Flashcards
Validation
Flashcards do conjunto (23)
- 1Real EffectsReal relationship between attributes and response variables
- 2Random EffectsRandom but looks like real effects
- 3Difference between Real and Random EffectsReal effects are the same in all data sets
Random effects are different in each data sets - 4Training Data SetLarger set of data to fit (build) the model
- 5Validation Data SetSmaller set of data to measure the model's effectiveness and choose a model
- 6Test Data SetTo Estimate performance of chosen model
- 7Splitting DataRandom
Rotation
e.g. 5 data point rotation sequence - 8Cross ValidationTo avoid problem that some important data only appears in the validation or test datasets
- 9K-Fold Cross ValidationOne of the Cross Validation methods
- 10How does K-fold CV workSeparate the whole data set to K parts and each of them will be act as training or validation dataset
- 11K-Fold CV, which model to pickDo not average the coefficients across the splits
Tran the model again using all the data - 12ClusteringGrouping Data Points
- 13Euclidean distanceaka Straight-line distance
2-norm
- 14Rectilinear distanceaka Manhattan distance
1-norm
- 15Minkowski distanceaka p-norm distance
the generalized formula of all distance formula
- 16Infinity norm distancethe largest (absolute) of a set of numbers
e.g. Machine go and put the box on top of a pile of boxes
- 17K-Means Clustering
- 18Why K-means Clustering is heuristic?Fast, good but not guaranteed to find absolute best solution
BUT K-means usually gets really close to the best solution - 19Expectation-Maximization algorithmK-mean clustering is an example of EM
EM: 1. Alternate expectation step (find cluster centers)
2. maximization step (assign points to clusters) - 20Compare different K-means clusteringElbow diagram
- 21Difference between Classification vs ClusteringClassification: We know what the correct classification is
Clustering: We DO NOT KNOW what the correct classification is - 22Supervised LearningCorrect answer (response) is known
- 23Unsupervised LearningCorrect answer (response) is unknown