Use LEFT and RIGHT arrow keys to navigate between flashcards;
Use UP and DOWN arrow keys to flip the card;
H to show hint;
A reads text to speech;
40 Cards in this Set
- Front
- Back
What are the 6 components of prediction? |
Question Input Feature Algorithm Parameters Evaluation |
|
What are the properties of good features? |
lead to data compression Retain relevant information Are created based on expert application knowledge |
|
What are common mistakes with features? |
Trying to automate feature selection not paying attention to data-specific quirks Throwing away information unnecessarily |
|
What criteria is used to evaluate machine learning methods? |
Interpretable Simple Accurate Fast Scalable |
|
What is sample error? |
The error rate you get on the same data set you used to build your predictor |
|
What is out of sample error? |
The error rate you get on new data set. |
|
What is another term for out of sample error? |
Generalization error |
|
What is more important, sample error or out of sample error? |
Out of sample error |
|
What is the reason for a larger out of sample error compared to in sample error? |
Overfitting |
|
Why does a machine learning method overfit? |
The algorithm captures both the signal and the noise |
|
What are the 6 steps in predictive study design? |
Define your error rate Split data into training, testing, validation Pick features Pick method Apply method to test data and refine Apply method to validation data |
|
How should training, test and validation data sets be separated? |
Randomly |
|
What are the different types of classification errors? |
True positive False positive True negative False negative |
|
What is the formula for sensitivity? |
True Positive / (True Positive + False Negative) |
|
What is the formula for specificity? |
True negative / (False Positive + True negative) |
|
How do you plot ROC curves? |
Y axis: True Positive X axis: False positive |
|
What does ROC stand for? |
Receiver operating characteristic |
|
How do you evaluate an ROC curve? |
The more area under the curve the better |
|
What is the main principal of cross validation? |
Train the model repeatedly with only a subset of the training data Use the excluded data to evaluate the models |
|
What are 4 use cases of cross validation? |
Picking variables to include in a model Picking the type of prediction function to use Picking the parameters in the prediction function Comparing different predictors |
|
What are 2 different cross validation methods? |
K-folds Leave one out |
|
Caret method to preprocess data? |
preProcess |
|
What are 4 useful caret functions to partition data? |
createDataPartition createResample createTimeSlices createFolds |
|
What caret method is used to train a model? |
train |
|
What caret method is used to make predictions using a model and data? |
predict |
|
What caret function is useful for comparing predicted results with actual results? |
confusionMatrix |
|
What property on a caret model give the results of the model? |
finalModel |
|
What are 2 metrics that can be used to evaluate continues models? |
Root mean squared error(RMSE) RSquared |
|
What are 2 metrics that can be used to evaluate categorical models? |
Accuracy(fraction correct) Kappa(measure of concordance) |
|
What caret method creates a plot of all features against an outcome? |
featurePlot |
|
What function breaks a group of data into quantiles based on number of bins? |
cut2(Hmisc package) |
|
What attributes can be set on the preProcess method to standardize features on a data set? |
method=c("center", "scale") |
|
What transformation tries to make continuous data look like normal data? |
Box Cox |
|
What plot shows sample quantiles vs theoretical quantiles? |
Normal Q-Q plot |
|
What algorithm is useful for imputing data? |
k nearest neighbors? |
|
What method must be set on preProcess to impute the data using k nearest neighbors? |
knnImpute |
|
What are the two levels of covariate creation? |
level 1: From raw data to covariate level 2: Transforming tidy covariates |
|
What must be considered what converting raw data into covariates? |
Summarization vs information loss |
|
What must categorical variables be for a ML algorithm to work? |
dummy variables (0 or 1) |
|
What caret function converts categorical variables into dummy variables? |
dummVars |