├── Part Three ├── PCA.m ├── main.m ├── example.m ├── output.m ├── Untitled2.m ├── labeltest.m ├── linearSVM.m ├── biaccuracy.m ├── cheby2filter.m ├── output_final.m ├── test_output.m ├── totaloutput.m ├── usebagging.m ├── test_validation.m ├── Testdatapreprocessing.m ├── fisheriris.m ├── average_output.m ├── topoplotEEG.m └── trainClassifier.m ├── images ├── Model.png └── Model_Identification.png ├── common ├── __pycache__ │ ├── utils.cpython-36.pyc │ ├── classifier.cpython-36.pyc │ ├── datawrapper.cpython-36.pyc │ ├── optimizer.cpython-36.pyc │ └── regressor.cpython-36.pyc ├── regressor.py ├── ButterWorth_filter.py ├── standard.py ├── optimizer.py ├── CSP(Spatial Filter).py ├── utils.py ├── datawrapper.py └── classifier.py ├── .idea ├── modules.xml ├── misc.xml ├── 代码.iml └── workspace.xml ├── README.md ├── p300_CNN_1_GRU_A.py └── EEG_Preprocessor.py /Part Three/PCA.m: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/chenzhuoshuai/EEG-P300Speller-Toolkit/HEAD/Part Three/PCA.m -------------------------------------------------------------------------------- /Part Three/main.m: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/chenzhuoshuai/EEG-P300Speller-Toolkit/HEAD/Part Three/main.m -------------------------------------------------------------------------------- /images/Model.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/chenzhuoshuai/EEG-P300Speller-Toolkit/HEAD/images/Model.png -------------------------------------------------------------------------------- /Part Three/example.m: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/chenzhuoshuai/EEG-P300Speller-Toolkit/HEAD/Part Three/example.m -------------------------------------------------------------------------------- /Part Three/output.m: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/chenzhuoshuai/EEG-P300Speller-Toolkit/HEAD/Part Three/output.m -------------------------------------------------------------------------------- /Part Three/Untitled2.m: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/chenzhuoshuai/EEG-P300Speller-Toolkit/HEAD/Part Three/Untitled2.m -------------------------------------------------------------------------------- /Part Three/labeltest.m: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/chenzhuoshuai/EEG-P300Speller-Toolkit/HEAD/Part Three/labeltest.m -------------------------------------------------------------------------------- /Part Three/linearSVM.m: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/chenzhuoshuai/EEG-P300Speller-Toolkit/HEAD/Part Three/linearSVM.m -------------------------------------------------------------------------------- /Part Three/biaccuracy.m: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/chenzhuoshuai/EEG-P300Speller-Toolkit/HEAD/Part Three/biaccuracy.m -------------------------------------------------------------------------------- /Part Three/cheby2filter.m: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/chenzhuoshuai/EEG-P300Speller-Toolkit/HEAD/Part Three/cheby2filter.m -------------------------------------------------------------------------------- /Part Three/output_final.m: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/chenzhuoshuai/EEG-P300Speller-Toolkit/HEAD/Part Three/output_final.m -------------------------------------------------------------------------------- /Part Three/test_output.m: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/chenzhuoshuai/EEG-P300Speller-Toolkit/HEAD/Part Three/test_output.m -------------------------------------------------------------------------------- /Part Three/totaloutput.m: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/chenzhuoshuai/EEG-P300Speller-Toolkit/HEAD/Part Three/totaloutput.m -------------------------------------------------------------------------------- /Part Three/usebagging.m: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/chenzhuoshuai/EEG-P300Speller-Toolkit/HEAD/Part Three/usebagging.m -------------------------------------------------------------------------------- /Part Three/test_validation.m: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/chenzhuoshuai/EEG-P300Speller-Toolkit/HEAD/Part Three/test_validation.m -------------------------------------------------------------------------------- /images/Model_Identification.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/chenzhuoshuai/EEG-P300Speller-Toolkit/HEAD/images/Model_Identification.png -------------------------------------------------------------------------------- /Part Three/Testdatapreprocessing.m: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/chenzhuoshuai/EEG-P300Speller-Toolkit/HEAD/Part Three/Testdatapreprocessing.m -------------------------------------------------------------------------------- /common/__pycache__/utils.cpython-36.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/chenzhuoshuai/EEG-P300Speller-Toolkit/HEAD/common/__pycache__/utils.cpython-36.pyc -------------------------------------------------------------------------------- /common/__pycache__/classifier.cpython-36.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/chenzhuoshuai/EEG-P300Speller-Toolkit/HEAD/common/__pycache__/classifier.cpython-36.pyc -------------------------------------------------------------------------------- /common/__pycache__/datawrapper.cpython-36.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/chenzhuoshuai/EEG-P300Speller-Toolkit/HEAD/common/__pycache__/datawrapper.cpython-36.pyc -------------------------------------------------------------------------------- /common/__pycache__/optimizer.cpython-36.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/chenzhuoshuai/EEG-P300Speller-Toolkit/HEAD/common/__pycache__/optimizer.cpython-36.pyc -------------------------------------------------------------------------------- /common/__pycache__/regressor.cpython-36.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/chenzhuoshuai/EEG-P300Speller-Toolkit/HEAD/common/__pycache__/regressor.cpython-36.pyc -------------------------------------------------------------------------------- /.idea/modules.xml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | -------------------------------------------------------------------------------- /.idea/misc.xml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 6 | 7 | -------------------------------------------------------------------------------- /Part Three/fisheriris.m: -------------------------------------------------------------------------------- 1 | load fisheriris 2 | data=meas; 3 | groups=ismember(species,'setosa'); 4 | [train,test]=crossvalind('holdOut',groups); 5 | model1=svmtrain(data(train,:),groups(train,:)); 6 | model2=svmtrain(data(test,:),groups(test,:)); 7 | classes1=svmclassify(model1,data); 8 | classes2=svmclassify(model2,data); 9 | 10 | classes=(classes1+classes2)/2; 11 | -------------------------------------------------------------------------------- /.idea/代码.iml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 12 | -------------------------------------------------------------------------------- /Part Three/average_output.m: -------------------------------------------------------------------------------- 1 | output_averaged=ones(size(labels_validation_total,1),1); 2 | for i=1:size(labels_validation_total,1) 3 | output_averaged(i)=mean(labels_validation_total(i,:)); 4 | end 5 | output_avg_final=ones(size(labels_validation_total,1),1); 6 | output_final=-ones(size(output_averaged)); 7 | for i=1:(size(labels_validation_total,1)/12) 8 | a=output_averaged((i-1)*12+1:12*i,:); 9 | [~,I]=sort(a); 10 | output_final((i-1)*12+I(11))=1; 11 | output_final((i-1)*12+I(12))=1; 12 | end 13 | 14 | -------------------------------------------------------------------------------- /common/regressor.py: -------------------------------------------------------------------------------- 1 | # -*- coding: utf-8 -*- 2 | 3 | import numpy as np 4 | 5 | 6 | ## Ridge regression 7 | # X: N by P feature matrix, N number of samples, P number of features 8 | # y: N by 1 target vector 9 | # b: P by 1 regression coefficients 10 | # b0: the intercept 11 | 12 | # 岭回归(英文名：ridge regression) 13 | # 是一种专用于共线性数据分析的有偏估计回归方法， 14 | # 实质上是一种改良的最小二乘估计法， 15 | # 通过放弃最小二乘法的无偏性，以损失部分信息、降低精度为代价 16 | # 获得回归系数更为符合实际、更可靠的结果， 17 | # 对病态数据的拟合要强于最小二乘法。 18 | def ridgereg(y, X, coeff = 1e-4): 19 | N, P = X.shape 20 | PHI = np.concatenate((np.ones([N,1]), X), axis=1) 21 | invC = np.linalg.inv(coeff*np.eye(P+1)+ np.matmul(PHI.T, PHI)) 22 | w = np.matmul(np.matmul(invC, PHI.T), y) 23 | b = w[1:] 24 | b0 = w[0] 25 | return b, b0 -------------------------------------------------------------------------------- /common/ButterWorth_filter.py: -------------------------------------------------------------------------------- 1 | # -*- coding: utf-8 -*- 2 | 3 | import numpy as np 4 | from scipy.signal import butter, lfilter, filtfilt, decimate 5 | 6 | 7 | def show_filter(b, a, fs=None): 8 | from scipy.signal import freqz 9 | import matplotlib.pyplot as plt 10 | w, h = freqz(b, a) 11 | if fs == None: 12 | plt.semilogx(w, 20 * np.log10(abs(h))) 13 | else: 14 | plt.plot((fs * 0.5 / np.pi) * w, 20 * np.log10(abs(h))) 15 | plt.title('Butterworth filter frequency response') 16 | plt.xlabel('Frequency (Hz)') 17 | plt.ylabel('Amplitude [dB]') 18 | plt.grid(True) 19 | plt.show() 20 | 21 | 22 | def show_filtering_result(b, a, x): 23 | import matplotlib.pyplot as plt 24 | x1 = lfilter(b, a, x) 25 | x2 = filtfilt(b, a, x) 26 | plt.plot(x, 'k-', label='input') 27 | plt.plot(x1, 'b-', label='lfilter') 28 | plt.plot(x2, 'c-', label='filtfilt') 29 | plt.legend(loc='best') 30 | plt.show() 31 | -------------------------------------------------------------------------------- /common/standard.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | ''' 3 | 1:正类 4 | 0:负类 5 | TP:预测为1,标签为1 6 | TN:预测为0,标签为0 7 | FP:预测为1,标签为0 8 | FN:预测为0,标签为1 9 | ''' 10 | 11 | 12 | def zhi_biao_result(predict, real): 13 | tp = sum([1 for i in range(len(real)) if predict[i] == 1 and real[i] == 1]) 14 | tn = sum([1 for i in range(len(real)) if predict[i] == 0 and real[i] == 0]) 15 | fp = sum([1 for i in range(len(real)) if predict[i] == 1 and real[i] == 0]) 16 | fn = sum([1 for i in range(len(real)) if predict[i] == 0 and real[i] == 1]) 17 | # 准确率 18 | acc = (tp + tn)/(tp + tn + fp + fn) 19 | # 查全率 20 | recall = tp/(tp + fn) 21 | silence = 1 - recall 22 | # 查准率 23 | precision = tp/(tp + fp) 24 | noise = 1 - precision 25 | # 错误率 26 | error = 1 - acc 27 | # F1度量 28 | f1_measure = 2*recall*precision/(precision + recall) 29 | return {"TP": tp, 30 | "TN": tn, 31 | "FP": fp, 32 | "FN": fn, 33 | "Acc": acc, 34 | "Recall": recall, 35 | "Silence": silence, 36 | "Precision": precision, 37 | "Noise": noise, 38 | "Error": error, 39 | "F1-measure": f1_measure} 40 | 41 | 42 | def zhi_biao(predict, real): 43 | if predict.size == real.size: 44 | return zhi_biao_result(predict, real) 45 | else: 46 | return "预测与标签数量不一致" 47 | 48 | 49 | if __name__ == "__main__": 50 | y_predict = np.array((1, 1, 1, 1, 0, 0)) 51 | y_real = np.array((1, 1, 1, 0, 0, 0)) 52 | print(zhi_biao(y_predict, y_real)) 53 | 54 | -------------------------------------------------------------------------------- /common/optimizer.py: -------------------------------------------------------------------------------- 1 | # -*- coding: utf-8 -*- 2 | 3 | import numpy as np 4 | from scipy.optimize import fmin_l_bfgs_b 5 | from abc import ABCMeta, abstractmethod 6 | 7 | class GradientDescentOptimizer(object): 8 | __metaclass__ = ABCMeta 9 | 10 | @abstractmethod 11 | def minimize(self, fojb, x0, args): 12 | raise NotImplementedError() 13 | 14 | class AdamOptimizer(GradientDescentOptimizer): 15 | def __init__(self, maxit=500, stopeps=1e-6): 16 | self.maxit = maxit 17 | self.stopeps = stopeps 18 | 19 | def minimize(self, fobj, x0, args): 20 | alpha = 0.01 21 | beta_1 = 0.9 22 | beta_2 = 0.999 # initialize the values of the parameters 23 | epsilon = 1e-8 24 | 25 | it = 0 26 | m_t = 0 27 | v_t = 0 28 | theta_0 = x0 29 | d_theta = np.Inf 30 | while (d_theta > self.stopeps) and (it < self.maxit): # till it gets converged 31 | it = it + 1 32 | theta_prev = theta_0 33 | f_t, g_t = fobj(theta_0, *args) 34 | m_t = beta_1 * m_t + (1 - beta_1) * g_t # updates the moving averages of the gradient 35 | v_t = beta_2 * v_t + (1 - beta_2) * (g_t * g_t) # updates the moving averages of the squared gradient 36 | m_cap = m_t / (1 - (beta_1 ** it)) # calculates the bias-corrected estimates 37 | v_cap = v_t / (1 - (beta_2 ** it)) # calculates the bias-corrected estimates 38 | theta_0 = theta_0 - (alpha * m_cap) / (np.sqrt(v_cap) + epsilon) # updates the parameters 39 | d_theta = np.linalg.norm(theta_0-theta_prev) 40 | print('Iteration %d: FuncValue = %f, d_theta = %f' % (it, f_t, d_theta)) 41 | 42 | return theta_0 43 | 44 | 45 | class LbfgsOptimizer(GradientDescentOptimizer): 46 | def __init__(self, maxit=500, stopeps=1e-5): 47 | self.maxit = maxit 48 | self.stopeps = stopeps 49 | 50 | def minimize(self, fobj, x0, args): 51 | theta, obj, info = fmin_l_bfgs_b(fobj, x0, args=args, maxiter=self.maxit, epsilon=self.stopeps, disp=1) 52 | return theta -------------------------------------------------------------------------------- /common/CSP(Spatial Filter).py: -------------------------------------------------------------------------------- 1 | # -*- coding: utf-8 -*- 2 | 3 | import numpy as np 4 | import scipy as sp 5 | 6 | 7 | """ Compute Common Spatial Patterns 8 | W: the transform matrix of CSP whose columns are the weight vector 9 | R1: the covariance matrix of one class n-by-n (number of channels) 10 | R1 = X*X'/trace(X*X'); normalized covariance matrix 11 | X is N-by-T matrix where N is the number of channels and T is the 12 | number of samples in time. 13 | R2: the covariance matrix of the other class with the same dimension 14 | """ 15 | def CSP1(R1, R2): 16 | R = R1 + R2 17 | V1, U1 = np.linalg.eig(R) 18 | P = np.dot(np.diag(V1**(-1/2)), U1.T) 19 | S1 = np.dot(np.dot(P, R1), P.T) 20 | # S2 = np.dot(P, np.dot(R2, P.T)) 21 | V2, U2 = np.linalg.eig(S1) 22 | W = np.dot(P.T, U2) 23 | ind = np.argsort(V2) 24 | W = W[:, ind] 25 | return W 26 | 27 | 28 | def CSP2(R1, R2): 29 | V, U = sp.linalg.eig(R1-R2, R1+R2) 30 | ind = np.argsort(V) 31 | W = U[:, ind] 32 | return W 33 | 34 | 35 | """ Train CSP filters 36 | Rs: variance of the multichannel signal 37 | label: class label -1, 1 38 | dim: number of eigenvectors used 39 | """ 40 | def trainCSP2(Rs, labels, dim): 41 | num_trials = Rs.shape[0] 42 | num_channels = Rs.shape[1] 43 | 44 | # sort to make sure that -1 corresponds to left 45 | cc = np.unique(labels) 46 | # cc = np.sort(np.unique(labels)) 47 | count_c1 = 0 48 | count_c2 = 0 49 | cov_c1 = np.zeros([num_channels, num_channels]) 50 | cov_c2 = np.zeros([num_channels, num_channels]) 51 | for i in range(num_trials): 52 | c = labels[i] 53 | if c == cc[0]: 54 | cov_c1 += Rs[i] 55 | count_c1 += 1 56 | elif c == cc[1]: 57 | cov_c2 += Rs[i] 58 | count_c2 += 1 59 | 60 | cov_c1 = cov_c1/count_c1 61 | cov_c2 = cov_c2/count_c2 62 | 63 | W_full = CSP1(cov_c1,cov_c2) 64 | 65 | if dim < 0: 66 | W = W_full 67 | else: 68 | W = np.concatenate((W_full[:, 0:int(dim/2)], W_full[:, num_channels-int(dim/2):]), axis=1) 69 | 70 | return W 71 | -------------------------------------------------------------------------------- /common/utils.py: -------------------------------------------------------------------------------- 1 | # -*- coding: utf-8 -*- 2 | 3 | import numpy as np 4 | 5 | 6 | ## sigmoid activation function 7 | def sigmoid(x): 8 | return 1. / (1. + np.exp(-x)) 9 | 10 | 11 | ## softmax activation function 12 | def softmax(X): 13 | expvx = np.exp(X - np.max(X, axis=1)[..., np.newaxis]) 14 | return expvx / np.sum(expvx, axis=1, keepdims=True) 15 | 16 | 17 | ## Compute numerical gradient for cost function J with respect to theta 18 | # theta: a vector of parameters 19 | # J: a function that outputs a real-number. Calling y = J(theta) will return the 20 | # function value at theta. 21 | def numgrad(J, theta, *args): 22 | epsilon = 1e-4 23 | # Initialize numgrad with zeros 24 | grad = np.zeros(theta.shape) 25 | for i in range(len(theta)): 26 | e = np.zeros(theta.shape) 27 | e[i] = 1 28 | costp, gradp = J(theta + e * epsilon, *args) 29 | costm, gradm = J(theta - e * epsilon, *args) 30 | grad[i] = (costp-costm)/(2*epsilon) 31 | return grad 32 | 33 | 34 | ## Generate a full matrix given the sparse representation 35 | def sparse(ind1, ind2, values): 36 | m = np.max(ind1) + 1 37 | n = np.max(ind2) + 1 38 | A = np.zeros([m,n]) 39 | for i in range(len(values)): 40 | A[ind1[i],ind2[i]] = values[i] 41 | return A 42 | 43 | 44 | ## Generate one-hot coded labels 45 | def onehot(labels, num_classes): 46 | num_labels = labels.shape[0] 47 | index_offset = np.arange(num_labels) * num_classes 48 | labels_onehot = np.zeros((num_labels, num_classes)) 49 | labels_onehot.flat[index_offset + labels.ravel()] = 1 50 | return labels_onehot 51 | 52 | 53 | ## Calculate the Woodbury identity 54 | # (A + BD^{-1}C)^{-1} = A^{-1} - A^{-1}B(D+CA^{-1}B)^{-1}CA^{-1} 55 | # which is useful when A is large and diagonal, and hence easy to invert, 56 | # while B has many rows but few columns (and conversely for C) so that 57 | # the right-hand side is much cheaper to evaluate than the left-hand side. 58 | # We consider this condition only. 59 | def woodburyinv(A,B,C,D): 60 | invA = np.diag(1/np.diag(A)) 61 | # WD = invA - invA*B*(D+C*invA*B)^(-1)*C*invA; 62 | aa = np.matmul(invA,B) 63 | bb = np.linalg.inv(D+np.matmul(C,aa)) 64 | cc = np.matmul(C,invA) 65 | WD = invA - np.matmul(np.matmul(aa,bb),cc) 66 | return WD 67 | 68 | 69 | if __name__ == "__main__": 70 | a = np.random.rand(5) 71 | A = sparse(range(5), range(5), a) 72 | X = np.random.rand(5,10) 73 | D = np.eye(10) 74 | 75 | Z1 = woodburyinv(A, X, X.T, D) 76 | Z2 = np.linalg.inv(A+np.matmul(np.matmul(X,np.linalg.inv(D)),X.T)) 77 | 78 | pass -------------------------------------------------------------------------------- /common/datawrapper.py: -------------------------------------------------------------------------------- 1 | # -*- coding: utf-8 -*- 2 | 3 | # import h5py 4 | import numpy as np 5 | import scipy.io as sio 6 | 7 | 8 | def read_matdata(filepath, keys): 9 | data = {} 10 | f = sio.loadmat(filepath) 11 | # f = h5py.File(filepath, 'r') 12 | for key in keys: 13 | data[key] = f[key] 14 | return data 15 | 16 | 17 | class Dataset(object): 18 | def __init__(self, inputs): 19 | self._data = inputs 20 | self._num_examples = self._data.shape[0] 21 | self._indices = np.arange(self._num_examples) 22 | self._epochs_completed = 0 23 | self._index_in_epoch = 0 24 | 25 | @property 26 | def data(self): 27 | return self._data 28 | 29 | @property 30 | def num_examples(self): 31 | return self._num_examples 32 | 33 | @property 34 | def epochs_completed(self): 35 | return self._epochs_completed 36 | 37 | def shuffle(self): 38 | perm = np.arange(self._num_examples) 39 | np.random.shuffle(perm) 40 | self._indices = perm 41 | 42 | def get_portiondata(self, indices): 43 | return self._data[indices] 44 | 45 | def get_subset(self, ratio, shuffle=True): 46 | ratio = ratio / np.sum(ratio) 47 | num_total = self.num_examples 48 | num_each = (num_total * ratio).astype(int) 49 | ends = np.cumsum(num_each) 50 | ends[-1] = num_total 51 | starts = np.copy(ends) 52 | starts[1:] = starts[0:-1] 53 | starts[0] = 0 54 | if shuffle: self.shuffle() 55 | return [Dataset(self.get_portiondata(self._indices[start:end])) for (start, end) in (starts, ends)] 56 | 57 | def next_batch(self, batch_size, shuffle=True): 58 | '''Return the next `batch_size` examples from this data set.''' 59 | start = self._index_in_epoch 60 | # Shuffle for the first epoch 61 | if self._epochs_completed == 0 and start == 0 and shuffle: 62 | self.shuffle() 63 | # Go to the next epoch 64 | if start + batch_size > self._num_examples: 65 | # Finished epoch 66 | self._epochs_completed += 1 67 | # Get the rest examples in this epoch 68 | rest_num_examples = self._num_examples - start 69 | indices_rest_part = self._indices[start:self._num_examples] 70 | # Shuffle the data 71 | if shuffle: self.shuffle() 72 | # Start next epoch 73 | start = 0 74 | self._index_in_epoch = batch_size - rest_num_examples 75 | end = self._index_in_epoch 76 | indices_new_part = self._indices[start:end] 77 | batch = self.get_portiondata(np.concatenate((indices_rest_part, indices_new_part), axis=0)) 78 | else: 79 | self._index_in_epoch += batch_size 80 | end = self._index_in_epoch 81 | batch = self.get_portiondata(self._indices[start:end]) 82 | return batch -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # EEG-P300Speller_Model-util 2 | This repository contains 3 part. Including PYTHON codes and MATLAB codes. 3 | 4 | # Part One: EEG Signal Classificatrion with Deep Learning Model (Stacked CNN and RNN)-Keras 5 | P300_CNN_1_GRU_A.py: A Python File in which is the model of a combination of CNN and GRU to determine if the EEG signal sequence contains P300 wave. 6 | 7 | The brief of the model-1 (Stacked CNN & RNN): 8 | 1, A CNN is responsbile for spatial domain feature extraction. 9 | 2, A GRU is repsonsible for temporal domain feature extraction. 10 | 3, Dropout is introduced to prevent overfitting and to improve the accuracy. 11 | 12 | ![Stacked RNN and CNN Model](/images/Model.png) 13 | 14 | The brief of model-2 (Stacked CNN)-not contained in repository: 15 | 1, A CNN is responsbile for spatial domain feature extraction. 16 | 2, A CNN is repsonsible for temporal domain feature extraction. 17 | 3, Dropout is introduced to prevent overfitting and to improve the accuracy. 18 | 19 | 20 | ![The Perfromance of Several Models](/images/Model_Identification.png) 21 | 22 | 23 | # Part Two: A Toolkit package for EEG Signals preprocessing. 24 | List and brief description: 25 | 1, EEG_Preprocessor.py: A python file where the code for a series of EEG Signals preprocessing procedure is included. 26 | Including: load_data, extract_eegdata, extract_feature, etc. 27 | 28 | Operations before classification: 29 | 30 | 1) band-pass butter-6 filter to remove the noises irrelevant to classification; 31 | 2) using sliding averaging window to downsample signal data; 32 | 33 | (the file below please see in the common folder) 34 | 2, classifier.py: A python file where the code for several simple classifiers for P300 classification is included. 35 | Including: 36 | 1) Fisher's Linear Discirminant Analysis; 37 | 2) Logistic Regression (Iterative reweighted least square (IRLS) by Newton-Raphson); 38 | 3) Logistic regression for binary classification using SGD algorithm; 39 | 4) Softmax Regression using SGD; 40 | 41 | 3, optimizer.py: A python file where the code for some classic optimizers is included. 42 | Including: 43 | 1) AdamOptimizer; 44 | 2) LbfgsOptimizaer; 45 | 46 | 4, regressor.py: A python file where the code for Ridge Regression is included. 47 | 48 | 5, standard.py: A python file where the code for some criterions for ML model evaluation is included. 49 | 50 | 6, CSP(Spatial_Filter).py: A python file where the code for Common Spatial Pattern, a powerful tool to realize signal data feature extraction, is included. 51 | For more about the CSP, please consult: 52 | https://blog.csdn.net/missxy_/article/details/81264953 53 | 54 | 7, ButterWorth(temporalfilter).py: A python file where the code for Common Spatial Pattern, a powerful tool to remove noises, is included. 55 | 56 | 8, utils.py: includes several machine learning gears. 57 | 58 | # Part Three: A Bagging Ensemble SVM to classify EEG Signals 59 | (the file below please see in the Part Three folder) 60 | It is all written in Matlab. 61 | Consult the main.m file for the code logics. 62 | 63 | 64 | 65 | -------------------------------------------------------------------------------- /.idea/workspace.xml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 30 | 31 | 32 | 33 | 34 | 51 | 52 | 53 | 54 | 55 | 56 | 57 | 58 | 59 | 60 | 61 | 62 | 63 | 1565610754467 64 | 72 | 73 | 74 | 75 | 77 | 78 | 79 | 80 | -------------------------------------------------------------------------------- /p300_CNN_1_GRU_A.py: -------------------------------------------------------------------------------- 1 | # -*- coding:utf-8 -*- 2 | 3 | import numpy as np 4 | from EEG_Preprocessor import * 5 | from common.regressor import * 6 | import random 7 | import tensorflow as tf 8 | 9 | # 6 × 6 字符矩阵 10 | matrix='ABCDEF'+'GHIJKL'+'MNOPQR'+'STUVWX'+'YZ1234'+'56789_' 11 | 12 | # 需要修改路径 13 | datapath = 'E:/bcicompetition/bci2005/II/' 14 | 15 | subject = 'Subject_A' 16 | featureTrain, labelTrain, targetTrain, featureTest, labelTest, targetTest = load_dataset(datapath, subject) 17 | # featureTrain.shape:(85, 12, 15, 1536) 18 | # labelTrain.shape:(85, 12, 15) 19 | # featureTest.shape:(100, 12, 15, 1536) 20 | # labelTest.shape:(100, 12, 15) 21 | 22 | num_train, num_chars, num_repeats, num_features = featureTrain.shape 23 | # num_train: 85 （Subject_B注视的85个字符） 24 | # num_chars： 12 （6行+6列=12种闪烁） 25 | # num_repeats： 15 （12次闪烁，重复15次，对应一个字符的完整输入） 26 | # num_features： 1536 （64个通道与24个时间步值） 27 | 28 | num_test = featureTest.shape[0] 29 | # featureTest.shape：(100, 12, 15, 1536) # 1536=24* 64 拉成一维向量 30 | X_train = np.reshape(featureTrain, [-1, 24, 64]) # 拉成二维 31 | # X_train:(85*12*15=15300, 1536) 32 | y_train = np.reshape(labelTrain, [-1]) 33 | # y_train.shape:(15300,) 34 | X_test = np.reshape(featureTest, [-1, 24, 64]) 35 | # X_test.shape :(100*12*15=18000, 1536) 36 | y_test = np.reshape(labelTest, [-1]) 37 | 38 | 39 | from keras.models import Sequential 40 | from keras.layers import Dense, Dropout, BatchNormalization, GRU 41 | from keras.layers import Conv1D, GlobalAveragePooling1D, MaxPooling1D,Conv2D,Flatten 42 | from keras.layers.core import Reshape 43 | 44 | # 每个信道降采样滤波后的信号长度 45 | seq_length = 24 46 | 47 | model = Sequential() 48 | #转为2D图像的形式 49 | model.add(Reshape((seq_length, 64, 1), input_shape=(seq_length, 64))) 50 | 51 | # 输入形状:3D张量与形状： (batch, steps, channels) 52 | model.add(Conv2D(filters=10, 53 | kernel_size=(1, 64), 54 | padding='valid', 55 | activation='sigmoid')) 56 | 57 | #输入已经变成了 24*1*10 58 | model.add(Reshape((seq_length, 10))) 59 | 60 | #为什么 GRU的时间维度上的输入 是对应着十个卷积核？ 61 | model.add(GRU(24)) 62 | 63 | model.add(Dropout(0.5)) 64 | 65 | # output layer 66 | model.add(Dense(1, activation='sigmoid')) 67 | 68 | 69 | model.compile(loss='binary_crossentropy', 70 | optimizer='rmsprop', 71 | metrics=['accuracy']) 72 | 73 | history = model.fit(X_train, y_train,epochs=10,batch_size=128,validation_data=(X_test, y_test)) 74 | 75 | score = model.evaluate(X_test, y_test, batch_size=128) 76 | print(score) 77 | y_predict = model.predict(X_test) 78 | 79 | targetPredict = np.zeros([num_test, num_repeats], dtype=np.str) 80 | # targetPredict.shape:(100, 15) 81 | 82 | for trial in range(num_test): #range(0, 100) 83 | ytrial = y_predict[trial*num_chars*num_repeats:(trial+1)*num_chars*num_repeats] 84 | # 如果trial=0, ytrial = y_predict[0:12*15=180] 85 | ytrial = np.reshape(ytrial, [num_chars, num_repeats]) 86 | # 改变形状：ytrial.shape :(12, 15) 87 | for repeat in range(num_repeats): # range(0, 15) 88 | yavg = np.mean(ytrial[:,0:repeat+1], axis=1) 89 | # 取repeat次重复后的ytrial在0:repeat+1轴的平均值 90 | row = np.argmax(yavg[6:]) 91 | col = np.argmax(yavg[0:6]) 92 | # 取行中最大的概率和列中最大的概率 93 | targetPredict[trial, repeat] = matrix[int(row*6+col)] 94 | # 预测字符 95 | 96 | accTest = np.zeros(num_repeats) 97 | # accTest.shape:(15,) 98 | for i in range(num_repeats): # range(0, 15) 99 | accTest[i] = np.mean(np.array(targetPredict[:,i] == targetTest).astype(int)) 100 | # 计算1，2，...，15次重复后的准确率的平均值 101 | 102 | # 计算指标 103 | from common.standard import * 104 | y = y_predict.copy() 105 | for i in range(len(y_predict)): 106 | if y[i] >= 0.5: 107 | y[i] = 1 108 | else: 109 | y[i] = 0 110 | print(zhi_biao(y, y_test.T)) 111 | 112 | # 画图 113 | import matplotlib.pyplot as plt 114 | plt.plot(np.arange(num_repeats)+1, accTest*100, 'k-') 115 | plt.title('Character Recognition Rate for ' + subject) 116 | plt.xlabel('Repeat [n]') 117 | plt.ylabel('Accuracy [%]') 118 | plt.grid(which='both', axis='both') 119 | plt.xlim(0, 16) 120 | plt.ylim(0, 100) 121 | plt.show() 122 | 123 | # 绘制训练 & 验证的准确率值 124 | plt.plot(history.history['acc']) 125 | plt.plot(history.history['val_acc']) 126 | plt.title('Model accuracy') 127 | plt.ylabel('Accuracy') 128 | plt.xlabel('Epoch') 129 | plt.legend(['Train', 'Test'], loc='upper left') 130 | plt.show() 131 | 132 | 133 | # 绘制训练 & 验证的损失值 134 | # plt.plot(history.history['loss']) 135 | # plt.plot(history.history['val_loss']) 136 | # plt.title('Model loss') 137 | # plt.ylabel('Loss') 138 | # plt.xlabel('Epoch') 139 | # plt.legend(['Train', 'Test'], loc='upper left') 140 | # plt.show() 141 | -------------------------------------------------------------------------------- /common/classifier.py: -------------------------------------------------------------------------------- 1 | # -*- coding: utf-8 -*- 2 | 3 | import numpy as np 4 | from .utils import * 5 | from .optimizer import * 6 | 7 | ## Fisher's Linear Discriminant Analysis 8 | # y: N by 1 labels 9 | # X: N by P matrix, N observation of P dimensional feature vectors 10 | # wd: weight decay coefficient 11 | def FLDA(y, X, wd = 1e-4): 12 | N, P = X.shape 13 | 14 | index1 = np.argwhere(y==1) 15 | index2 = np.argwhere(y==-1) 16 | N1 = len(index1) 17 | N2 = len(index2) 18 | 19 | X1 = X[index1,:] 20 | X2 = X[index2,:] 21 | 22 | mu1 = np.squeeze(np.mean(X1, axis=0)) 23 | mu2 = np.squeeze(np.mean(X2, axis=0)) 24 | 25 | Sw = np.cov(np.transpose(X)) 26 | 27 | b = np.dot(np.linalg.inv(Sw + wd*np.eye(P)), (mu1 - mu2).T) 28 | b0 = -np.dot(mu1 + mu2, b)/2 29 | 30 | return b, b0 31 | 32 | 33 | ## Logistic regression for binary classification (Page 205-208 of PRML) 34 | # Iterative reweighted least square (IRLS) by Newton-Raphson 35 | # iterative optimization scheme. 36 | # w_new = w_old - (PHI'*R*PHI)^(-1)*PHI'*(y-t); 37 | # 38 | # X: N by P design matrix with N samples of M features 39 | # y: N by 1 target values {0,1} 40 | # wd: weight decay coefficient 41 | # b: P by 1 weight vector 42 | # b0: bias 43 | def logistic(y, X, wd = 1e-4): 44 | # add a constant column to cope with bias 45 | PHI = np.concatenate((np.ones([X.shape[0],1]),X),axis=1) 46 | N, P = PHI.shape 47 | y[y==-1] = 0 # the class label should be 1 or 0 48 | # initialization 49 | w = np.zeros(P) # rough initialization 50 | w[0] = np.log(np.mean(y)/(1-np.mean(y))) 51 | # stop conditions 52 | d_w = np.Inf 53 | maxit = 500 54 | stopeps = 1e-6 55 | 56 | i = 1 57 | while (d_w > stopeps) and (i < maxit) : 58 | wold = w 59 | 60 | t = 1/(1+np.exp(-np.dot(PHI,w))) # predicted target value 61 | R = np.diag(np.squeeze(t*(1-t))) # the variance matrix of target value 62 | # update with a norm2 regularization of w 63 | # H = PHI'*R*PHI + wd*eye(P); 64 | if (P > N): 65 | invH = woodburyinv(wd*np.eye(P), PHI.T, PHI, R) 66 | else: 67 | invH = np.linalg.inv(wd*np.eye(P) + np.dot(np.dot(PHI.T,R),PHI)) 68 | 69 | w = w - np.dot(invH, np.dot(PHI.T,t-y) + wd*w) 70 | d_w = np.linalg.norm(wold-w) 71 | 72 | print('Iteration %d: wchange = %f' % (i, d_w)) 73 | i = i + 1 74 | 75 | if (i >= maxit): 76 | print('Optimization finished with maximum iterations = %d' % maxit) 77 | 78 | return w[1:], w[0] 79 | 80 | 81 | ## Logistic regression for binary classification using SGD algorithm 82 | # X: N by P design matrix with N samples of M features 83 | # y: N by 1 target values {0,1} 84 | # wd: weight decay coefficient 85 | # b: P by 1 weight vector 86 | # b0: bias 87 | # need package pylbfgs installed 88 | def logistic_sgd(y, X, wd=1e-4, optimizer=LbfgsOptimizer()): 89 | ## min sum(log(1 + exp(-t.*(PHI * W)))) + wd *norm(w) 90 | def _logisticCost(w, *args): 91 | wd, PHI, y = args 92 | y = np.squeeze(y) 93 | z = y * np.matmul(PHI, w) 94 | t = 1 / (1 + np.exp(-z)) 95 | grad = np.matmul(-PHI.T, y * (1 - t)) + wd * w 96 | cost = -np.sum(np.log(t)) + 0.5 * wd * np.dot(w.T, w) 97 | return cost, grad 98 | 99 | # add a constant column to cope with bias 100 | PHI = np.concatenate((np.ones([X.shape[0],1]),X),axis=1) 101 | N, P = PHI.shape 102 | # y[y==-1] = 0 # the class label should be 1 or 0 103 | # initialization 104 | w = np.zeros(P) # rough initialization 105 | 106 | # cost, grad = _logisticCost(w, wd, PHI, y) 107 | # grad1 = numgrad(_logisticCost, w, wd, PHI, y) 108 | # diff = np.linalg.norm(grad1 - grad) / np.linalg.norm(grad1 + grad) 109 | # print(diff) 110 | 111 | w = optimizer.minimize(_logisticCost, w, args=(wd, PHI, y)) 112 | return w[1:], w[0] 113 | 114 | 115 | ## softmax activation function 116 | def _softmax(X): 117 | expvx = np.exp(X - np.max(X, axis=1)[..., np.newaxis]) 118 | return expvx / np.sum(expvx, axis=1, keepdims=True) 119 | 120 | ## Softmax regression using stocastic gradient descent algorithm 121 | # X: N by P feature matrix, N number of samples, P number of features 122 | # y: N by 1 class labels (t=k indicate belong to class k) 123 | # wd: weight decay coefficient 124 | # W: P by K regression coefficients 125 | def softmax_train(y, X, wd=1e-4, optimizer=LbfgsOptimizer()): 126 | ## Cross entropy error cost function 127 | def _softmaxCost(theta, *args): 128 | wd, PHI, y = args 129 | N, P = PHI.shape 130 | W = np.reshape(theta, [P,-1]) 131 | t = _softmax(np.matmul(PHI, W)) 132 | grad = (1./N)*np.matmul(PHI.T,t-y) + wd*W 133 | grad = grad.flatten() 134 | cost = -(1./N)*np.dot(y.flatten().T,np.log(t.flatten())) + 0.5*wd*np.sum(W.flatten()**2) 135 | return cost, grad 136 | 137 | K = len(np.unique(y)) 138 | if len(y.shape) == 1 or y.shape[1] == 1: 139 | y = onehot(y, K) 140 | # add a constant column to cope with bias 141 | PHI = np.concatenate((np.ones([X.shape[0], 1]), X), axis=1) 142 | N, P = PHI.shape 143 | W = np.ones([P, K]) # rough initialization 144 | theta = W.flatten() 145 | 146 | # cost, grad = _softmaxCost(theta, wd, PHI, y) 147 | # grad1 = numgrad(_softmaxCost, theta, wd, PHI, y) 148 | # diff = np.linalg.norm(grad1 - grad) / np.linalg.norm(grad1 + grad) 149 | # print(diff) 150 | 151 | opttheta = optimizer.minimize(_softmaxCost, theta, args=(wd, PHI, y)) 152 | W = np.reshape(opttheta, W.shape) 153 | return W 154 | 155 | def softmax_predict(X, W): 156 | PHI = np.concatenate((np.ones([X.shape[0], 1]), X), axis=1) 157 | t = _softmax(np.matmul(PHI, W)) 158 | y = np.argmax(t, axis=1) 159 | return y, t 160 | 161 | 162 | if __name__ == "__main__": 163 | 164 | X = np.random.rand(5, 50) 165 | y = np.array([0, 1, 2, 3, 4]) 166 | W = softmax_train(y, X) -------------------------------------------------------------------------------- /EEG_Preprocessor.py: -------------------------------------------------------------------------------- 1 | # -*- coding: utf-8 -*- 2 | 3 | import numpy as np 4 | from common.datawrapper import * 5 | from common.ButterWorth_filter import * 6 | from scipy.stats import zscore 7 | 8 | 9 | def load_traindata(filepath, filter): 10 | data = read_matdata(filepath, ['Signal', 'Flashing', 'StimulusCode', 'TargetChar']) 11 | signal = data['Signal'] 12 | flashing = data['Flashing'] 13 | stimuluscode = data['StimulusCode'] 14 | targetchar = data['TargetChar'][0] 15 | return extract_eegdata(signal, flashing, stimuluscode, targetchar, filter) 16 | 17 | 18 | def load_testdata(filepath, labelpath, filter): 19 | data = read_matdata(filepath, ['Signal', 'Flashing', 'StimulusCode']) 20 | signal = data['Signal'] 21 | flashing = data['Flashing'] 22 | stimuluscode = data['StimulusCode'] 23 | with open(labelpath, 'r') as myfile: 24 | targetchar = myfile.read().replace('\n', '') 25 | # 去掉换行和空格 26 | return extract_eegdata(signal, flashing, stimuluscode, targetchar, filter) 27 | 28 | 29 | def extract_eegdata(signal, flashing, stimuluscode, targetchar, filter): 30 | num_chars = 12 31 | num_repeats = 15 32 | num_samples = 240 33 | num_trials = signal.shape[0] 34 | # 85个字符 35 | num_channels = signal.shape[2] 36 | # 64个通道 37 | 38 | fb, fa = filter 39 | # show_filtering_result(fb, fa, signal[0,:,0]) 40 | 41 | signal_filtered = np.zeros(signal.shape) 42 | for i in range(num_trials): # 85个字符 43 | for j in range(num_channels): # 64个通道 44 | signal_channel = signal[i,:,j] 45 | signal_filtered[i,:,j] = filtfilt(fb, fa, signal_channel) # 信号滤波 46 | 47 | target = np.array(list(targetchar)) 48 | data = np.zeros([num_trials, num_chars, num_repeats, num_samples, num_channels]) 49 | for i in range(num_trials): 50 | repeat = np.zeros([num_chars,1], dtype=int) 51 | for n in range(1, signal.shape[1]): 52 | if flashing[i, n-1] == 0 and flashing[i, n] == 1: 53 | event = int(stimuluscode[i, n]) 54 | data[i, event-1, repeat[event-1], :, :] = signal_filtered[i, n:n+num_samples, :] 55 | repeat[event - 1] += 1 56 | 57 | return data, target 58 | 59 | 60 | def extract_feature(data, target, sampleseg, chanset, dfs): 61 | 62 | num_trials, num_chars, num_repeats, num_samples, num_channels = data.shape 63 | # data.shape: (85, 12, 15, 240, 64) 64 | sample_begin = sampleseg[0] # 0 65 | sample_end = sampleseg[1] # 144 66 | num_samples_used = int(np.ceil((sample_end - sample_begin) / dfs)) 67 | # np.ceil((144-0)/6 = 24)对于输入24，返回最小的整数24使得 24>=24。 68 | num_channel_used = len(chanset) 69 | # 64个通道 70 | num_features = num_samples_used * num_channel_used 71 | 72 | np.seterr(divide='ignore', invalid='ignore') 73 | labels = np.zeros([num_trials, num_chars, num_repeats]) 74 | feature = np.zeros([num_trials, num_chars, num_repeats, num_features]) 75 | for trial in range(num_trials): 76 | target_index = matrix.find(target[trial]) 77 | target_row = int(np.floor((target_index)/6)) 78 | target_col = target_index - target_row * 6 79 | labels[trial, (target_col, target_row + 6), :] = 1 80 | 81 | signal_trial = data[trial] 82 | for char in range(num_chars): 83 | for repeat in range(num_repeats): 84 | signal_epoch = signal_trial[char, repeat, :, :] 85 | signal_filtered = signal_epoch[sample_begin:sample_end, chanset] 86 | # signal_filtered.shape:[144,64] 87 | signal_downsampled = np.transpose(decimate(signal_filtered.T, dfs, zero_phase=True)) 88 | # signal_downsampled.shape：[144/6=24,64] 89 | # scipy.signal.decimate（要下采样的信号，降采样因子（等价地表示采样率变成原来的几分之一，通常不需要相移所以True） 90 | signal_normalized = np.zeros(signal_downsampled.shape) 91 | for c in range(num_channel_used): 92 | # 如果通道仅包含直流成分，则将值设置为零。 93 | if not (np.max(signal_downsampled[:, c]) == np.min(signal_downsampled[:, c])): 94 | signal_normalized[:, c] = zscore(signal_downsampled[:, c]) 95 | feature[trial, char, repeat, :] = np.reshape(signal_normalized, [-1]) 96 | # np.reshape(signal_normalized, [-1]):把【24,64】的矩阵拉成一维向量 97 | return feature, labels 98 | 99 | 100 | def load_dataset(datapath, subject): 101 | f = np.load(datapath+'processed/'+subject+'.npz') 102 | return f['featureTrain'], f['labelTrain'], f['targetTrain'], f['featureTest'], f['labelTest'], f['targetTest'] 103 | 104 | 105 | if __name__ == '__main__': 106 | # 6 by 6 matrix 107 | matrix = 'ABCDEF' + 'GHIJKL' + 'MNOPQR' + 'STUVWX' + 'YZ1234' + '56789_' 108 | 109 | datapath = 'E:/bcicompetition/bci2005/II/' 110 | 111 | import os 112 | if not os.path.isdir(datapath + 'processed/'): 113 | os.mkdir(datapath + 'processed/') 114 | 115 | fs = 240 116 | # 采样频率 117 | f2 = 20 118 | # 截止频率 119 | order = 6 120 | # 滤波器的阶数 121 | fb, fa = butter(order, 2 * f2 / fs, btype='low') 122 | # 归一化频率的具体计算方法是（2 * 截止频率) / 采样频率 123 | # btype='low'：低通滤波器 124 | # fb, fa : Butterworth滤波器系数,分子（b）和分母（a）IIR滤波器的多项式 125 | # show_filter(fb, fa, fs) 126 | 127 | dfs = 6 128 | sampleseg = [0, int(0.6 * fs)] 129 | # 序列长度【0,0.6秒*240HZ=144】 130 | chanset = np.arange(64) 131 | # chanset:[0,1,2,...,63] 132 | 133 | subject = 'Subject_A' 134 | file_train = datapath + subject + '_Train.mat' 135 | file_test = datapath + subject + '_Test.mat' 136 | file_label = datapath + 'true_labels_' + subject[-1].lower() + '.txt' 137 | 138 | print('Load and extract continuous EEG into epochs for train data') 139 | dataTrain, targetTrain = load_traindata(file_train, [fb, fa]) 140 | print('Extract P300 features from epochs for train data') 141 | featureTrain, labelTrain = extract_feature(dataTrain, targetTrain, sampleseg, chanset, dfs) 142 | 143 | print('Load and extract continuous EEG into epochs for test data') 144 | dataTest, targetTest = load_testdata(file_test, file_label, [fb, fa]) 145 | print('Extract P300 features from epochs for test data') 146 | featureTest, labelTest = extract_feature(dataTest, targetTest, sampleseg, chanset, dfs) 147 | 148 | np.savez(datapath+'processed/'+subject+'.npz', 149 | featureTrain=featureTrain, labelTrain=labelTrain, targetTrain=targetTrain, 150 | featureTest=featureTest, labelTest=labelTest, targetTest=targetTest) 151 | 152 | 153 | -------------------------------------------------------------------------------- /Part Three/topoplotEEG.m: -------------------------------------------------------------------------------- 1 | % topoplot() - plot a topographic map of an EEG field as a 2-D 2 | % circular view (looking down at the top of the head) 3 | % using cointerpolation on a fine cartesian grid. 4 | % Usage: 5 | % >> topoplot(datavector,'eloc_file'); 6 | % >> topoplot(datavector,'eloc_file', 'Param1','Value1', ...) 7 | % Inputs: 8 | % datavector = vector of values at the corresponding locations. 9 | % 'eloc_file' = name of an EEG electrode position file {0 -> 'chan_file'} 10 | % 11 | % Optional Parameters & Values (in any order): 12 | % Param Value 13 | % 'colormap' - any sized colormap 14 | % 'interplimits' - 'electrodes' to furthest electrode 15 | % 'head' to edge of head 16 | % {default 'head'} 17 | % 'gridscale' - scaling grid size {default 67} 18 | % 'maplimits' - 'absmax' +/- the absolute-max 19 | % 'maxmin' scale to data range 20 | % [clim1,clim2] user-definined lo/hi 21 | % {default = 'absmax'} 22 | % 'style' - 'straight' colormap only 23 | % 'contour' contour lines only 24 | % 'both' both colormap and contour lines 25 | % 'fill' constant color between lines 26 | % 'blank' just head and electrodes 27 | % {default = 'both'} 28 | % 'numcontour' - number of contour lines 29 | % {default = 6} 30 | % 'shading' - 'flat','interp' {default = 'flat'} 31 | % 'headcolor' - Color of head cartoon {default black} 32 | % 'electrodes' - 'on','off','labels','numbers' 33 | % 'efontsize','electcolor','emarker','emarkersize' - details 34 | % 35 | % Note: topoplot() only works when map limits are >= the max and min 36 | % interpolated data values. 37 | % Eloc_file format: 38 | % chan_number degrees radius reject_level amp_gain channel_name 39 | % (Angle-0 = Cz-to-Fz; C3-angle =-90; Radius at edge of image = 0.5) 40 | % 41 | % For a sample eloc file: >> topoplot('example') 42 | % 43 | % Note: topoplot() will ignore any electrode with a position outside 44 | % the head (radius > 0.5) 45 | 46 | % Topoplot Version 2.1 47 | 48 | % Begun by Andy Spydell, NHRC, 7-23-96 49 | % 8-96 Revised by Colin Humphries, CNL / Salk Institute, La Jolla CA 50 | % -changed surf command to imagesc (faster) 51 | % -can now handle arbitrary scaling of electrode distances 52 | % -can now handle non integer angles in eloc_file 53 | % 4-4-97 Revised again by Colin Humphries, reformat by SM 54 | % -added parameters 55 | % -changed eloc_file format 56 | % 2-26-98 Revised by Colin 57 | % -changed image back to surface command 58 | % -added fill and blank styles 59 | % -removed extra background colormap entry (now use any colormap) 60 | % -added parameters for electrode colors and labels 61 | % -now each topoplot axes use the caxis command again. 62 | % -removed OUTPUT parameter 63 | % 3-11-98 changed default emarkersize, improve help msg -sm 64 | 65 | function handle = topoplot(Vl,loc_file,p1,v1,p2,v2,p3,v3,p4,v4,p5,v5,p6,v6,p7,v7,p8,v8,p9,v9) 66 | 67 | % User Defined Defaults: 68 | MAXCHANS = 256; 69 | DEFAULT_ELOC = 'eloc64.txt'; 70 | INTERPLIMITS = 'head'; % head, electrodes 71 | MAPLIMITS = 'absmax'; % absmax, maxmin, [values] 72 | GRID_SCALE = 67; 73 | CONTOURNUM = 6; 74 | STYLE = 'both'; % both,straight,fill,contour,blank 75 | HCOLOR = [0 0 0]; 76 | ECOLOR = [0 0 0]; 77 | CONTCOLOR = [0 0 0]; 78 | ELECTROD = 'on'; % ON OFF LABEL 79 | EMARKERSIZE = 6; 80 | EFSIZE = get(0,'DefaultAxesFontSize'); 81 | HLINEWIDTH = 2; 82 | EMARKER = '.'; 83 | SHADING = 'flat'; % flat or interp 84 | 85 | %%%%%%%%%%%%%%%%%%%%%%% 86 | nargs = nargin; 87 | if nargs < 2 88 | loc_file = DEFAULT_ELOC; 89 | end 90 | if nargs == 1 91 | if isstr(Vl) 92 | if any(strcmp(lower(Vl),{'example','demo'})) 93 | fprintf(['This is an example of an electrode location file,\n',... 94 | 'an ascii file consisting of the following four columns:\n',... 95 | ' channel_number degrees arc_length channel_name\n\n',... 96 | 'Example:\n',... 97 | ' 1 -18 .352 Fp1.\n',... 98 | ' 2 18 .352 Fp2.\n',... 99 | ' 5 -90 .181 C3..\n',... 100 | ' 6 90 .181 C4..\n',... 101 | ' 7 -90 .500 A1..\n',... 102 | ' 8 90 .500 A2..\n',... 103 | ' 9 -142 .231 P3..\n',... 104 | '10 142 .231 P4..\n',... 105 | '11 0 .181 Fz..\n',... 106 | '12 0 0 Cz..\n',... 107 | '13 180 .181 Pz..\n\n',... 108 | 'The model head sphere has a diameter of 1.\n',... 109 | 'The vertex (Cz) has arc length 0. Channels with arc \n',... 110 | 'lengths > 0.5 are not plotted nor used for interpolation.\n'... 111 | 'Zero degrees is towards the nasion. Positive angles\n',... 112 | 'point to the right hemisphere; negative to the left.\n',... 113 | 'Channel names should each be four chars, padded with\n',... 114 | 'periods (in place of spaces).\n']) 115 | return 116 | 117 | end 118 | end 119 | end 120 | if isempty(loc_file) 121 | loc_file = 0; 122 | end 123 | if loc_file == 0 124 | loc_file = DEFAULT_ELOC; 125 | end 126 | 127 | if nargs > 2 128 | if ~(round(nargs/2) == nargs/2) 129 | error('topoplot(): Odd number of inputs?') 130 | end 131 | for i = 3:2:nargs 132 | Param = eval(['p',int2str((i-3)/2 +1)]); 133 | Value = eval(['v',int2str((i-3)/2 +1)]); 134 | if ~isstr(Param) 135 | error('topoplot(): Parameter must be a string') 136 | end 137 | Param = lower(Param); 138 | switch lower(Param) 139 | case 'colormap' 140 | if size(Value,2)~=3 141 | error('topoplot(): Colormap must be a n x 3 matrix') 142 | end 143 | colormap(Value) 144 | case {'interplimits','headlimits'} 145 | if ~isstr(Value) 146 | error('topoplot(): interplimits value must be a string') 147 | end 148 | Value = lower(Value); 149 | if ~strcmp(Value,'electrodes') & ~strcmp(Value,'head') 150 | error('topoplot(): Incorrect value for interplimits') 151 | end 152 | INTERPLIMITS = Value; 153 | case 'maplimits' 154 | MAPLIMITS = Value; 155 | case 'gridscale' 156 | GRID_SCALE = Value; 157 | case 'style' 158 | STYLE = lower(Value); 159 | case 'numcontour' 160 | CONTOURNUM = Value; 161 | case 'electrodes' 162 | ELECTROD = lower(Value); 163 | case 'emarker' 164 | EMARKER = Value; 165 | case {'headcolor','hcolor'} 166 | HCOLOR = Value; 167 | case {'electcolor','ecolor'} 168 | ECOLOR = Value; 169 | case {'emarkersize','emsize'} 170 | EMARKERSIZE = Value; 171 | case {'efontsize','efsize'} 172 | EFSIZE = Value; 173 | case 'shading' 174 | SHADING = lower(Value); 175 | if ~any(strcmp(SHADING,{'flat','interp'})) 176 | error('Invalid Shading Parameter') 177 | end 178 | otherwise 179 | error('Unknown parameter.') 180 | end 181 | end 182 | end 183 | 184 | [r,c] = size(Vl); 185 | if r>1 & c>1, 186 | error('topoplot(): data should be a single vector\n'); 187 | end 188 | fid = fopen(loc_file); 189 | if fid<1, 190 | fprintf('topoplot(): cannot open eloc_file (%s).\n',loc_file); 191 | return 192 | end 193 | A = fscanf(fid,'%d %f %f %s',[7 MAXCHANS]); 194 | fclose(fid); 195 | 196 | A = A'; 197 | 198 | if length(Vl) ~= size(A,1), 199 | fprintf(... 200 | 'topoplot(): data vector must have the same rows (%d) as eloc_file (%d)\n',... 201 | length(Vl),size(A,1)); 202 | A 203 | error(''); 204 | end 205 | 206 | labels = setstr(A(:,4:7)); 207 | idx = find(labels == '.'); % some labels have dots 208 | labels(idx) = setstr(abs(' ')*ones(size(idx))); % replace them with spaces 209 | 210 | Th = pi/180*A(:,2); % convert degrees to radians 211 | Rd = A(:,3); 212 | ii = find(Rd <= 0.5); % interpolate on-head channels only 213 | Th = Th(ii); 214 | Rd = Rd(ii); 215 | Vl = Vl(ii); 216 | labels = labels(ii,:); 217 | 218 | [x,y] = pol2cart(Th,Rd); % transform from polar to cartesian coordinates 219 | rmax = 0.5; 220 | 221 | ha = gca; 222 | cla 223 | hold on 224 | 225 | if ~strcmp(STYLE,'blank') 226 | % find limits for interpolation 227 | if strcmp(INTERPLIMITS,'head') 228 | xmin = min(-.5,min(x)); xmax = max(0.5,max(x)); 229 | ymin = min(-.5,min(y)); ymax = max(0.5,max(y)); 230 | else 231 | xmin = max(-.5,min(x)); xmax = min(0.5,max(x)); 232 | ymin = max(-.5,min(y)); ymax = min(0.5,max(y)); 233 | end 234 | 235 | xi = linspace(xmin,xmax,GRID_SCALE); % x-axis description (row vector) 236 | yi = linspace(ymin,ymax,GRID_SCALE); % y-axis description (row vector) 237 | 238 | [Xi,Yi,Zi] = griddata(y,x,Vl,yi',xi,'invdist'); % Interpolate data 239 | 240 | % Take data within head 241 | mask = (sqrt(Xi.^2+Yi.^2) <= rmax); 242 | ii = find(mask == 0); 243 | Zi(ii) = NaN; 244 | 245 | % calculate colormap limits 246 | m = size(colormap,1); 247 | if isstr(MAPLIMITS) 248 | if strcmp(MAPLIMITS,'absmax') 249 | amin = -max(max(abs(Zi))); 250 | amax = max(max(abs(Zi))); 251 | elseif strcmp(MAPLIMITS,'maxmin') 252 | amin = min(min(Zi)); 253 | amax = max(max(Zi)); 254 | end 255 | else 256 | amin = MAPLIMITS(1); 257 | amax = MAPLIMITS(2); 258 | end 259 | delta = xi(2)-xi(1); % length of grid entry 260 | 261 | % Draw topoplot on head 262 | if strcmp(STYLE,'contour') 263 | contour(Xi,Yi,Zi,CONTOURNUM,'k'); 264 | elseif strcmp(STYLE,'both') 265 | surface(Xi-delta/2,Yi-delta/2,zeros(size(Zi)),Zi,'EdgeColor','none',... 266 | 'FaceColor',SHADING); 267 | contour(Xi,Yi,Zi,CONTOURNUM,'k'); 268 | elseif strcmp(STYLE,'straight') 269 | surface(Xi-delta/2,Yi-delta/2,zeros(size(Zi)),Zi,'EdgeColor','none',... 270 | 'FaceColor',SHADING); 271 | elseif strcmp(STYLE,'fill') 272 | contourf(Xi,Yi,Zi,CONTOURNUM,'k'); 273 | else 274 | error('Invalid style') 275 | end 276 | caxis([amin amax]) % set coloraxis 277 | end 278 | 279 | set(ha,'Xlim',[-rmax*1.3 rmax*1.3],'Ylim',[-rmax*1.3 rmax*1.3]) 280 | 281 | % %%% Draw Head %%%% 282 | l = 0:2*pi/100:2*pi; 283 | basex = .18*rmax; 284 | tip = rmax*1.15; base = rmax-.004; 285 | EarX = [.497 .510 .518 .5299 .5419 .54 .547 .532 .510 .489]; 286 | EarY = [.0555 .0775 .0783 .0746 .0555 -.0055 -.0932 -.1313 -.1384 -.1199]; 287 | 288 | % Plot Electrodes 289 | if strcmp(ELECTROD,'on') 290 | hp2 = plot(y,x,EMARKER,'Color',ECOLOR,'markersize',EMARKERSIZE); 291 | elseif strcmp(ELECTROD,'labels') 292 | for i = 1:size(labels,1) 293 | text(y(i),x(i),labels(i,:),'HorizontalAlignment','center',... 294 | 'VerticalAlignment','middle','Color',ECOLOR,... 295 | 'FontSize',EFSIZE) 296 | end 297 | elseif strcmp(ELECTROD,'numbers') 298 | whos y x 299 | for i = 1:size(labels,1) 300 | text(y(i),x(i),int2str(i),'HorizontalAlignment','center',... 301 | 'VerticalAlignment','middle','Color',ECOLOR,... 302 | 'FontSize',EFSIZE) 303 | end 304 | end 305 | 306 | % Plot Head, Ears, Nose 307 | plot(cos(l).*rmax,sin(l).*rmax,... 308 | 'color',HCOLOR,'Linestyle','-','LineWidth',HLINEWIDTH); 309 | 310 | plot([.18*rmax;0;-.18*rmax],[base;tip;base],... 311 | 'Color',HCOLOR,'LineWidth',HLINEWIDTH); 312 | 313 | plot(EarX,EarY,'color',HCOLOR,'LineWidth',HLINEWIDTH) 314 | plot(-EarX,EarY,'color',HCOLOR,'LineWidth',HLINEWIDTH) 315 | 316 | hold off 317 | axis off 318 | 319 | -------------------------------------------------------------------------------- /Part Three/trainClassifier.m: -------------------------------------------------------------------------------- 1 | function [trainedClassifier, validationAccuracy] = trainClassifier(trainingData) 2 | % trainClassifier(trainingData) 3 | % returns a trained classifier and its accuracy. 4 | % This code recreates the classification model trained in 5 | % Classification Learner app. 6 | % 7 | % Input: 8 | % trainingData: the training data of same data type as imported 9 | % in the app (table or matrix). 10 | % 11 | % Output: 12 | % trainedClassifier: a struct containing the trained classifier. 13 | % The struct contains various fields with information about the 14 | % trained classifier. 15 | % 16 | % trainedClassifier.predictFcn: a function to make predictions 17 | % on new data. It takes an input of the same form as this training 18 | % code (table or matrix) and returns predictions for the response. 19 | % If you supply a matrix, include only the predictors columns (or 20 | % rows). 21 | % 22 | % validationAccuracy: a double containing the accuracy in 23 | % percent. In the app, the History list displays this 24 | % overall accuracy score for each model. 25 | % 26 | % Use the code to train the model with new data. 27 | % To retrain your classifier, call the function from the command line 28 | % with your original data or new data as the input argument trainingData. 29 | % 30 | % For example, to retrain a classifier trained with the original data set 31 | % T, enter: 32 | % [trainedClassifier, validationAccuracy] = trainClassifier(T) 33 | % 34 | % To make predictions with the returned 'trainedClassifier' on new data T, 35 | % use 36 | % yfit = trainedClassifier.predictFcn(T) 37 | % 38 | % To automate training the same classifier with new data, or to learn how 39 | % to programmatically train classifiers, examine the generated code. 40 | 41 | % Auto-generated by MATLAB on 2019-01-01 19:20:30 42 | 43 | 44 | % Extract predictors and response 45 | % This code processes the data into the right shape for training the 46 | % classifier. 47 | % Convert input to table 48 | inputTable = array2table(trainingData, 'VariableNames', {'column_1', 'column_2', 'column_3', 'column_4', 'column_5', 'column_6', 'column_7', 'column_8', 'column_9', 'column_10', 'column_11', 'column_12', 'column_13', 'column_14', 'column_15', 'column_16', 'column_17', 'column_18', 'column_19', 'column_20', 'column_21', 'column_22', 'column_23', 'column_24', 'column_25', 'column_26', 'column_27', 'column_28', 'column_29', 'column_30', 'column_31', 'column_32', 'column_33', 'column_34', 'column_35', 'column_36', 'column_37', 'column_38', 'column_39', 'column_40', 'column_41', 'column_42', 'column_43', 'column_44', 'column_45', 'column_46', 'column_47', 'column_48', 'column_49', 'column_50', 'column_51', 'column_52', 'column_53', 'column_54', 'column_55', 'column_56', 'column_57', 'column_58', 'column_59', 'column_60', 'column_61', 'column_62', 'column_63', 'column_64', 'column_65', 'column_66', 'column_67', 'column_68', 'column_69', 'column_70', 'column_71', 'column_72', 'column_73', 'column_74', 'column_75', 'column_76', 'column_77', 'column_78', 'column_79', 'column_80', 'column_81', 'column_82', 'column_83', 'column_84', 'column_85', 'column_86', 'column_87', 'column_88', 'column_89', 'column_90', 'column_91', 'column_92', 'column_93', 'column_94', 'column_95', 'column_96', 'column_97', 'column_98', 'column_99', 'column_100', 'column_101', 'column_102', 'column_103', 'column_104', 'column_105', 'column_106', 'column_107', 'column_108', 'column_109', 'column_110', 'column_111', 'column_112', 'column_113', 'column_114', 'column_115', 'column_116', 'column_117', 'column_118', 'column_119', 'column_120', 'column_121', 'column_122', 'column_123', 'column_124', 'column_125', 'column_126', 'column_127', 'column_128', 'column_129', 'column_130', 'column_131', 'column_132', 'column_133', 'column_134', 'column_135', 'column_136', 'column_137', 'column_138', 'column_139', 'column_140', 'column_141', 'column_142', 'column_143', 'column_144', 'column_145', 'column_146', 'column_147', 'column_148', 'column_149', 'column_150', 'column_151', 'column_152', 'column_153', 'column_154', 'column_155', 'column_156', 'column_157', 'column_158', 'column_159', 'column_160', 'column_161', 'column_162', 'column_163', 'column_164', 'column_165', 'column_166', 'column_167', 'column_168', 'column_169', 'column_170', 'column_171', 'column_172', 'column_173', 'column_174', 'column_175', 'column_176', 'column_177', 'column_178', 'column_179', 'column_180', 'column_181', 'column_182', 'column_183', 'column_184', 'column_185', 'column_186', 'column_187', 'column_188', 'column_189', 'column_190', 'column_191', 'column_192', 'column_193', 'column_194', 'column_195', 'column_196', 'column_197', 'column_198', 'column_199', 'column_200', 'column_201', 'column_202', 'column_203', 'column_204', 'column_205', 'column_206', 'column_207', 'column_208', 'column_209', 'column_210', 'column_211', 'column_212', 'column_213', 'column_214', 'column_215', 'column_216', 'column_217', 'column_218', 'column_219', 'column_220', 'column_221', 'column_222', 'column_223', 'column_224', 'column_225', 'column_226', 'column_227', 'column_228', 'column_229', 'column_230', 'column_231', 'column_232', 'column_233', 'column_234', 'column_235', 'column_236', 'column_237', 'column_238', 'column_239', 'column_240', 'column_241', 'column_242', 'column_243', 'column_244', 'column_245', 'column_246', 'column_247', 'column_248', 'column_249', 'column_250', 'column_251', 'column_252', 'column_253', 'column_254', 'column_255', 'column_256', 'column_257', 'column_258', 'column_259', 'column_260', 'column_261', 'column_262', 'column_263', 'column_264', 'column_265', 'column_266', 'column_267', 'column_268', 'column_269', 'column_270', 'column_271', 'column_272', 'column_273', 'column_274', 'column_275', 'column_276', 'column_277', 'column_278', 'column_279', 'column_280', 'column_281', 'column_282', 'column_283', 'column_284', 'column_285', 'column_286', 'column_287', 'column_288', 'column_289', 'column_290', 'column_291', 'column_292', 'column_293', 'column_294', 'column_295', 'column_296', 'column_297', 'column_298', 'column_299', 'column_300', 'column_301', 'column_302', 'column_303', 'column_304', 'column_305', 'column_306', 'column_307', 'column_308', 'column_309', 'column_310', 'column_311', 'column_312', 'column_313', 'column_314', 'column_315', 'column_316', 'column_317', 'column_318', 'column_319', 'column_320', 'column_321', 'column_322', 'column_323', 'column_324', 'column_325', 'column_326', 'column_327', 'column_328', 'column_329', 'column_330', 'column_331', 'column_332', 'column_333', 'column_334', 'column_335', 'column_336', 'column_337', 'column_338', 'column_339', 'column_340', 'column_341', 'column_342', 'column_343', 'column_344', 'column_345', 'column_346', 'column_347', 'column_348', 'column_349', 'column_350', 'column_351', 'column_352', 'column_353', 'column_354', 'column_355', 'column_356', 'column_357', 'column_358', 'column_359', 'column_360', 'column_361', 'column_362', 'column_363', 'column_364', 'column_365', 'column_366', 'column_367', 'column_368', 'column_369', 'column_370', 'column_371', 'column_372', 'column_373', 'column_374', 'column_375', 'column_376', 'column_377', 'column_378', 'column_379', 'column_380', 'column_381', 'column_382', 'column_383', 'column_384', 'column_385', 'column_386', 'column_387', 'column_388', 'column_389', 'column_390', 'column_391', 'column_392', 'column_393', 'column_394', 'column_395', 'column_396', 'column_397', 'column_398', 'column_399', 'column_400', 'column_401', 'column_402', 'column_403', 'column_404', 'column_405', 'column_406', 'column_407', 'column_408', 'column_409', 'column_410', 'column_411', 'column_412', 'column_413', 'column_414', 'column_415', 'column_416', 'column_417', 'column_418', 'column_419', 'column_420', 'column_421', 'column_422', 'column_423', 'column_424', 'column_425', 'column_426', 'column_427', 'column_428', 'column_429', 'column_430', 'column_431', 'column_432', 'column_433', 'column_434', 'column_435', 'column_436', 'column_437', 'column_438', 'column_439', 'column_440', 'column_441', 'column_442', 'column_443', 'column_444', 'column_445', 'column_446', 'column_447', 'column_448', 'column_449', 'column_450', 'column_451', 'column_452', 'column_453', 'column_454', 'column_455', 'column_456', 'column_457', 'column_458', 'column_459', 'column_460', 'column_461', 'column_462', 'column_463', 'column_464', 'column_465', 'column_466', 'column_467', 'column_468', 'column_469', 'column_470', 'column_471', 'column_472', 'column_473', 'column_474', 'column_475', 'column_476', 'column_477', 'column_478', 'column_479', 'column_480', 'column_481', 'column_482', 'column_483', 'column_484', 'column_485', 'column_486', 'column_487', 'column_488', 'column_489', 'column_490', 'column_491', 'column_492', 'column_493', 'column_494', 'column_495', 'column_496', 'column_497', 'column_498', 'column_499', 'column_500', 'column_501', 'column_502', 'column_503', 'column_504', 'column_505', 'column_506', 'column_507', 'column_508', 'column_509', 'column_510', 'column_511', 'column_512', 'column_513', 'column_514', 'column_515', 'column_516', 'column_517', 'column_518', 'column_519', 'column_520', 'column_521', 'column_522', 'column_523', 'column_524', 'column_525', 'column_526', 'column_527', 'column_528', 'column_529', 'column_530', 'column_531', 'column_532', 'column_533', 'column_534', 'column_535', 'column_536', 'column_537', 'column_538', 'column_539', 'column_540', 'column_541', 'column_542', 'column_543', 'column_544', 'column_545', 'column_546', 'column_547', 'column_548', 'column_549', 'column_550', 'column_551', 'column_552', 'column_553', 'column_554', 'column_555', 'column_556', 'column_557', 'column_558', 'column_559', 'column_560', 'column_561', 'column_562', 'column_563', 'column_564', 'column_565', 'column_566', 'column_567', 'column_568', 'column_569', 'column_570', 'column_571', 'column_572', 'column_573', 'column_574', 'column_575', 'column_576', 'column_577', 'column_578', 'column_579', 'column_580', 'column_581', 'column_582', 'column_583', 'column_584', 'column_585', 'column_586', 'column_587', 'column_588', 'column_589', 'column_590', 'column_591', 'column_592', 'column_593', 'column_594', 'column_595', 'column_596', 'column_597', 'column_598', 'column_599', 'column_600', 'column_601', 'column_602', 'column_603', 'column_604', 'column_605', 'column_606', 'column_607', 'column_608', 'column_609', 'column_610', 'column_611', 'column_612', 'column_613', 'column_614', 'column_615', 'column_616', 'column_617', 'column_618', 'column_619', 'column_620', 'column_621', 'column_622', 'column_623', 'column_624', 'column_625', 'column_626', 'column_627', 'column_628', 'column_629', 'column_630', 'column_631', 'column_632', 'column_633', 'column_634', 'column_635', 'column_636', 'column_637', 'column_638', 'column_639', 'column_640', 'column_641', 'column_642', 'column_643', 'column_644', 'column_645', 'column_646', 'column_647', 'column_648', 'column_649', 'column_650', 'column_651', 'column_652', 'column_653', 'column_654', 'column_655', 'column_656', 'column_657', 'column_658', 'column_659', 'column_660', 'column_661', 'column_662', 'column_663', 'column_664', 'column_665', 'column_666', 'column_667', 'column_668', 'column_669', 'column_670', 'column_671', 'column_672', 'column_673', 'column_674', 'column_675', 'column_676', 'column_677', 'column_678', 'column_679', 'column_680', 'column_681', 'column_682', 'column_683', 'column_684', 'column_685', 'column_686', 'column_687', 'column_688', 'column_689', 'column_690', 'column_691', 'column_692', 'column_693', 'column_694', 'column_695', 'column_696', 'column_697', 'column_698', 'column_699', 'column_700', 'column_701', 'column_702', 'column_703', 'column_704', 'column_705', 'column_706', 'column_707', 'column_708', 'column_709', 'column_710', 'column_711', 'column_712', 'column_713', 'column_714', 'column_715', 'column_716', 'column_717', 'column_718', 'column_719', 'column_720', 'column_721', 'column_722', 'column_723', 'column_724', 'column_725', 'column_726', 'column_727', 'column_728', 'column_729', 'column_730', 'column_731', 'column_732', 'column_733', 'column_734', 'column_735', 'column_736', 'column_737', 'column_738', 'column_739', 'column_740', 'column_741', 'column_742', 'column_743', 'column_744', 'column_745', 'column_746', 'column_747', 'column_748', 'column_749', 'column_750', 'column_751', 'column_752', 'column_753', 'column_754', 'column_755', 'column_756', 'column_757', 'column_758', 'column_759', 'column_760', 'column_761', 'column_762', 'column_763', 'column_764', 'column_765', 'column_766', 'column_767', 'column_768', 'column_769', 'column_770', 'column_771', 'column_772', 'column_773', 'column_774', 'column_775', 'column_776', 'column_777', 'column_778', 'column_779', 'column_780', 'column_781', 'column_782', 'column_783', 'column_784', 'column_785', 'column_786', 'column_787', 'column_788', 'column_789', 'column_790', 'column_791', 'column_792', 'column_793', 'column_794', 'column_795', 'column_796', 'column_797', 'column_798', 'column_799'}); 49 | 50 | predictorNames = {'column_1', 'column_2', 'column_3', 'column_4', 'column_5', 'column_6', 'column_7', 'column_8', 'column_9', 'column_10', 'column_11', 'column_12', 'column_13', 'column_14', 'column_15', 'column_16', 'column_17', 'column_18', 'column_19', 'column_20', 'column_21', 'column_22', 'column_23', 'column_24', 'column_25', 'column_26', 'column_27', 'column_28', 'column_29', 'column_30', 'column_31', 'column_32', 'column_33', 'column_34', 'column_35', 'column_36', 'column_37', 'column_38', 'column_39', 'column_40', 'column_41', 'column_42', 'column_43', 'column_44', 'column_45', 'column_46', 'column_47', 'column_48', 'column_49', 'column_50', 'column_51', 'column_52', 'column_53', 'column_54', 'column_55', 'column_56', 'column_57', 'column_58', 'column_59', 'column_60', 'column_61', 'column_62', 'column_63', 'column_64', 'column_65', 'column_66', 'column_67', 'column_68', 'column_69', 'column_70', 'column_71', 'column_72', 'column_73', 'column_74', 'column_75', 'column_76', 'column_77', 'column_78', 'column_79', 'column_80', 'column_81', 'column_82', 'column_83', 'column_84', 'column_85', 'column_86', 'column_87', 'column_88', 'column_89', 'column_90', 'column_91', 'column_92', 'column_93', 'column_94', 'column_95', 'column_96', 'column_97', 'column_98', 'column_99', 'column_100', 'column_101', 'column_102', 'column_103', 'column_104', 'column_105', 'column_106', 'column_107', 'column_108', 'column_109', 'column_110', 'column_111', 'column_112', 'column_113', 'column_114', 'column_115', 'column_116', 'column_117', 'column_118', 'column_119', 'column_120', 'column_121', 'column_122', 'column_123', 'column_124', 'column_125', 'column_126', 'column_127', 'column_128', 'column_129', 'column_130', 'column_131', 'column_132', 'column_133', 'column_134', 'column_135', 'column_136', 'column_137', 'column_138', 'column_139', 'column_140', 'column_141', 'column_142', 'column_143', 'column_144', 'column_145', 'column_146', 'column_147', 'column_148', 'column_149', 'column_150', 'column_151', 'column_152', 'column_153', 'column_154', 'column_155', 'column_156', 'column_157', 'column_158', 'column_159', 'column_160', 'column_161', 'column_162', 'column_163', 'column_164', 'column_165', 'column_166', 'column_167', 'column_168', 'column_169', 'column_170', 'column_171', 'column_172', 'column_173', 'column_174', 'column_175', 'column_176', 'column_177', 'column_178', 'column_179', 'column_180', 'column_181', 'column_182', 'column_183', 'column_184', 'column_185', 'column_186', 'column_187', 'column_188', 'column_189', 'column_190', 'column_191', 'column_192', 'column_193', 'column_194', 'column_195', 'column_196', 'column_197', 'column_198', 'column_199', 'column_200', 'column_201', 'column_202', 'column_203', 'column_204', 'column_205', 'column_206', 'column_207', 'column_208', 'column_209', 'column_210', 'column_211', 'column_212', 'column_213', 'column_214', 'column_215', 'column_216', 'column_217', 'column_218', 'column_219', 'column_220', 'column_221', 'column_222', 'column_223', 'column_224', 'column_225', 'column_226', 'column_227', 'column_228', 'column_229', 'column_230', 'column_231', 'column_232', 'column_233', 'column_234', 'column_235', 'column_236', 'column_237', 'column_238', 'column_239', 'column_240', 'column_241', 'column_242', 'column_243', 'column_244', 'column_245', 'column_246', 'column_247', 'column_248', 'column_249', 'column_250', 'column_251', 'column_252', 'column_253', 'column_254', 'column_255', 'column_256', 'column_257', 'column_258', 'column_259', 'column_260', 'column_261', 'column_262', 'column_263', 'column_264', 'column_265', 'column_266', 'column_267', 'column_268', 'column_269', 'column_270', 'column_271', 'column_272', 'column_273', 'column_274', 'column_275', 'column_276', 'column_277', 'column_278', 'column_279', 'column_280', 'column_281', 'column_282', 'column_283', 'column_284', 'column_285', 'column_286', 'column_287', 'column_288', 'column_289', 'column_290', 'column_291', 'column_292', 'column_293', 'column_294', 'column_295', 'column_296', 'column_297', 'column_298', 'column_299', 'column_300', 'column_301', 'column_302', 'column_303', 'column_304', 'column_305', 'column_306', 'column_307', 'column_308', 'column_309', 'column_310', 'column_311', 'column_312', 'column_313', 'column_314', 'column_315', 'column_316', 'column_317', 'column_318', 'column_319', 'column_320', 'column_321', 'column_322', 'column_323', 'column_324', 'column_325', 'column_326', 'column_327', 'column_328', 'column_329', 'column_330', 'column_331', 'column_332', 'column_333', 'column_334', 'column_335', 'column_336', 'column_337', 'column_338', 'column_339', 'column_340', 'column_341', 'column_342', 'column_343', 'column_344', 'column_345', 'column_346', 'column_347', 'column_348', 'column_349', 'column_350', 'column_351', 'column_352', 'column_353', 'column_354', 'column_355', 'column_356', 'column_357', 'column_358', 'column_359', 'column_360', 'column_361', 'column_362', 'column_363', 'column_364', 'column_365', 'column_366', 'column_367', 'column_368', 'column_369', 'column_370', 'column_371', 'column_372', 'column_373', 'column_374', 'column_375', 'column_376', 'column_377', 'column_378', 'column_379', 'column_380', 'column_381', 'column_382', 'column_383', 'column_384', 'column_385', 'column_386', 'column_387', 'column_388', 'column_389', 'column_390', 'column_391', 'column_392', 'column_393', 'column_394', 'column_395', 'column_396', 'column_397', 'column_398', 'column_399', 'column_400', 'column_401', 'column_402', 'column_403', 'column_404', 'column_405', 'column_406', 'column_407', 'column_408', 'column_409', 'column_410', 'column_411', 'column_412', 'column_413', 'column_414', 'column_415', 'column_416', 'column_417', 'column_418', 'column_419', 'column_420', 'column_421', 'column_422', 'column_423', 'column_424', 'column_425', 'column_426', 'column_427', 'column_428', 'column_429', 'column_430', 'column_431', 'column_432', 'column_433', 'column_434', 'column_435', 'column_436', 'column_437', 'column_438', 'column_439', 'column_440', 'column_441', 'column_442', 'column_443', 'column_444', 'column_445', 'column_446', 'column_447', 'column_448', 'column_449', 'column_450', 'column_451', 'column_452', 'column_453', 'column_454', 'column_455', 'column_456', 'column_457', 'column_458', 'column_459', 'column_460', 'column_461', 'column_462', 'column_463', 'column_464', 'column_465', 'column_466', 'column_467', 'column_468', 'column_469', 'column_470', 'column_471', 'column_472', 'column_473', 'column_474', 'column_475', 'column_476', 'column_477', 'column_478', 'column_479', 'column_480', 'column_481', 'column_482', 'column_483', 'column_484', 'column_485', 'column_486', 'column_487', 'column_488', 'column_489', 'column_490', 'column_491', 'column_492', 'column_493', 'column_494', 'column_495', 'column_496', 'column_497', 'column_498', 'column_499', 'column_500', 'column_501', 'column_502', 'column_503', 'column_504', 'column_505', 'column_506', 'column_507', 'column_508', 'column_509', 'column_510', 'column_511', 'column_512', 'column_513', 'column_514', 'column_515', 'column_516', 'column_517', 'column_518', 'column_519', 'column_520', 'column_521', 'column_522', 'column_523', 'column_524', 'column_525', 'column_526', 'column_527', 'column_528', 'column_529', 'column_530', 'column_531', 'column_532', 'column_533', 'column_534', 'column_535', 'column_536', 'column_537', 'column_538', 'column_539', 'column_540', 'column_541', 'column_542', 'column_543', 'column_544', 'column_545', 'column_546', 'column_547', 'column_548', 'column_549', 'column_550', 'column_551', 'column_552', 'column_553', 'column_554', 'column_555', 'column_556', 'column_557', 'column_558', 'column_559', 'column_560', 'column_561', 'column_562', 'column_563', 'column_564', 'column_565', 'column_566', 'column_567', 'column_568', 'column_569', 'column_570', 'column_571', 'column_572', 'column_573', 'column_574', 'column_575', 'column_576', 'column_577', 'column_578', 'column_579', 'column_580', 'column_581', 'column_582', 'column_583', 'column_584', 'column_585', 'column_586', 'column_587', 'column_588', 'column_589', 'column_590', 'column_591', 'column_592', 'column_593', 'column_594', 'column_595', 'column_596', 'column_597', 'column_598', 'column_599', 'column_600', 'column_601', 'column_602', 'column_603', 'column_604', 'column_605', 'column_606', 'column_607', 'column_608', 'column_609', 'column_610', 'column_611', 'column_612', 'column_613', 'column_614', 'column_615', 'column_616', 'column_617', 'column_618', 'column_619', 'column_620', 'column_621', 'column_622', 'column_623', 'column_624', 'column_625', 'column_626', 'column_627', 'column_628', 'column_629', 'column_630', 'column_631', 'column_632', 'column_633', 'column_634', 'column_635', 'column_636', 'column_637', 'column_638', 'column_639', 'column_640', 'column_641', 'column_642', 'column_643', 'column_644', 'column_645', 'column_646', 'column_647', 'column_648', 'column_649', 'column_650', 'column_651', 'column_652', 'column_653', 'column_654', 'column_655', 'column_656', 'column_657', 'column_658', 'column_659', 'column_660', 'column_661', 'column_662', 'column_663', 'column_664', 'column_665', 'column_666', 'column_667', 'column_668', 'column_669', 'column_670', 'column_671', 'column_672', 'column_673', 'column_674', 'column_675', 'column_676', 'column_677', 'column_678', 'column_679', 'column_680', 'column_681', 'column_682', 'column_683', 'column_684', 'column_685', 'column_686', 'column_687', 'column_688', 'column_689', 'column_690', 'column_691', 'column_692', 'column_693', 'column_694', 'column_695', 'column_696', 'column_697', 'column_698', 'column_699', 'column_700', 'column_701', 'column_702', 'column_703', 'column_704', 'column_705', 'column_706', 'column_707', 'column_708', 'column_709', 'column_710', 'column_711', 'column_712', 'column_713', 'column_714', 'column_715', 'column_716', 'column_717', 'column_718', 'column_719', 'column_720', 'column_721', 'column_722', 'column_723', 'column_724', 'column_725', 'column_726', 'column_727', 'column_728', 'column_729', 'column_730', 'column_731', 'column_732', 'column_733', 'column_734', 'column_735', 'column_736', 'column_737', 'column_738', 'column_739', 'column_740', 'column_741', 'column_742', 'column_743', 'column_744', 'column_745', 'column_746', 'column_747', 'column_748', 'column_749', 'column_750', 'column_751', 'column_752', 'column_753', 'column_754', 'column_755', 'column_756', 'column_757', 'column_758', 'column_759', 'column_760', 'column_761', 'column_762', 'column_763', 'column_764', 'column_765', 'column_766', 'column_767', 'column_768', 'column_769', 'column_770', 'column_771', 'column_772', 'column_773', 'column_774', 'column_775', 'column_776', 'column_777', 'column_778', 'column_779', 'column_780', 'column_781', 'column_782', 'column_783', 'column_784', 'column_785', 'column_786', 'column_787', 'column_788', 'column_789', 'column_790', 'column_791', 'column_792', 'column_793', 'column_794', 'column_795', 'column_796', 'column_797', 'column_798'}; 51 | predictors = inputTable(:, predictorNames); 52 | response = inputTable.column_799; 53 | isCategoricalPredictor = [false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false]; 54 | 55 | % Train a classifier 56 | % This code specifies all the classifier options and trains the classifier. 57 | classificationSVM = fitcsvm(... 58 | predictors, ... 59 | response, ... 60 | 'KernelFunction', 'gaussian', ... 61 | 'PolynomialOrder', [], ... 62 | 'KernelScale', 'auto', ... 63 | 'BoxConstraint', 1, ... 64 | 'Standardize', true, ... 65 | 'ClassNames', [-1; 1]); 66 | 67 | % Create the result struct with predict function 68 | predictorExtractionFcn = @(x) array2table(x, 'VariableNames', predictorNames); 69 | svmPredictFcn = @(x) predict(classificationSVM, x); 70 | trainedClassifier.predictFcn = @(x) svmPredictFcn(predictorExtractionFcn(x)); 71 | 72 | % Add additional fields to the result struct 73 | trainedClassifier.ClassificationSVM = classificationSVM; 74 | trainedClassifier.About = 'This struct is a trained classifier exported from Classification Learner R2016b.'; 75 | trainedClassifier.HowToPredict = sprintf('To make predictions on a new predictor column matrix, X, use: \n yfit = c.predictFcn(X) \nreplacing ''c'' with the name of the variable that is this struct, e.g. ''trainedClassifier''. \n \nX must contain exactly 798 columns because this classifier was trained using 798 predictors. \nX must contain only predictor columns in exactly the same order and format as your training \ndata. Do not include the response column or any columns you did not import into \nClassification Learner. \n \nFor more information, see How to predict using an exported model.'); 76 | 77 | % Extract predictors and response 78 | % This code processes the data into the right shape for training the 79 | % classifier. 80 | % Convert input to table 81 | inputTable = array2table(trainingData, 'VariableNames', {'column_1', 'column_2', 'column_3', 'column_4', 'column_5', 'column_6', 'column_7', 'column_8', 'column_9', 'column_10', 'column_11', 'column_12', 'column_13', 'column_14', 'column_15', 'column_16', 'column_17', 'column_18', 'column_19', 'column_20', 'column_21', 'column_22', 'column_23', 'column_24', 'column_25', 'column_26', 'column_27', 'column_28', 'column_29', 'column_30', 'column_31', 'column_32', 'column_33', 'column_34', 'column_35', 'column_36', 'column_37', 'column_38', 'column_39', 'column_40', 'column_41', 'column_42', 'column_43', 'column_44', 'column_45', 'column_46', 'column_47', 'column_48', 'column_49', 'column_50', 'column_51', 'column_52', 'column_53', 'column_54', 'column_55', 'column_56', 'column_57', 'column_58', 'column_59', 'column_60', 'column_61', 'column_62', 'column_63', 'column_64', 'column_65', 'column_66', 'column_67', 'column_68', 'column_69', 'column_70', 'column_71', 'column_72', 'column_73', 'column_74', 'column_75', 'column_76', 'column_77', 'column_78', 'column_79', 'column_80', 'column_81', 'column_82', 'column_83', 'column_84', 'column_85', 'column_86', 'column_87', 'column_88', 'column_89', 'column_90', 'column_91', 'column_92', 'column_93', 'column_94', 'column_95', 'column_96', 'column_97', 'column_98', 'column_99', 'column_100', 'column_101', 'column_102', 'column_103', 'column_104', 'column_105', 'column_106', 'column_107', 'column_108', 'column_109', 'column_110', 'column_111', 'column_112', 'column_113', 'column_114', 'column_115', 'column_116', 'column_117', 'column_118', 'column_119', 'column_120', 'column_121', 'column_122', 'column_123', 'column_124', 'column_125', 'column_126', 'column_127', 'column_128', 'column_129', 'column_130', 'column_131', 'column_132', 'column_133', 'column_134', 'column_135', 'column_136', 'column_137', 'column_138', 'column_139', 'column_140', 'column_141', 'column_142', 'column_143', 'column_144', 'column_145', 'column_146', 'column_147', 'column_148', 'column_149', 'column_150', 'column_151', 'column_152', 'column_153', 'column_154', 'column_155', 'column_156', 'column_157', 'column_158', 'column_159', 'column_160', 'column_161', 'column_162', 'column_163', 'column_164', 'column_165', 'column_166', 'column_167', 'column_168', 'column_169', 'column_170', 'column_171', 'column_172', 'column_173', 'column_174', 'column_175', 'column_176', 'column_177', 'column_178', 'column_179', 'column_180', 'column_181', 'column_182', 'column_183', 'column_184', 'column_185', 'column_186', 'column_187', 'column_188', 'column_189', 'column_190', 'column_191', 'column_192', 'column_193', 'column_194', 'column_195', 'column_196', 'column_197', 'column_198', 'column_199', 'column_200', 'column_201', 'column_202', 'column_203', 'column_204', 'column_205', 'column_206', 'column_207', 'column_208', 'column_209', 'column_210', 'column_211', 'column_212', 'column_213', 'column_214', 'column_215', 'column_216', 'column_217', 'column_218', 'column_219', 'column_220', 'column_221', 'column_222', 'column_223', 'column_224', 'column_225', 'column_226', 'column_227', 'column_228', 'column_229', 'column_230', 'column_231', 'column_232', 'column_233', 'column_234', 'column_235', 'column_236', 'column_237', 'column_238', 'column_239', 'column_240', 'column_241', 'column_242', 'column_243', 'column_244', 'column_245', 'column_246', 'column_247', 'column_248', 'column_249', 'column_250', 'column_251', 'column_252', 'column_253', 'column_254', 'column_255', 'column_256', 'column_257', 'column_258', 'column_259', 'column_260', 'column_261', 'column_262', 'column_263', 'column_264', 'column_265', 'column_266', 'column_267', 'column_268', 'column_269', 'column_270', 'column_271', 'column_272', 'column_273', 'column_274', 'column_275', 'column_276', 'column_277', 'column_278', 'column_279', 'column_280', 'column_281', 'column_282', 'column_283', 'column_284', 'column_285', 'column_286', 'column_287', 'column_288', 'column_289', 'column_290', 'column_291', 'column_292', 'column_293', 'column_294', 'column_295', 'column_296', 'column_297', 'column_298', 'column_299', 'column_300', 'column_301', 'column_302', 'column_303', 'column_304', 'column_305', 'column_306', 'column_307', 'column_308', 'column_309', 'column_310', 'column_311', 'column_312', 'column_313', 'column_314', 'column_315', 'column_316', 'column_317', 'column_318', 'column_319', 'column_320', 'column_321', 'column_322', 'column_323', 'column_324', 'column_325', 'column_326', 'column_327', 'column_328', 'column_329', 'column_330', 'column_331', 'column_332', 'column_333', 'column_334', 'column_335', 'column_336', 'column_337', 'column_338', 'column_339', 'column_340', 'column_341', 'column_342', 'column_343', 'column_344', 'column_345', 'column_346', 'column_347', 'column_348', 'column_349', 'column_350', 'column_351', 'column_352', 'column_353', 'column_354', 'column_355', 'column_356', 'column_357', 'column_358', 'column_359', 'column_360', 'column_361', 'column_362', 'column_363', 'column_364', 'column_365', 'column_366', 'column_367', 'column_368', 'column_369', 'column_370', 'column_371', 'column_372', 'column_373', 'column_374', 'column_375', 'column_376', 'column_377', 'column_378', 'column_379', 'column_380', 'column_381', 'column_382', 'column_383', 'column_384', 'column_385', 'column_386', 'column_387', 'column_388', 'column_389', 'column_390', 'column_391', 'column_392', 'column_393', 'column_394', 'column_395', 'column_396', 'column_397', 'column_398', 'column_399', 'column_400', 'column_401', 'column_402', 'column_403', 'column_404', 'column_405', 'column_406', 'column_407', 'column_408', 'column_409', 'column_410', 'column_411', 'column_412', 'column_413', 'column_414', 'column_415', 'column_416', 'column_417', 'column_418', 'column_419', 'column_420', 'column_421', 'column_422', 'column_423', 'column_424', 'column_425', 'column_426', 'column_427', 'column_428', 'column_429', 'column_430', 'column_431', 'column_432', 'column_433', 'column_434', 'column_435', 'column_436', 'column_437', 'column_438', 'column_439', 'column_440', 'column_441', 'column_442', 'column_443', 'column_444', 'column_445', 'column_446', 'column_447', 'column_448', 'column_449', 'column_450', 'column_451', 'column_452', 'column_453', 'column_454', 'column_455', 'column_456', 'column_457', 'column_458', 'column_459', 'column_460', 'column_461', 'column_462', 'column_463', 'column_464', 'column_465', 'column_466', 'column_467', 'column_468', 'column_469', 'column_470', 'column_471', 'column_472', 'column_473', 'column_474', 'column_475', 'column_476', 'column_477', 'column_478', 'column_479', 'column_480', 'column_481', 'column_482', 'column_483', 'column_484', 'column_485', 'column_486', 'column_487', 'column_488', 'column_489', 'column_490', 'column_491', 'column_492', 'column_493', 'column_494', 'column_495', 'column_496', 'column_497', 'column_498', 'column_499', 'column_500', 'column_501', 'column_502', 'column_503', 'column_504', 'column_505', 'column_506', 'column_507', 'column_508', 'column_509', 'column_510', 'column_511', 'column_512', 'column_513', 'column_514', 'column_515', 'column_516', 'column_517', 'column_518', 'column_519', 'column_520', 'column_521', 'column_522', 'column_523', 'column_524', 'column_525', 'column_526', 'column_527', 'column_528', 'column_529', 'column_530', 'column_531', 'column_532', 'column_533', 'column_534', 'column_535', 'column_536', 'column_537', 'column_538', 'column_539', 'column_540', 'column_541', 'column_542', 'column_543', 'column_544', 'column_545', 'column_546', 'column_547', 'column_548', 'column_549', 'column_550', 'column_551', 'column_552', 'column_553', 'column_554', 'column_555', 'column_556', 'column_557', 'column_558', 'column_559', 'column_560', 'column_561', 'column_562', 'column_563', 'column_564', 'column_565', 'column_566', 'column_567', 'column_568', 'column_569', 'column_570', 'column_571', 'column_572', 'column_573', 'column_574', 'column_575', 'column_576', 'column_577', 'column_578', 'column_579', 'column_580', 'column_581', 'column_582', 'column_583', 'column_584', 'column_585', 'column_586', 'column_587', 'column_588', 'column_589', 'column_590', 'column_591', 'column_592', 'column_593', 'column_594', 'column_595', 'column_596', 'column_597', 'column_598', 'column_599', 'column_600', 'column_601', 'column_602', 'column_603', 'column_604', 'column_605', 'column_606', 'column_607', 'column_608', 'column_609', 'column_610', 'column_611', 'column_612', 'column_613', 'column_614', 'column_615', 'column_616', 'column_617', 'column_618', 'column_619', 'column_620', 'column_621', 'column_622', 'column_623', 'column_624', 'column_625', 'column_626', 'column_627', 'column_628', 'column_629', 'column_630', 'column_631', 'column_632', 'column_633', 'column_634', 'column_635', 'column_636', 'column_637', 'column_638', 'column_639', 'column_640', 'column_641', 'column_642', 'column_643', 'column_644', 'column_645', 'column_646', 'column_647', 'column_648', 'column_649', 'column_650', 'column_651', 'column_652', 'column_653', 'column_654', 'column_655', 'column_656', 'column_657', 'column_658', 'column_659', 'column_660', 'column_661', 'column_662', 'column_663', 'column_664', 'column_665', 'column_666', 'column_667', 'column_668', 'column_669', 'column_670', 'column_671', 'column_672', 'column_673', 'column_674', 'column_675', 'column_676', 'column_677', 'column_678', 'column_679', 'column_680', 'column_681', 'column_682', 'column_683', 'column_684', 'column_685', 'column_686', 'column_687', 'column_688', 'column_689', 'column_690', 'column_691', 'column_692', 'column_693', 'column_694', 'column_695', 'column_696', 'column_697', 'column_698', 'column_699', 'column_700', 'column_701', 'column_702', 'column_703', 'column_704', 'column_705', 'column_706', 'column_707', 'column_708', 'column_709', 'column_710', 'column_711', 'column_712', 'column_713', 'column_714', 'column_715', 'column_716', 'column_717', 'column_718', 'column_719', 'column_720', 'column_721', 'column_722', 'column_723', 'column_724', 'column_725', 'column_726', 'column_727', 'column_728', 'column_729', 'column_730', 'column_731', 'column_732', 'column_733', 'column_734', 'column_735', 'column_736', 'column_737', 'column_738', 'column_739', 'column_740', 'column_741', 'column_742', 'column_743', 'column_744', 'column_745', 'column_746', 'column_747', 'column_748', 'column_749', 'column_750', 'column_751', 'column_752', 'column_753', 'column_754', 'column_755', 'column_756', 'column_757', 'column_758', 'column_759', 'column_760', 'column_761', 'column_762', 'column_763', 'column_764', 'column_765', 'column_766', 'column_767', 'column_768', 'column_769', 'column_770', 'column_771', 'column_772', 'column_773', 'column_774', 'column_775', 'column_776', 'column_777', 'column_778', 'column_779', 'column_780', 'column_781', 'column_782', 'column_783', 'column_784', 'column_785', 'column_786', 'column_787', 'column_788', 'column_789', 'column_790', 'column_791', 'column_792', 'column_793', 'column_794', 'column_795', 'column_796', 'column_797', 'column_798', 'column_799'}); 82 | 83 | predictorNames = {'column_1', 'column_2', 'column_3', 'column_4', 'column_5', 'column_6', 'column_7', 'column_8', 'column_9', 'column_10', 'column_11', 'column_12', 'column_13', 'column_14', 'column_15', 'column_16', 'column_17', 'column_18', 'column_19', 'column_20', 'column_21', 'column_22', 'column_23', 'column_24', 'column_25', 'column_26', 'column_27', 'column_28', 'column_29', 'column_30', 'column_31', 'column_32', 'column_33', 'column_34', 'column_35', 'column_36', 'column_37', 'column_38', 'column_39', 'column_40', 'column_41', 'column_42', 'column_43', 'column_44', 'column_45', 'column_46', 'column_47', 'column_48', 'column_49', 'column_50', 'column_51', 'column_52', 'column_53', 'column_54', 'column_55', 'column_56', 'column_57', 'column_58', 'column_59', 'column_60', 'column_61', 'column_62', 'column_63', 'column_64', 'column_65', 'column_66', 'column_67', 'column_68', 'column_69', 'column_70', 'column_71', 'column_72', 'column_73', 'column_74', 'column_75', 'column_76', 'column_77', 'column_78', 'column_79', 'column_80', 'column_81', 'column_82', 'column_83', 'column_84', 'column_85', 'column_86', 'column_87', 'column_88', 'column_89', 'column_90', 'column_91', 'column_92', 'column_93', 'column_94', 'column_95', 'column_96', 'column_97', 'column_98', 'column_99', 'column_100', 'column_101', 'column_102', 'column_103', 'column_104', 'column_105', 'column_106', 'column_107', 'column_108', 'column_109', 'column_110', 'column_111', 'column_112', 'column_113', 'column_114', 'column_115', 'column_116', 'column_117', 'column_118', 'column_119', 'column_120', 'column_121', 'column_122', 'column_123', 'column_124', 'column_125', 'column_126', 'column_127', 'column_128', 'column_129', 'column_130', 'column_131', 'column_132', 'column_133', 'column_134', 'column_135', 'column_136', 'column_137', 'column_138', 'column_139', 'column_140', 'column_141', 'column_142', 'column_143', 'column_144', 'column_145', 'column_146', 'column_147', 'column_148', 'column_149', 'column_150', 'column_151', 'column_152', 'column_153', 'column_154', 'column_155', 'column_156', 'column_157', 'column_158', 'column_159', 'column_160', 'column_161', 'column_162', 'column_163', 'column_164', 'column_165', 'column_166', 'column_167', 'column_168', 'column_169', 'column_170', 'column_171', 'column_172', 'column_173', 'column_174', 'column_175', 'column_176', 'column_177', 'column_178', 'column_179', 'column_180', 'column_181', 'column_182', 'column_183', 'column_184', 'column_185', 'column_186', 'column_187', 'column_188', 'column_189', 'column_190', 'column_191', 'column_192', 'column_193', 'column_194', 'column_195', 'column_196', 'column_197', 'column_198', 'column_199', 'column_200', 'column_201', 'column_202', 'column_203', 'column_204', 'column_205', 'column_206', 'column_207', 'column_208', 'column_209', 'column_210', 'column_211', 'column_212', 'column_213', 'column_214', 'column_215', 'column_216', 'column_217', 'column_218', 'column_219', 'column_220', 'column_221', 'column_222', 'column_223', 'column_224', 'column_225', 'column_226', 'column_227', 'column_228', 'column_229', 'column_230', 'column_231', 'column_232', 'column_233', 'column_234', 'column_235', 'column_236', 'column_237', 'column_238', 'column_239', 'column_240', 'column_241', 'column_242', 'column_243', 'column_244', 'column_245', 'column_246', 'column_247', 'column_248', 'column_249', 'column_250', 'column_251', 'column_252', 'column_253', 'column_254', 'column_255', 'column_256', 'column_257', 'column_258', 'column_259', 'column_260', 'column_261', 'column_262', 'column_263', 'column_264', 'column_265', 'column_266', 'column_267', 'column_268', 'column_269', 'column_270', 'column_271', 'column_272', 'column_273', 'column_274', 'column_275', 'column_276', 'column_277', 'column_278', 'column_279', 'column_280', 'column_281', 'column_282', 'column_283', 'column_284', 'column_285', 'column_286', 'column_287', 'column_288', 'column_289', 'column_290', 'column_291', 'column_292', 'column_293', 'column_294', 'column_295', 'column_296', 'column_297', 'column_298', 'column_299', 'column_300', 'column_301', 'column_302', 'column_303', 'column_304', 'column_305', 'column_306', 'column_307', 'column_308', 'column_309', 'column_310', 'column_311', 'column_312', 'column_313', 'column_314', 'column_315', 'column_316', 'column_317', 'column_318', 'column_319', 'column_320', 'column_321', 'column_322', 'column_323', 'column_324', 'column_325', 'column_326', 'column_327', 'column_328', 'column_329', 'column_330', 'column_331', 'column_332', 'column_333', 'column_334', 'column_335', 'column_336', 'column_337', 'column_338', 'column_339', 'column_340', 'column_341', 'column_342', 'column_343', 'column_344', 'column_345', 'column_346', 'column_347', 'column_348', 'column_349', 'column_350', 'column_351', 'column_352', 'column_353', 'column_354', 'column_355', 'column_356', 'column_357', 'column_358', 'column_359', 'column_360', 'column_361', 'column_362', 'column_363', 'column_364', 'column_365', 'column_366', 'column_367', 'column_368', 'column_369', 'column_370', 'column_371', 'column_372', 'column_373', 'column_374', 'column_375', 'column_376', 'column_377', 'column_378', 'column_379', 'column_380', 'column_381', 'column_382', 'column_383', 'column_384', 'column_385', 'column_386', 'column_387', 'column_388', 'column_389', 'column_390', 'column_391', 'column_392', 'column_393', 'column_394', 'column_395', 'column_396', 'column_397', 'column_398', 'column_399', 'column_400', 'column_401', 'column_402', 'column_403', 'column_404', 'column_405', 'column_406', 'column_407', 'column_408', 'column_409', 'column_410', 'column_411', 'column_412', 'column_413', 'column_414', 'column_415', 'column_416', 'column_417', 'column_418', 'column_419', 'column_420', 'column_421', 'column_422', 'column_423', 'column_424', 'column_425', 'column_426', 'column_427', 'column_428', 'column_429', 'column_430', 'column_431', 'column_432', 'column_433', 'column_434', 'column_435', 'column_436', 'column_437', 'column_438', 'column_439', 'column_440', 'column_441', 'column_442', 'column_443', 'column_444', 'column_445', 'column_446', 'column_447', 'column_448', 'column_449', 'column_450', 'column_451', 'column_452', 'column_453', 'column_454', 'column_455', 'column_456', 'column_457', 'column_458', 'column_459', 'column_460', 'column_461', 'column_462', 'column_463', 'column_464', 'column_465', 'column_466', 'column_467', 'column_468', 'column_469', 'column_470', 'column_471', 'column_472', 'column_473', 'column_474', 'column_475', 'column_476', 'column_477', 'column_478', 'column_479', 'column_480', 'column_481', 'column_482', 'column_483', 'column_484', 'column_485', 'column_486', 'column_487', 'column_488', 'column_489', 'column_490', 'column_491', 'column_492', 'column_493', 'column_494', 'column_495', 'column_496', 'column_497', 'column_498', 'column_499', 'column_500', 'column_501', 'column_502', 'column_503', 'column_504', 'column_505', 'column_506', 'column_507', 'column_508', 'column_509', 'column_510', 'column_511', 'column_512', 'column_513', 'column_514', 'column_515', 'column_516', 'column_517', 'column_518', 'column_519', 'column_520', 'column_521', 'column_522', 'column_523', 'column_524', 'column_525', 'column_526', 'column_527', 'column_528', 'column_529', 'column_530', 'column_531', 'column_532', 'column_533', 'column_534', 'column_535', 'column_536', 'column_537', 'column_538', 'column_539', 'column_540', 'column_541', 'column_542', 'column_543', 'column_544', 'column_545', 'column_546', 'column_547', 'column_548', 'column_549', 'column_550', 'column_551', 'column_552', 'column_553', 'column_554', 'column_555', 'column_556', 'column_557', 'column_558', 'column_559', 'column_560', 'column_561', 'column_562', 'column_563', 'column_564', 'column_565', 'column_566', 'column_567', 'column_568', 'column_569', 'column_570', 'column_571', 'column_572', 'column_573', 'column_574', 'column_575', 'column_576', 'column_577', 'column_578', 'column_579', 'column_580', 'column_581', 'column_582', 'column_583', 'column_584', 'column_585', 'column_586', 'column_587', 'column_588', 'column_589', 'column_590', 'column_591', 'column_592', 'column_593', 'column_594', 'column_595', 'column_596', 'column_597', 'column_598', 'column_599', 'column_600', 'column_601', 'column_602', 'column_603', 'column_604', 'column_605', 'column_606', 'column_607', 'column_608', 'column_609', 'column_610', 'column_611', 'column_612', 'column_613', 'column_614', 'column_615', 'column_616', 'column_617', 'column_618', 'column_619', 'column_620', 'column_621', 'column_622', 'column_623', 'column_624', 'column_625', 'column_626', 'column_627', 'column_628', 'column_629', 'column_630', 'column_631', 'column_632', 'column_633', 'column_634', 'column_635', 'column_636', 'column_637', 'column_638', 'column_639', 'column_640', 'column_641', 'column_642', 'column_643', 'column_644', 'column_645', 'column_646', 'column_647', 'column_648', 'column_649', 'column_650', 'column_651', 'column_652', 'column_653', 'column_654', 'column_655', 'column_656', 'column_657', 'column_658', 'column_659', 'column_660', 'column_661', 'column_662', 'column_663', 'column_664', 'column_665', 'column_666', 'column_667', 'column_668', 'column_669', 'column_670', 'column_671', 'column_672', 'column_673', 'column_674', 'column_675', 'column_676', 'column_677', 'column_678', 'column_679', 'column_680', 'column_681', 'column_682', 'column_683', 'column_684', 'column_685', 'column_686', 'column_687', 'column_688', 'column_689', 'column_690', 'column_691', 'column_692', 'column_693', 'column_694', 'column_695', 'column_696', 'column_697', 'column_698', 'column_699', 'column_700', 'column_701', 'column_702', 'column_703', 'column_704', 'column_705', 'column_706', 'column_707', 'column_708', 'column_709', 'column_710', 'column_711', 'column_712', 'column_713', 'column_714', 'column_715', 'column_716', 'column_717', 'column_718', 'column_719', 'column_720', 'column_721', 'column_722', 'column_723', 'column_724', 'column_725', 'column_726', 'column_727', 'column_728', 'column_729', 'column_730', 'column_731', 'column_732', 'column_733', 'column_734', 'column_735', 'column_736', 'column_737', 'column_738', 'column_739', 'column_740', 'column_741', 'column_742', 'column_743', 'column_744', 'column_745', 'column_746', 'column_747', 'column_748', 'column_749', 'column_750', 'column_751', 'column_752', 'column_753', 'column_754', 'column_755', 'column_756', 'column_757', 'column_758', 'column_759', 'column_760', 'column_761', 'column_762', 'column_763', 'column_764', 'column_765', 'column_766', 'column_767', 'column_768', 'column_769', 'column_770', 'column_771', 'column_772', 'column_773', 'column_774', 'column_775', 'column_776', 'column_777', 'column_778', 'column_779', 'column_780', 'column_781', 'column_782', 'column_783', 'column_784', 'column_785', 'column_786', 'column_787', 'column_788', 'column_789', 'column_790', 'column_791', 'column_792', 'column_793', 'column_794', 'column_795', 'column_796', 'column_797', 'column_798'}; 84 | predictors = inputTable(:, predictorNames); 85 | response = inputTable.column_799; 86 | isCategoricalPredictor = [false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false]; 87 | 88 | % Perform cross-validation 89 | partitionedModel = crossval(trainedClassifier.ClassificationSVM, 'KFold', 5); 90 | 91 | % Compute validation accuracy 92 | validationAccuracy = 1 - kfoldLoss(partitionedModel, 'LossFun', 'ClassifError'); 93 | 94 | % Compute validation predictions and scores 95 | [validationPredictions, validationScores] = kfoldPredict(partitionedModel); 96 | --------------------------------------------------------------------------------