Interview Prep

Bias is error due to erroneous or overly simplistic assumptions in algorithm. This can lead to the model underfitting your data, making it hard to generalize your knowledge from the training set to the test set. Variance is error due to too much complexity in algorithm. Leads to the algorithm being highly sensitive and overfitting.

Supervised = used data that is labeled Unsupervised = not labeled

KNN = supervised classification algorithm K-means clustering is unsupervised

Receiver operating characteristic –> representation of true positive rates vs false positive rates

Email spam Precision = TP / predicted positive (actual cancer patients / pred cancer patients) Recall = TP / Real positive (Actual cancer patients / actual cancer patients + cancer patients not predicated)

Type I = False positive Type II = False negative

Holding out different parts of the data to test the model on

Branches that have lower predictive power are removed in order to reduce complexity

weighted avg between recall and precision 1 = best 0 = worst

Keep models simple Use regularization (Lasso / Ridge) Use cross-validation

Decision Tree + Boosting

Delete row or decide to replace with 0 or another value (mean or median) or prediction

1) Define cost function (total squared error) to minimize 2) Initialize gradient descent 3) Iterate on gradient decent based on alpha (learning rate) and number of iterations 4) Analyze results

Ensure all data is on the same scale = x - mean / SD

1) Get access to data 2) EDA to get familiar with data + understand any issues 3) Determine how to prepare the data 4) Explore models to use 5) Execute, visualize, and inform

Variance = how apart numbers are from mean Covariance = how the mean of two variables move together

with a large enough sample size, the sample mean and sd will match the population – you can use z-score to get prob

• There is a linear relationship between the dependent variables and the regressors • The errors or residuals of the data are normally distributed and independent from each other • There is minimal multicollinearity between explanatory variables • Homoscedasticity. This means the variance around the regression line is the same for all values of the

Lists, Dictionaries, Sets, Strings, Tuples, Boolean, Numeric Tuples are immutable (along with frozensets)

Use train, test, and validation – validation allows you to fine tune parameters before running final test 80:20 typically appropriate but depends on the data you have

Linear regression Regression trees Neural Network

Support Vector Machines Decisions Trees KNN

KNN calculates similarity scores SMV - looks at difference between hyperplanes

git pull git add file/folder git commit -m “comment” git push