Python metrics.zero_one_loss函数代码示例

def test_grid(features, target):
    '''
    Given a list of models for each genre, run the features through the models to
    predict target labels, and compare the predictions to the true target labels.
    '''
    genre_list = ['animated', 'action', 'comedy', 'drama', 'family', 'fantasy', \
    'horror', 'musical', 'mystery', 'romance', 'sci-fi', 'thriller', 'war', 'western']
    ypred_mat = np.empty([target.shape[0], target.shape[1]])
    for i in xrange(target.shape[1]):
        filename = '../data/is_' + genre_list[i] + '.pkl'
        ypred = test_prediction(filename, features, target[:,i])
        for j, prob in enumerate(ypred):
            ypred_mat[j,i] = prob
    with open('../data/grid_pkl_500.txt','w') as f:
        f.write("Model rounded by .25\n")
        yrd = round_by(ypred_mat, .25)
        f.write( metrics.classification_report(target, yrd) )
        f.write( "Percent of misclassification: {}\n".format(metrics.zero_one_loss(target, yrd)) )
        f.write("\nModel rounded by .3\n")
        yrd = round_by(ypred_mat, .3)
        f.write( metrics.classification_report(target, yrd) )
        f.write( "Percent of misclassification: {}\n".format(metrics.zero_one_loss(target, yrd)) )
        f.write("\nModel rounded by .2\n")
        yrd = round_by(ypred_mat, .2)
        f.write( metrics.classification_report(target, yrd) )
        f.write( "Percent of misclassification: {}\n".format(metrics.zero_one_loss(target, yrd)) )
        f.write("\nModel rounded by .1\n")
        yrd = round_by(ypred_mat, .1)
        f.write( metrics.classification_report(target, yrd) )
        f.write( "Percent of misclassification: {}\n".format(metrics.zero_one_loss(target, yrd)) )

def cross_validation_example():
    """ Slightly more complex example : Perform grid search cross-validation to find optimal parameters for MinCq using
    rbf kernels as voters.
    """
    # We load iris dataset, We convert the labels to be -1 or 1, and we split it in two parts: train and test.
    dataset = load_iris()
    dataset.target[dataset.target == 0] = -1
    dataset.target[dataset.target == 2] = -1
    X_train, X_test, y_train, y_test = train_test_split(dataset.data, dataset.target, random_state=42)

    # The learning algorithm and its parameters.
    learner = MinCqLearner(mu=0.0001, voters_type='kernel', kernel='rbf', gamma=0.0)
    learner_params = {'mu': [0.0001, 0.001, 0.01],
                      'gamma': [0.0, 0.1, 1.0, 10]}

    cv_classifier = GridSearchCV(learner, learner_params, scoring=accuracy_scorer)
    cv_classifier = cv_classifier.fit(X_train, y_train)

    predictions_train = cv_classifier.predict(X_train)
    predictions_test = cv_classifier.predict(X_test)

    print_sklearn_grid_scores("Iris", "RbfMinCq", learner_params, cv_classifier.grid_scores_)

    print("Best parameters: {}".format(str(cv_classifier.best_params_)))
    print("Training set risk: {:.4f}".format(zero_one_loss(y_train, predictions_train)))
    print("Testing set risk: {:.4f}".format(zero_one_loss(y_test, predictions_test)))

def drawLearningCurve(model, x_train, y_train, x_test, y_test, num_points = 50):
# adapted from http://sachithdhanushka.blogspot.de/2013/09/learning-curve-generator-for-learning.html
    
    train_error = np.zeros(num_points)
    crossval_error = np.zeros(num_points)
    
    sizes = np.linspace(2, len(x_train), num=num_points).astype(int)
    for i,size in enumerate(sizes):
         
        #getting the predicted results of the model
        model.fit(x_train[:size], y_train[:size])
         
        #compute the validation error
        y_pred = model.predict(x_test[:size])
        crossval_error[i] = zero_one_loss(y_test[:size], y_pred, normalize=True)
         
        #compute the training error
        y_pred = model.predict(x_train[:size])
        train_error[i] = zero_one_loss(y_train[:size], y_pred, normalize=True)

    #draw the plot
    print crossval_error
    print train_error
    fig,ax = plt.subplots()
    ax.plot(sizes,crossval_error,lw = 2, label='cross validation error')
    ax.plot(sizes,train_error, lw = 4, label='training error')
    ax.set_xlabel('cross val error')
    ax.set_ylabel('rms error')
    ax.legend(loc = 0)
    ax.set_title('Learning Curve' )
    return fig

def plot_adaclassifier(classifier, n_estimators, X_train, X_test, y_train, y_test):

    fig = plt.figure()
    ax = fig.add_subplot(111)

    #ax.plot([1, n_estimators], [dt_stump_err] * 2, 'k-',
    #        label='Decision Stump Error')
    #ax.plot([1, n_estimators], [dt_err] * 2, 'k--',
    #        label='Decision Tree Error')

    ada_err_test = np.zeros((n_estimators,))
    for i, y_pred in enumerate(classifier.staged_predict(X_test)):
        ada_err_test[i] = zero_one_loss(y_pred, y_test)

    ada_err_train = np.zeros((n_estimators,))
    for i, y_pred in enumerate(classifier.staged_predict(X_train)):
        ada_err_train[i] = zero_one_loss(y_pred, y_train)

    ax.plot(np.arange(n_estimators) + 1, ada_err_test,
            label='AdaBoost Test Error',
            color='red')
    ax.plot(np.arange(n_estimators) + 1, ada_err_train,
            label='AdaBoost Train Error',
            color='blue')

    ax.set_ylim((0.0, 1.0))
    ax.set_xlabel('n_estimators')
    ax.set_ylabel('error rate')

    leg = ax.legend(loc='upper right', fancybox=True)
    leg.get_frame().set_alpha(0.7)


    return fig

def run_gamma(x, y):
    perc = 0.6
    n = x.shape[0]
    gamma_list = (np.power(2.0, range(-4, 12))/(n*perc)).tolist()
    n_iter = 2
    train_err_libsvm = np.zeros((len(gamma_list), n_iter))
    test_err_libsvm = np.zeros((len(gamma_list), n_iter))
    train_err_dsvm = np.zeros((len(gamma_list), n_iter))
    test_err_dsvm = np.zeros((len(gamma_list), n_iter))
    train_err_pegasos = np.zeros((len(gamma_list), n_iter))
    test_err_pegasos = np.zeros((len(gamma_list), n_iter))
    ss = cv.StratifiedShuffleSplit(y, n_iter=n_iter, test_size=1-perc, train_size=None, random_state=0)
    for k, (train, test) in enumerate(ss):
        ntr = len(train)
        lmda = 1.0 / ntr
        print "#iter: %d" % k
        x_train, x_test, y_train, y_test = x[train], x[test], y[train], y[test]
        mM_scale = preprocessing.MinMaxScaler(feature_range=(-1, 1))
        x_train = mM_scale.fit_transform(x_train)
        x_test = mM_scale.transform(x_test)
        for j, gm in enumerate(gamma_list):
            print "check lamda %f, gamma %f" % (lmda, gm)
            clf = svm.SVC(C=lmda * ntr, kernel='rbf', gamma=gm, cache_size=600)
            clf.fit(x_train, y_train)

            pred = clf.predict(x_train)
            train_err_libsvm[j, k] = zero_one_loss(y_train, pred)
            pred = clf.predict(x_test)
            test_err_libsvm[j, k] = zero_one_loss(y_test, pred)
            dsvm = DualKSVM(lmda=lmda, gm=gm, kernelstr='rbf', nsweep=ntr/2, b=5, c=1)
            dsvm.fit(x_train, y_train, x_test, y_test, )
            train_err_dsvm[j, k] = dsvm.err_tr[-1]
            test_err_dsvm[j, k] = dsvm.err_te[-1]
            kpega = Pegasos(ntr, lmda, gm, nsweep=2, batchsize=2)
            kpega.train_test(x_train, y_train, x_test, y_test)
            train_err_pegasos[j, k] = kpega.err_tr[-1]
            test_err_pegasos[j, k] = kpega.err_te[-1]
    avg_train_err_libsvm = np.mean(train_err_libsvm, axis=1)
    avg_test_err_libsvm = np.mean(test_err_libsvm, axis=1)
    avg_train_err_dsvm = np.mean(train_err_dsvm, axis=1)
    avg_test_err_dsvm = np.mean(test_err_dsvm, axis=1)
    avg_train_err_pegasos = np.mean(train_err_pegasos, axis=1)
    avg_test_err_pegasos = np.mean(test_err_pegasos, axis=1)
    plt.figure()
    # color_list = ['b', 'r', 'g', 'c', ]
    # marker_list = ['o', 'x', '>', 's']

    plt.loglog(gamma_list, avg_train_err_libsvm, 'bo-', label='libsvm train')
    plt.loglog(gamma_list, avg_test_err_libsvm, 'ro-', label='libsvm test')
    plt.loglog(gamma_list, avg_train_err_dsvm, 'gx-', label='dsvm train')
    plt.loglog(gamma_list, avg_test_err_dsvm, 'cx-', label='dsvm test')
    plt.loglog(gamma_list, avg_train_err_pegasos, 'mD-', label='pegasos train')
    plt.loglog(gamma_list, avg_test_err_pegasos, 'kD-', label='pegasos test')
    plt.legend(bbox_to_anchor=(0, 1.17, 1, .1), loc=2, ncol=2, mode="expand", borderaxespad=0)
    plt.savefig('../output/usps_diff_gamma.pdf')

def build_tree(clf,type,i,X_train, X_test, y_train, y_test,attribute_names,class_names):
    print("------------Run "+type+ "_"+str(i)+"----------")
    clf.fit(X_train, y_train)
    print("Training error =", zero_one_loss(y_train, clf.predict(X_train)))
    predicted_test = clf.predict(X_test)
    print("Test error =",zero_one_loss(y_test, predicted_test ) )
    figure_name = type+"_"+str(i)
    visualize_tree(clf,attribute_names,class_names,figure_name)
    print(classification_report(  y_test,predicted_test ))
    print(confusion_matrix(y_test,predicted_test))
    return zero_one_loss(y_test, predicted_test )

def simple_classification_example():
    """ Simple example : with fixed hyperparameters, run four versions of MinCq on a single dataset.
    """
    # MinCq parameters, fixed to a given value as this is a simple example.
    mu = 0.001

    # We load iris dataset, We convert the labels to be -1 or 1, and we split it in two parts: train and test.
    dataset = load_iris()
    dataset.target[dataset.target == 0] = -1
    dataset.target[dataset.target == 2] = -1
    X_train, X_test, y_train, y_test = train_test_split(dataset.data, dataset.target, random_state=42)

    # We train MinCq using decision stumps as voters, on the training set.
    learner = MinCqLearner(mu, voters_type='stumps')
    learner.fit(X_train, y_train)

    # We predict the train and test labels and print the risk.
    predictions_train = learner.predict(X_train)
    predictions_test = learner.predict(X_test)

    print("\nStumpsMinCq")
    print("-----------")
    print("Training set risk: {:.4f}".format(zero_one_loss(y_train, predictions_train)))
    print("Testing set risk: {:.4f}\n".format(zero_one_loss(y_test, predictions_test)))

    # We do the same again, now with a linear kernel.
    learner = MinCqLearner(mu, voters_type='kernel', kernel='linear')
    learner.fit(X_train, y_train)

    predictions_train = learner.predict(X_train)
    predictions_test = learner.predict(X_test)

    print("\nLinearMinCq")
    print("-----------")
    print("Training set risk: {:.4f}".format(zero_one_loss(y_train, predictions_train)))
    print("Testing set risk: {:.4f}\n".format(zero_one_loss(y_test, predictions_test)))

    # We do the same again, now with a polynomial kernel.
    learner = MinCqLearner(mu, voters_type='kernel', kernel='poly')
    learner.fit(X_train, y_train)

    predictions_train = learner.predict(X_train)
    predictions_test = learner.predict(X_test)

    print("\nPolyMinCq")
    print("-----------")
    print("Training set risk: {:.4f}".format(zero_one_loss(y_train, predictions_train)))
    print("Testing set risk: {:.4f}\n".format(zero_one_loss(y_test, predictions_test)))

    # We do the same again, now with an RBF kernel.
    learner = MinCqLearner(mu, voters_type='kernel', kernel='rbf', gamma=0.0)
    learner.fit(X_train, y_train)

    predictions_train = learner.predict(X_train)
    predictions_test = learner.predict(X_test)

    print("\nRbfMinCq")
    print("--------")
    print("Training set risk: {:.4f}".format(zero_one_loss(y_train, predictions_train)))
    print("Testing set risk: {:.4f}\n".format(zero_one_loss(y_test, predictions_test)))

def test_losses():
    """Test loss functions"""
    y_true, y_pred, _ = make_prediction(binary=True)
    n_samples = y_true.shape[0]
    n_classes = np.size(unique_labels(y_true))

    # Classification
    # --------------
    with warnings.catch_warnings(record=True):
    # Throw deprecated warning
        assert_equal(zero_one(y_true, y_pred), 11)

    assert_almost_equal(zero_one_loss(y_true, y_pred),
                        11 / float(n_samples), 2)
    assert_equal(zero_one_loss(y_true, y_pred, normalize=False), 11)

    assert_almost_equal(zero_one_loss(y_true, y_true), 0.0, 2)
    assert_almost_equal(hamming_loss(y_true, y_pred),
                        2 * 11. / (n_samples * n_classes), 2)

    assert_equal(accuracy_score(y_true, y_pred),
                 1 - zero_one_loss(y_true, y_pred))

    with warnings.catch_warnings(True):
    # Throw deprecated warning
        assert_equal(zero_one_score(y_true, y_pred),
                     1 - zero_one_loss(y_true, y_pred))

    # Regression
    # ----------
    assert_almost_equal(mean_squared_error(y_true, y_pred),
                        10.999 / n_samples, 2)
    assert_almost_equal(mean_squared_error(y_true, y_true),
                        0.00, 2)

    # mean_absolute_error and mean_squared_error are equal because
    # it is a binary problem.
    assert_almost_equal(mean_absolute_error(y_true, y_pred),
                        10.999 / n_samples, 2)
    assert_almost_equal(mean_absolute_error(y_true, y_true), 0.00, 2)

    assert_almost_equal(explained_variance_score(y_true, y_pred), 0.16, 2)
    assert_almost_equal(explained_variance_score(y_true, y_true), 1.00, 2)
    assert_equal(explained_variance_score([0, 0, 0], [0, 1, 1]), 0.0)

    assert_almost_equal(r2_score(y_true, y_pred), 0.12, 2)
    assert_almost_equal(r2_score(y_true, y_true), 1.00, 2)
    assert_equal(r2_score([0, 0, 0], [0, 0, 0]), 1.0)
    assert_equal(r2_score([0, 0, 0], [0, 1, 1]), 0.0)

def experiment_neighbors_k_nearest_neighbors():
    avgError = []
    x_learners = []
    for k_neighbors in range(1, 20, 1):
        k = 10
        skf = StratifiedKFold(labels,n_folds=k)
        averageError = 0.0
        for train_index, test_index in skf:
            X_train, X_test = mfcc[:,train_index], mfcc[:,test_index]
            y_train, y_test = labels[train_index], labels[test_index]
            knc = KNeighborsClassifier(n_neighbors=k_neighbors, weights='distance')
            knc.fit(X_train.T,y_train)
            y_pred = knc.predict(X_test.T)
            error = zero_one_loss(y_pred,y_test)
            print error
            averageError += (1./k) * error
        print "Average error: %4.2f%s" % (100 * averageError,'%')
        avgError.append(averageError)
        x_learners.append(k_neighbors)

    plt.plot(x_learners, avgError)
    plt.ylabel('Average Error (k=10)')
    plt.xlabel('Number of Neighbors')
    plt.title('Error as a function of the number of neighbors taken into consideration')
    plt.show()

def cross_valid(h, y, ratio_list):
    """
    cross validation to tune the best cap probability for soft-margin boosting
    """
    print " find optimal ratio"
    n_samples = h.shape[0]
    n_folds = 4
    ntr = n_samples/n_folds
    ratio_list = ratio_list[ratio_list >= 1.0/ntr]
    kf = cv.KFold(n=n_samples, n_folds=n_folds)
    err_tr = np.zeros((n_folds, len(ratio_list)))
    err_te = np.zeros((n_folds, len(ratio_list)))
    k = 0
    for tr_ind, te_ind in kf:
        print "nfold: %d" % (k)
        xtr, ytr, xte, yte = h[tr_ind, :], y[tr_ind], h[te_ind, :], y[te_ind]
        for i, r in enumerate(ratio_list):
            pd = ParaBoost(epsi=0.005, has_dcap=True, ratio=r)
            pd.train(xtr, ytr)
            pred = pd.test_h(xte)
            err_te[k, i] = zero_one_loss(y_true=yte, y_pred=pred)
            err_tr[k, i] = pd.err_tr[-1]
        k += 1
    err_te_avg = np.mean(err_te, axis=0)
    err_tr_avg = np.mean(err_tr, axis=0)
    arg = np.argmin(err_te_avg)
    best_ratio = ratio_list[arg]
    err = err_te_avg[arg]
    return best_ratio

def apply_dbn(files, main_driver=1):
    """
    Applies DBN for identifying trips which are not from the driver of interest
    """
    (X_train, Y_train, weight, X, driver_trip_arr) = \
        get_train_data(files, main_driver)
    a = np.empty(shape=[0, 2])

    net = DBN([len(COL), 10, 2],
              learn_rates=0.3,
              learn_rate_decays=0.9,
              epochs=10,
              verbose=0)
    net.fit(X_train, Y_train)

    Y_dbn = net.predict(X_train)
    print main_driver, ':', 1 - zero_one_loss(Y_train, Y_dbn)
    # print "Classification report:"
    # print classification_report(Y_train, preds)

    i = 0
    Y = net.predict(X)
    for y in Y:
        driver_trip = driver_trip_arr[i][0]
        prob = str(int(Y[i]))
        a = np.append(a, np.array([[driver_trip, prob]]), axis=0)
        i = i + 1

    print main_driver, ': ', sum([1 for p in a if p[1] == '1'])

    return a

def experiment_learners_random_forest():
    avgError = []
    x_learners = []
    for maxLearners in range(10, 150, 20):
        k = 10
        skf = StratifiedKFold(labels,n_folds=k)
        averageError = 0.0
        for train_index, test_index in skf:
            X_train, X_test = mfcc[:,train_index], mfcc[:,test_index]
            y_train, y_test = labels[train_index], labels[test_index]
            rf = RandomForestClassifier(n_estimators=maxLearners, max_depth = maxDepth, warm_start = False)
            rf.fit(X_train.T,y_train)
            y_pred = rf.predict(X_test.T)
            error = zero_one_loss(y_pred,y_test)
            print error
            averageError += (1./k) * error
        print "Average error: %4.2f%s" % (100 * averageError,'%')
        avgError.append(averageError)
        x_learners.append(maxLearners)

    plt.plot(x_learners, avgError)
    plt.ylabel('Average Error (k=10)')
    plt.xlabel('Max Learners')
    plt.title('Error as a function of the number of learners')
    plt.show()

def experiment_estimators_AdaBoostRandomForest():
    avgError = []
    x_learners = []
    rf = RandomForestClassifier(n_estimators=maxLearners, max_depth = maxDepth, warm_start = False)
    for lr in frange(0.01, 1., 0.25):
        k = 10
        skf = StratifiedKFold(labels,n_folds=k)
        averageError = 0.0
        for train_index, test_index in skf:
            X_train, X_test = mfcc[:,train_index], mfcc[:,test_index]
            y_train, y_test = labels[train_index], labels[test_index]
            adb = AdaBoostClassifier(base_estimator=rf, n_estimators=100, learning_rate=lr)
            adb.fit(X_train.T,y_train)
            y_pred = adb.predict(X_test.T)
            error = zero_one_loss(y_pred,y_test)
            print error
            averageError += (1./k) * error
        print "Average error: %4.2f%s" % (100 * averageError,'%')
        avgError.append(averageError)
        x_learners.append(lr)
    # graph the errors now.
    plt.plot(x_learners, avgError)
    plt.ylabel('Average Error (k=10)')
    plt.xlabel('Learning Rate')
    plt.title('Error as a function of the learning rate')
    plt.show()

def experiment_pca_n_components_random_forest():
    pca = decomposition.PCA()
    rf = RandomForestClassifier(n_estimators=maxLearners, max_depth = maxDepth, warm_start = False)
    pipe = Pipeline(steps=[('pca', pca), ('rf', rf)])
    avgError = []
    x_learners = []
    for k_components in range(10, 100, 10):
        k = 10
        skf = StratifiedKFold(labels,n_folds=k)
        averageError = 0.0
        for train_index, test_index in skf:
            X_train, X_test = mfcc[:,train_index], mfcc[:,test_index]
            y_train, y_test = labels[train_index], labels[test_index]
            estimator = GridSearchCV(pipe, dict(pca__n_components=[k_components]))
            estimator.fit(X_train.T,y_train)
            y_pred = estimator.predict(X_test.T)
            error = zero_one_loss(y_pred,y_test)
            print error
            averageError += (1./k) * error
        print "Average error: %4.2f%s" % (100 * averageError,'%')
        avgError.append(averageError)
        x_learners.append(k_components)

    plt.plot(x_learners, avgError)
    plt.ylabel('Average Error (k=10)')
    plt.xlabel('Number of Components')
    plt.title('Error as a function of the number of components')
    plt.show()

 def classify(self, model, test_y, test_x):
     pred = model.predict(test_x)
     if not self.multi:
         rec, spec, acc = self.score(pred, test_y)
         return rec, spec, acc
     else:
         return 1 - zero_one_loss(test_y, pred)