Created by Ali Poursotoudeh Tehrani
about 2 years ago
|
||
Question | Answer |
What is machine learning | 1.) Building models that automatically improve through experience e.g. the use of data to improve in solving problems without being explicitly programmed to |
How could you categorize Machine learning | 1.) By problem statement: Classification , Regression 2.) By data handling : Supervised unsupervised 3.) By the five tribes : Symbolist, Connectionists, Evolutionaries, Bayesians, Analogizers |
Describe: Data Data point Data set | Data: Variable + Value Datapoint = Multiple Data about one resource (vector) Dataset = Collection of multiple datapoints ( matrix) |
What problems with data do often occur in Machine learning. How could they be solved. | Data is too big : one hot encoding, Feature reduction, pca lda Data is not from same measurements and therefore hard to compare : Use scaling and normalization |
Describe: Min Max Scaling Z- Score Normalization Log transformation | Mix max: Brining data into an specific range X = X - Xmin / Xmax - Xmin Z- score: Bring data into gaussian X - mean/ stddev Log transformation: Just add log to remove skewness x = log(x) |
Descirbe: Clustering Soft vs Hard clustering K-means K-means ++ | Clustering: Putting similiar objects in to an group with an unsupervised algorithm Soft clustering: Obeject has probabilities for each group Hard clustering : Object definitly belongs into a group K-means: 1.) Randomly assign k centroids 2.) Assign to points to group in neares centroid 3.) Recalculate centroid 4.) repeat until nothing changes anymore K-means++ : Advanced initialziation algorithm for k-means, first centroid = random datapoint, p( point = next centroid) is proptional to distance of last choosen centroid |
Write down distance metrics: L1 norm manhattan L2 norm euclidian Cos distance | Just lookup google |
Describe hierachical clustering Agglomerative (bottom up) Divisive ( top down) | Agglomerative: In each iteration merge most similiar cluster until only 1 big cluster remains Divisive: Start with 1 big cluster and remove farthest points from centeruu |
Want to create your own Flashcards for free with GoConqr? Learn more.