Test
Machine Learning
Crea una cuenta gratis para terminar este test
Guarda tu progreso y consulta tu resultado final.
¿Ya tienes cuenta? Inicia sesión
Preguntas del test (20)
1What is a ROC curve?
- AGraphical plot that illustrates the performance of a binary classifier, as its threshold is varied
- BGraphical plot that illustrates the performance of a classifier, as its threshold is varied, representing the relation between True positive rate
and False positive rate
2Select the one which is TRUE:
- AA more reliable mapping between input features and response can be built by combining the output of multiple experts or models
- BWhen you can build classifiers that are more accurate than chance and, more importantly, that are dependent from each other, then you can build an ensemble
3Why is Naive Bayes called Naive?
- ABecause it assumes that the decision boundary to separate the classes is linear
- BBecause it assumes that data is normally distributed
- CBecause it assumes independence between the features
- DBecause it assumes dependence between the features
4What is the difference between LDA and QDA?:
- ALDA is more flexible
- BQDA is less flexible
- CQDA presents lower variance
- DLDA presents lower variance
5Select the one which is NOT true when it comes to PCA:
- AIt seeks to find linear combinations of the predictors which capture the most possible variance
- BIt is unsupervised
- CFinds the linear combinations of the original variables that gives the best possible linear separation between the classes
- DSmall eigenvalues are related to irrelevant variables
6If you have independent events (Select the one that is true):
- AYou cannot compute its conditional probability
- BTheir conditional and joint probabilities are equal
- CTheir joint probability is equal to the product of the individual event probabilities (marginal probabilities)
- DTheir joint probability is o, since they do not depend on each other
7How could you try to build a classifier that would match a complicated (highly non-linear) decision boundary separating the positive and negative classes in a binary classification problem?
- AUsing a Nearest Neighbor Method
- BBy increasing the regularization parameter to reduce the variance
- CUsing a simpler model such as Logistic Regression
- DAvoiding tree-based models
8How can it be avoided the overfitting of tree-based models?
- ATree-based models do not overfit.
- BUsing cross-entropy as cost function
- CIncluding a cost complexity factor as regularization
- DUsing a boosting method such as XGBoost
9Eigenvalues:
- AMeasure the eigeness of the data
- BGive us the directions of the principal axes of the cloud of data points
- CPoint along the direction of greatest variance
- DGive us the overall test MSE over the principal components
10Select from the following the one which is NOT true about KNN
- AIt is a lazy methodology
- BYou should always scale/standardize feature values
- CIt is a distance-based classifier
- DIt finds the decision boundary with the largest margin
11In what relies PCA to reduce the dimension of a dataset?
- AThe correlation between variables
- BThe independence between variables
- CIn finding the separation function with the largest margin
- DIn finding the hyperplane that produces the lower variance
12Select the one that better describes how a KDE works:
- AEstimates the density directly from the data without assuming a particular form for the underlying distribution
- BIt is not a generative method
- CDensity estimators are very similar to KNN
- DKDE classifiers allows, for any observation "x", and its label "y", to compute the likelihood "Ply x)"
13Select the one which is NOT true when it comes to cross-validation:
- ACross-validation is considered a re-sampling method to avoid optimistic error estimates
- BIt is mainly used to estimate how accurately a predictive model will perform in practice
- CThe goal of cross-validation is to test the model's ability to predict new data that was not used during the training phase
- DIn k-fold cross-validation, the original sample is partitioned into subsamples of "k" elements, each
14Select the one which is NOT true:
- ATree based methods find informative attributes from data
- BTree based methods can be used for classification and regression
- CTrees produce multivariate unsupervised segmentation
- DTrees are non parametric classifiers
15"All models are wrong but some are useful" What do we mean by that?:
- AYou cannot trust your machine learning models
- BYou cannot trust your data
- CYou can make foolish assumptions and still get an useful model
- DYou expect to be lucky when you train a machine learning model
16Select the one which is NOT true:
- APrecission is the fraction of predictions that are actually in the class
- BRecall is the fraction of observations in the class that are actually detected
- CPrecission is a measure of confirmation (how often my model predictions are incorrect)
- DRecall is a measure of utility (how much my model finds what it has to find)
17When using wrapper methods for model selection (select the one that applies)
- AUsing indirect techniques is always better than directly measuring the estimate error over a cross validation data set
- BComputational resources can be a problem, and must be carefully considered before selecting cross validation over less computationally expensive indirect techniques
- CDirectly measuring the cross-validation error is always better than using indirect techniques
- DWe always have to estimate the error variance in cross validation
18Select from the following which is NOT a method to reduce the variance of a predictive model:
- ARegularization
- BBagging
- CBoosting
- DCross-Validation
19Select form the following the one which is NOT true:
- AIf you do not know what to try first, fit a linear regression to your data and evaluate the outcome
- BRandom Forest, as any other methodology based on Decision Trees, is preferred when you suspect that linear combinations of features are critical
- CRidge regression will include all p predictors in the final model because it does not perform subset selection
- DMore complex models reduce the bias at the cost of increasing the variance
20If you have an unbalanced distribution of your target variable, Which evaluation metric would you use?:
- AMatthew's Correlation Coefficient
- BAccuracy
- CRecall
- DF1
Información
- Descripción
- Test sobre Machine Learning, creado por Catharina Doria el 29/03/2022.
- Autor
- Catharina Doria
- Idioma
- Português
- Visitas
- 2
- Etiquetas
- Valoración
- Aún sin valoraciones Valora este test