Python metrics.matthews_corrcoef函数代码示例

def plot_conf_matrix(y_true, y_pred, normed=True, heatmap_color ='Blues', **kwargs):

    ## check to make sure that y_pred is an array of integers if y_true is a bunch of integers
    true_int_check = all(isinstance(a,int) for a in y_true)
    pred_int_check = all(isinstance(a,int) for a in y_pred)
    if true_int_check and not pred_int_check: # convert the y_pred values to integers
        if isinstance(y_pred, pd.Series):
            y_pred = y_pred.astype(int)

    my_c = metrics.confusion_matrix(y_true, y_pred)

    print metrics.matthews_corrcoef(y_true, y_pred)
    if normed:
        cm_normalized = my_c.astype('float') / my_c.sum(axis=1)[:, np.newaxis]
        my_c = cm_normalized
        plt.title('Normalized RF Classifier Confusion Matrix')
    else:
        plt.title('Random Forest Classifier Confusion Matrix')

    sns.heatmap(my_c, annot=True,  fmt='',cmap=heatmap_color, **kwargs)
    plt.ylabel('True')
    plt.xlabel('Assigned')
    plt.show()

    return

def melodiness_metrics(m_train, m_test, y_train, y_test):
    """ Compute metrics on melodiness score

    Parameters
    ----------
    m_train : np.array [n_samples]
        melodiness scores for training set
    m_test : np.array [n_samples]
        melodiness scores for testing set
    y_train : np.array [n_samples]
        Training labels.
    y_test : np.array [n_samples]
        Testing labels.

    Returns
    -------
    melodiness_scores : dict
        melodiness scores for training set
    """
    m_bin_train = 1*(m_train >= 1)
    m_bin_test = 1*(m_test >= 1)

    train_scores = {}
    test_scores = {}

    train_scores['accuracy'] = metrics.accuracy_score(y_train, m_bin_train)
    test_scores['accuracy'] = metrics.accuracy_score(y_test, m_bin_test)

    train_scores['mcc'] = metrics.matthews_corrcoef(y_train, m_bin_train)
    test_scores['mcc'] = metrics.matthews_corrcoef(y_test, m_bin_test)

    (p, r, f, s) = metrics.precision_recall_fscore_support(y_train,
                                                           m_bin_train)
    train_scores['precision'] = p
    train_scores['recall'] = r
    train_scores['f1'] = f
    train_scores['support'] = s

    (p, r, f, s) = metrics.precision_recall_fscore_support(y_test,
                                                           m_bin_test)
    test_scores['precision'] = p
    test_scores['recall'] = r
    test_scores['f1'] = f
    test_scores['support'] = s

    train_scores['confusion matrix'] = \
        metrics.confusion_matrix(y_train, m_bin_train, labels=[0, 1])
    test_scores['confusion matrix'] = \
        metrics.confusion_matrix(y_test, m_bin_test, labels=[0, 1])

    train_scores['auc score'] = \
        metrics.roc_auc_score(y_train, m_train + 1, average='weighted')
    test_scores['auc score'] = \
        metrics.roc_auc_score(y_test, m_test + 1, average='weighted')

    melodiness_scores = {'train': train_scores, 'test': test_scores}

    return melodiness_scores

def clf_metrics(p_train, p_test, y_train, y_test):
    """ Compute metrics on classifier predictions

    Parameters
    ----------
    p_train : np.array [n_samples]
        predicted probabilities for training set
    p_test : np.array [n_samples]
        predicted probabilities for testing set
    y_train : np.array [n_samples]
        Training labels.
    y_test : np.array [n_samples]
        Testing labels.

    Returns
    -------
    clf_scores : dict
        classifier scores for training set
    """
    y_pred_train = 1*(p_train >= 0.5)
    y_pred_test = 1*(p_test >= 0.5)

    train_scores = {}
    test_scores = {}

    train_scores['accuracy'] = metrics.accuracy_score(y_train, y_pred_train)
    test_scores['accuracy'] = metrics.accuracy_score(y_test, y_pred_test)

    train_scores['mcc'] = metrics.matthews_corrcoef(y_train, y_pred_train)
    test_scores['mcc'] = metrics.matthews_corrcoef(y_test, y_pred_test)

    (p, r, f, s) = metrics.precision_recall_fscore_support(y_train,
                                                           y_pred_train)
    train_scores['precision'] = p
    train_scores['recall'] = r
    train_scores['f1'] = f
    train_scores['support'] = s

    (p, r, f, s) = metrics.precision_recall_fscore_support(y_test,
                                                           y_pred_test)
    test_scores['precision'] = p
    test_scores['recall'] = r
    test_scores['f1'] = f
    test_scores['support'] = s

    train_scores['confusion matrix'] = \
        metrics.confusion_matrix(y_train, y_pred_train, labels=[0, 1])
    test_scores['confusion matrix'] = \
        metrics.confusion_matrix(y_test, y_pred_test, labels=[0, 1])

    train_scores['auc score'] = \
        metrics.roc_auc_score(y_train, p_train + 1, average='weighted')
    test_scores['auc score'] = \
        metrics.roc_auc_score(y_test, p_test + 1, average='weighted')

    clf_scores = {'train': train_scores, 'test': test_scores}

    return clf_scores

	def printAnalysis(self,true_pred,y_pred1):

		print "########## Analysing the Model result ##########################"

		math_corr = matthews_corrcoef( true_pred,y_pred1)
		roc_auc = roc_auc_score( true_pred,y_pred1)

		print(classification_report( true_pred,y_pred1))
		print("Matthews correlation :" + str(matthews_corrcoef( true_pred,y_pred1)))
		print("ROC AUC score :" + str(roc_auc_score( true_pred,y_pred1)))

def _show_classification_results(y_test, y_pred):
    """ Prints performance metrics for a classifier """

    print metrics.classification_report(y_test, y_pred)
    print
    print 'Confusion matrix:'
    print metrics.confusion_matrix(y_test, y_pred)
    print
    print 'Matthew\'s correlation coefficient:',
    print metrics.matthews_corrcoef(y_test, y_pred)
    print 'F1 score:',
    print metrics.f1_score(y_test, y_pred)
    print

def score_MCC(ground_truth, scores):
    '''
    assuming the model output is the probability of being default,
    then this probability can be used for ranking. Then using the fraction of
    default in validation data to assign the proper threshold to the prediction
    '''

    if isinstance(scores, pd.Series):
        scores = scores.values

    if isinstance(ground_truth, pd.Series):
        ground_truth = ground_truth.values

    tmp_ground_truth = np.copy(ground_truth)
    fault_frac = tmp_ground_truth.mean()
    #print 'score shape:', scores.shape, 
    print 'mean of groud truth:', fault_frac
    thres_value = np.percentile(scores, 100.*(1-fault_frac), axis=0)
    print 'threshold for preds:', thres_value
    binary_scores = scores > thres_value
    binary_scores = binary_scores.astype(int)
    ## convert to sk-learn format
    np.place(binary_scores, binary_scores==0, -1)
    np.place(tmp_ground_truth, tmp_ground_truth==0, -1)

    return matthews_corrcoef(tmp_ground_truth, binary_scores)

def calcualte_threshold(positives, negatives, measure="SPC", measure_threshold=0.95, thresholds=None, attempt=0):
    """Plot the TPR the FPR vs threshold values
 
    Input:
    postives - list of scores of postive runs
    negatives - list of scores of negative runs
    measure - choose coffectiong by 95% Specificity ("SPC"), or matthews_corrcoef ("MCC")
    """
    assert measure in ["TPR", "FPR", "SPC", "MCC", "PPV", "FDR", "ACC"]
    y_true = [1]*len(positives)+[0]*len(negatives)
    values = {name:[] for name in ["TPR", "FPR", "SPC", "MCC", "PPV", "FDR", "ACC"]}
    saveThreshold = None
    saveValue = 1.0
    thresholds = list(sorted(thresholds or map(lambda i: i/10., xrange(1,10000))))

    for threshold in thresholds:
        TN = sum([1 for score in negatives if score < threshold])
        FP = sum([1 for score in negatives if score >= threshold])
        TP = sum([1 for score in positives if score >= threshold])
        FN = sum([1 for score in positives if score < threshold])

        values["FPR"].append(float(FP)/(FP+TN))
        values["TPR"].append(float(TP)/(TP+FN))
        values["SPC"].append(float(TN)/(FP+TN))

        y_pred = [int(score >= threshold) for scores in (positives, negatives) for score in scores]
        values["MCC"].append(matthews_corrcoef(y_true, y_pred))
        values["PPV"].append(float(TP)/(TP+FP) if TP+FP>0 else 0.0)
        values["FDR"].append(float(FP)/(TP+FP) if TP+FP>0 else 0.0)
        values["ACC"].append(float(TP+TN)/(len(positives)+len(negatives)))
        
    specificity_curve_inverse = interp1d(values[measure], thresholds)
    saveThreshold = specificity_curve_inverse(measure_threshold) #modified by Alexey 0.95 => measure_threshold
                
    return saveThreshold, thresholds, values

    def KFold_method(self):
        
        kf = KFold(n_splits=10)
        for train_index, test_index in kf.split(self.FeatureSet):
            X_train = []
            X_test = []
            y_train = []
            y_test = []
            for trainid in train_index.tolist():
                X_train.append(self.FeatureSet[trainid])
                y_train.append(self.Label[trainid])

            for testid in test_index.tolist():
                X_test.append(self.FeatureSet[testid])
                y_test.append(self.Label[testid])
            #clf = tree.DecisionTreeClassifier()        
            #clf = clf.fit(X_train, y_train)
            #pre_labels = clf.predict(X_test)
            clf = AdaBoostClassifier(n_estimators=100)
            clf = clf.fit(X_train, y_train)
            pre_labels = clf.predict(X_test)
            # Modeal Evaluation
            ACC = metrics.accuracy_score(y_test, pre_labels)
            MCC = metrics.matthews_corrcoef(y_test, pre_labels)
            SN = self.performance(y_test, pre_labels)
            print ACC, SN

    def evalmetric(pred, truth):
        return 'auc_mine', metrics.roc_auc_score(truth.get_label(), pred)

        thresholds =  np.arange(99.6, 99.9, 0.025)
        bestScore =  0
        bestT = 0
        bestAcc = 0
        bestCf = np.zeros((2,2))

        thresholds = [0.10]
        for t in thresholds:
            temp = np.copy(pred)
            temp[np.where(pred > np.percentile(pred, t))] = 1
            temp[np.where(pred <= np.percentile(pred, t))] = 0
            score = metrics.matthews_corrcoef(truth.get_label(), temp)
            

            if score > bestScore:
                bestScore = score
                bestT = np.percentile(pred, t)
                bestAuc = metrics.roc_auc_score(truth.get_label(), temp, reorder=True)
                bestCf = metrics.confusion_matrix(truth.get_label(), temp)

        
        print('threshold {} mcc {} auc {} TN {} FP {} FN {} TP {}\n'.format(bestT, bestScore, bestAcc, bestCf[0][0], bestCf[0][1], bestCf[1][0], bestCf[1][1]))

        return 'mcc', -1 * bestScore

def compute_MCC(y_true, y_score, threshold_num=500):
    """Compute the Matthews Correlation Coefficient.
    
    :param y_true: true binary labels in range {0, 1}
    :type y_true: numpy array
    :param y_score: the probability estimates of the positive class
    :type y_score: numpy array
    :param threshold_num: the number of thresholds
    :type threshold_num: int
    :return: the maximum Matthews Correlation Coefficient
    :rtype: float
    """

    # Get the ranks
    ranks = get_ranks(y_score)

    # Generate the array which contains the value of thresholds
    threshold_array = np.linspace(np.min(ranks) - 1, np.max(ranks) + 1, num=threshold_num)

    # Generate MCC values
    MCC_list = []
    for threshold in threshold_array:
        MCC_list.append(matthews_corrcoef(y_true, ranks > threshold))
    MCC_array = np.array(MCC_list)

    # Illustrate threshold and MCC values
    # pylab.figure()
    # pylab.plot(threshold_array / np.max(ranks), MCC_array)
    # pylab.show()

    return np.max(MCC_array)

    def Bootstrap_method(self):
        rs = cross_validation.ShuffleSplit(
            len(self.FeatureSet), 10, 0.25, random_state=0)
        clf = tree.DecisionTreeClassifier()
        for train_index, test_index in rs:
            X_train = []
            X_test = []
            y_train = []
            y_test = []
            for trainid in train_index.tolist():
                X_train.append(self.FeatureSet[trainid])
                y_train.append(self.Label[trainid])

            for testid in test_index.tolist():
                X_test.append(self.FeatureSet[testid])
                y_test.append(self.Label[testid])

            #clf = clf.fit(X_train, y_train)
           # pre_labels = clf.predict(X_test)
            clf = AdaBoostClassifier(n_estimators=100)
            clf = clf.fit(X_train, y_train)
            pre_labels = clf.predict(X_test)
            # Modeal Evaluation
            ACC = metrics.accuracy_score(y_test, pre_labels)
            MCC = metrics.matthews_corrcoef(y_test, pre_labels)
            SN = self.performance(y_test, pre_labels)
            print ACC,SN

    def fit_knn(config):
        ### Prepare result holders ###b
        values = {}
        results = {}
        monitors = {}
        E = {"config": config, "results": results, "monitors":monitors, "values":values}

        ### Print experiment header ###
        print_exp_name(config)

        ### Train ###
        monitors["acc_fold"] = []
        monitors["mcc_fold"] = []
        monitors["wac_fold"] = []
        monitors["cm"] = [] # confusion matrix

        results["mean_acc"] = 0
        results["mean_mcc"] = 0


        values["mean_cls"] = Y.mean()
        values["transformers"] = []

        for fold in D["folds"]:
            if config["use_embedding"] == 0:
                tr_id, ts_id = fold["train_id"], fold["test_id"]
                X_train, Y_train, X_test, Y_test = X[tr_id].todense(), Y[tr_id], X[ts_id].todense(), Y[ts_id]
            else:
                X_train, Y_train, X_test, Y_test = fold["X_train"], fold["Y_train"], fold["X_test"], fold["Y_test"]

            if config["use_embedding"] == 0:
                clf = KNeighborsClassifier(n_neighbors=config["KNN_K"], metric="jaccard")
                clf.fit(X_train, Y_train)
                Y_pred = clf.predict(X_test)
            else: # Looking at the similarity of the closest example and getting K=1 from arbitrary K :)
                Y_pred = []
                for x in X_test:
                    Y_pred.append(1 if x[-4] > x[-2] else -1)
                Y_pred = np.array(Y_pred)

            acc_fold, mcc_fold = accuracy_score(Y_test, Y_pred), matthews_corrcoef(Y_test, Y_pred)
            cm = confusion_matrix(Y_test, Y_pred)
            tp, fn, fp, tn = cm[1,1], cm[1,0], cm[0,1], cm[0,0]

            monitors["cm"].append(cm)
            monitors["wac_fold"].append(0.5*tp/float(tp+fn) + 0.5*tn/float(tn+fp))
            monitors["acc_fold"].append(acc_fold)
            monitors["mcc_fold"].append(mcc_fold)

        monitors["acc_fold"] = np.array(monitors["acc_fold"])
        monitors["mcc_fold"] = np.array(monitors["mcc_fold"])
        monitors["wac_fold"] = np.array(monitors["wac_fold"])

        results["mean_acc"] = monitors["acc_fold"].mean()
        results["mean_mcc"] = monitors["mcc_fold"].mean()
        results["mean_wac"] = monitors["wac_fold"].mean()

        logger.info(results)

        return E

def eval_mcc(y_true, y_prob, show=False):
    idx = np.argsort(y_prob)
    y_true = np.array(y_true, dtype=int)
    y_true_sort = y_true[idx]
    n = y_true.shape[0]
    nump = 1.0 * np.sum(y_true)  # number of positive
    numn = n - nump  # number of negative
    tp = nump
    tn = 0.0
    fp = numn
    fn = 0.0
    best_mcc = 0.0
    best_id = -1
    mccs = np.zeros(n)
    for i in range(n):
        if y_true_sort[i] == 1:
            tp -= 1.0
            fn += 1.0
        else:
            fp -= 1.0
            tn += 1.0
        new_mcc = mcc(tp, tn, fp, fn)
        mccs[i] = new_mcc
        if new_mcc >= best_mcc:
            best_mcc = new_mcc
            best_id = i
    best_proba = y_prob[idx[best_id]]
    y_pred = (y_prob > best_proba).astype(int)
    final_mcc = matthews_corrcoef(y_true, y_pred)
    if show:
        return best_proba, final_mcc, y_pred
    else:
        return final_mcc

    def run(self):
        with LpcApocResultTask(self.tname,
                               self.qname,
                               self.subset).output().open('r') as f:
            apoc_parser = ApocResultParer(f.read())

        f = self.output().open('w')

        data = {}
        data['tname'] = self.tname
        data['qname'] = self.qname
        data['t pocket'] = LpcPocketPathTask(self.tname).output().path
        data['q pocket'] = LpcPocketPathTask(self.qname).output().path
        data['Apoc result'] = LpcApocResultTask(self.tname, self.qname, self.subset).output().path
        data['Kcombu result'] = LpcKcombuResult(self.tname, self.qname, self.subset).output().path

        kcombu_data = self._kcombu_results().data
        data['Kcombu tanimoto'] = kcombu_data.tanimoto

        t_coords, q_coords = self._select_ligand_atom_coords()

        global_alignment = apoc_parser.queryGlobal(self.tname, self.qname)
        data['seq identity'] = global_alignment.seq_identity

        pocket_alignment = apoc_parser.queryPocket(self.tname, self.qname)
        if pocket_alignment.has_pocket_alignment:
            t_prt_coords, t_prt_names = self._select_residues(self.tname,
                                                              pocket_alignment.template_chainid,
                                                              pocket_alignment.template_res)

            q_prt_coords, q_prt_names = self._select_residues(self.qname,
                                                              pocket_alignment.query_chainid,
                                                              pocket_alignment.query_res)

            try:
                assert(t_prt_names == q_prt_names)
            except AssertionError:
                raise AssertionError("%s and %s protein residues do not match" % (self.tname, self.qname))

            t_contact = buildArrayOfContact(t_prt_coords, t_coords)
            q_contact = buildArrayOfContact(q_prt_coords, q_coords)
            cms = matthews_corrcoef(t_contact, q_contact)

            data['# residues'] = len(pocket_alignment.template_res)
            data['# ligand atoms'] = len(t_coords)
            data['Apoc ps-score'] = pocket_alignment.ps_score
            data['Apoc p-value'] = pocket_alignment.p_value
            data['Apoc z-score'] = pocket_alignment.z_score
            data['# residue atoms'] = len(t_prt_coords)
            data['t contact'] = t_contact
            data['q contact'] = q_contact
            data['xcms'] = cms

        to_write = json.dumps(data, sort_keys=True, indent=4, separators=(',', ': '))
        f.write(to_write)
        f.close()

        print "xcms output %s" % (self.output().path)

def svc_amino(X, y, score_type):
    """

    :param X:
    :param y:
    :param score_type:
    """
    

    if (score_type=="split"):
        X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=0)

        C = 70  # SVM regularization parameter
        rbf_svc = svm.SVC(kernel='rbf', gamma=0.07, C=C)
        rbf_svc.fit(X_train, y_train)
        y_score = np.array(rbf_svc.predict(X_test))
        y_test = np.array(y_test)
        tn = 0.0
        fp = 0.0

        for i in range(y_score.shape[0]):
            if y_test[i]==-1:
                if y_score[i]==-1:
                    tn = tn+1
                else: fp = fp+1
        spec = tn/(tn+fp)
        print "sensitivity:"
        print recall_score(y_test,y_score)
        print "specificity:"
        print spec
        print "accuracy:"
        print accuracy_score(y_test,y_score)
        print "MCC:"
        print matthews_corrcoef(y_test,y_score)

        


        return "ciao"
        #con binary fa 0.78 con pssm fa 0.80
    else:
        if(score_type=="cross"):
            scores = cross_validation.cross_val_score(rbf_svc, X, np.array(y), cv=5)
            return scores