max'] - self.statistics[feature]['min']
            self.normDF[feature] = (self.dataframe[feature] - self.statistics[feature]['mean']) / range
        return self.normDF

    def convertCategorialFeatureTo01(self, feature):
        series = pd.Series(self.dataframe[feature])
        print('There such unique categorial values: {}'.format(sorted(series.unique())))
        convert_dict = {entry: [0] * sorted(series.unique()).index(entry) + [1] + [0] * (
                len(series.unique()) - 1 - sorted(series.unique()).index(entry)) for entry in series.unique()}
        series = series.map(convert_dict)
        return series

    def covarianceMatrix(self, features):
        pass # TODO визуализация матрицы ковариации

    #  TODO Класс делегирует методы визуализации в модуль plotting. Класс делегирует методы сериализации и
    #  TODO десереализации своих атрибутов в служебный класс ObjectSerialize.


class ParticipantEEGFeatureVectors(ParticipantFeatureVectors):
    def __init__(self, nParticipant, from_file = False):
        self.nParticipant = nParticipant
        self.events = EventLogger()
        self.ratings = ParticipantRatings(self.nParticipant)
        self.featureVectors = {}  #
        self.featureDF = pd.DataFrame  # use special function
        self.Y = {}
        for trial in range(1, 41):
            self.featureVectors[trial] = {}
            self.Y[trial] = 0
        #  seed and variables for splitting data
        self.randomSeed = random.randint(1, 1000000)
        self.X_train = {}
        self.X_test = {}
        self.X_validation = {}
        self.Y_train = {}
        self.Y_test = {}
        self.Y_validation = {}
        #  if we not use precomputed feature from file - compute it from raw signals
        if not from_file:
            self.featureVectors = self.unpackEEGDataToVectors()
        else:
            self.loadFeatureVectorsFromCSV()
            self.convertFeatureVectorsToDataFrame()

    def unpackEEGDataToVectors(self):
        eegSpectral = ClearedEEGParticipantFeatures(self.nParticipant).computeEEGSpectralPower()
        for trial in range(1, 41):
            for electrode in range(len(EEG_CHANELS)):
                for timeSegment in eegSpectral[trial][electrode+1].keys():
                    for band in ASYM_BANDS.keys():
                        feature_name = EEG_CHANELS[electrode] + '_' + band + '_' + str(timeSegment)
                        self.featureVectors[trial][feature_name] = eegSpectral[trial][electrode+1][timeSegment][band]
        return self.featureVectors

    def saveFeatureVectorToCSV(self):
        filename = 'feature_vectors/4/FV{}.csv'.format(str(self.nParticipant))
        pd.DataFrame.from_dict(self.featureVectors, orient='index').to_csv(filename)

    #  in most cases we no need to recalculate features from data, so it's necessary to load the previously computed
    #  (and saved to *.csv) featureVectors
    def loadFeatureVectorsFromCSV(self):
        filename = 'feature_vectors/4/FV{}.csv'.format(str(self.nParticipant))
        self.featureVectors = pd.DataFrame.from_csv(filename).to_dict(orient='index')


def saveHugeFeatureVector(seed):
    X_huge_train, Y_huge_train, X_huge_test, Y_huge_test = createHugeFeatureVector(seed)
    np.save('feature_vectors/huge/' + str(seed) + 'X_train', X_huge_train)
    np.save('feature_vectors/huge/' + str(seed) + 'Y_train', Y_huge_train)
    np.save('feature_vectors/huge/' + str(seed) + 'X_test', X_huge_test)
    np.save('feature_vectors/huge/' + str(seed) + 'Y_test', Y_huge_test)


def convertCategorialFeatureTo01(series):
    print('There such unique categorial values: {}'.format(sorted(series.unique())))
    convert_dict = {entry: [0] * sorted(series.unique()).index(entry) + [1] + [0] * (
            len(series.unique()) - 1 - sorted(series.unique()).index(entry)) for entry in series.unique()}
    series = series.map(convert_dict)
    return series


def combineFV():
    # features = []
    for type in ['arousal', 'dominance', 'liking', 'valence']:
        Ys = []
        for part in range(1, 33):
            # features.append(pd.DataFrame.from_csv('feature_vectors/FV'+str(part)+'.csv'))
            Ys.append(pd.DataFrame.from_csv('feature_vectors/YV' + str(part) + type[0] + '.csv'))
        result = pd.concat(Ys)
        print(result)
        filename = 'feature_vectors/huge/YV_' + type + '.csv'
        result.to_csv(filename)
    return result

def alltogether():
    df = []
    for n in range(1, 33):
        if n not in [2, 15, 23]:
            df.append(pd.DataFrame.from_csv('feature_vectors/4/FV{}.csv'.format(str(n))))
    result = pd.concat(df)
    result.to_csv('feature_vectors/4/FV_all_f.csv')


if __name__ == "__main__":
    pass
    # alltogether()


    # p.saveFeatureVectorToCSV()
    # pass
    # y = pd.read_csv('feature_vectors/huge/YV_familiarity.csv', index_col=0).astype(int)
    # print(y)
    # y.replace({1: 0, 2: 0, 3: 1, 4: 1, 5: 1}, inplace=True)
    # print(y)
    # y.to_csv('feature_vectors/huge/YV_familiarity_binary.csv')


    # starttime = datetime.today()
    # print('Star for {} participant at {}'.format(str(2), str(starttime)))
    # p = ParticipantFeatureVectors(2)
    # finishtime = datetime.today()
    # print('ended after '+str(finishtime-starttime))
    #
    # firststarttime = datetime.today()
    # print('Star at {}'.format(str(firststarttime)))
    # for i in range(1, 33):
    #     starttime = datetime.today()
    #     print('Star for {} participant at {}'.format(str(i), str(starttime)))
    #     p = ParticipantFeatureVectors(i)
    #     p.fillFeatureVectors()
    #     p.saveFeatureVectorToCSV()
    #     p.createYVector('save')
    #     finishtime = datetime.today()
    #     print('ended after ' + str(finishtime - starttime))
    # print('Last ended after ' + str(finishtime - firststarttime))