Data Science Fundemtals 2

An ___ feature has values that are unaffected by other features.

An ___ feature has values that are unaffected by other features.

Input

An ___ feature has values affected by other features.

Output

Residual Error

The difference between the observed and predicted value.

Extrapolation

A prediction that is far beyond the range of the original data.

Simple Linear Regression =

f(0)+mx

Sum of Squared Errors (SSE)

The sum of the squares of all residuals.

Least Squares Regression Line

The simple linear regression formula that minimizes SSE.

Correlation Coefficient

Measures the direction and strength of a linear relationship as a value between 0 and 1.

Fitted vs. Residuals Plots

Displays the predicted values against the residuals.

Normal Q-Q Plot

Displays the sample quantiles against the theoretical quantiles.

Multiple Linear Regression =

f(0) x_0 + f(1) x_1 + … + f(k) x_k

Simple Polynomial Regression =

f(0) x^{0} + f(1) x^{1} + … + f(k) x^{k}

Polynomial Regression Model

A regression model that displays a polynomial relationship between two features.

Interaction Term

A term in a regression model that contains multiple input features.

Logistic Regression =

\frac{e^{b_0 + b_1 x}}{1+e^{b_0 + b_1 x}}

Hot Encoding

Transforming a categorical feature into a numeric feature.

Log-Odds = ln(\frac{p}{1-p}) =

b_0 + b_1 x

Odds Ratio

Compares the relative odds of a outcome given a feature.

A model is ___ if it is too simple to fit the given data.

Underfit

A model is ___ if it is too complex to fit the given data.

Overfit

Ideally, a model ___ pass through every point on a graph.

Shouldn’t

The ___ complex model is preferred over the ___ complex model.

Least, More

Total Error

How much the observed values differ from predicted values.

Bias

How much a model’s prediction differs from the observed values.

Variance

How spread out a model’s predictions are.

Irreducible Error

Error inherent to the situation, unaffected by the model.

A complex model will have more ___ than ___.

Variance, Bias

A simple model will have more ___ than ___.

Bias, Variance

Machine Learning Algorithm

Uses data to build a model that makes predictions.

Regression

A machine learning model used to predict numerical values.

Classification

A machine learning model used to predict categorical values.

Model Training

The process of estimating model parameters used to make a prediction.

___ data is used to fit a model.

Training

___ data is used to evaluate a model’s performance while working on the model.

Validation

___ data is used to evaluatethe final model’s performance compared to other models.

Test

Loss Function

Quantifies the difference between a model’s predictions and the observed values.

Regression Metric

The value returned by a loss function.

The lower the regression metric, the ___ the model is.

Better

Mean Squared Error =

\frac{1}{n} \sum (y_i - \hat{y}_{i})^{2}

Mean Squared Error

A direct measure of a model’s variance.

Mean Absolute Error =

\frac{1}{n} \sum |y_i - \hat{y}_{i}|^{2}

Mean Absolute Error

Like Mean Squared Error, but is less influenced by outliers.

Absolute Loss

Quantifies the loss due to uncertainty.

L_{abs}(y,\hat{p})=|y-\hat{p}| where y is the ___ and \hat{p} is the ___.

Observed class, Predicted probability

An instance is ___ if the output feature’s value is known for that instance.

Labeled

Supervised Learning

Training a model to predict a labeled output feature.

A model is ___ if the relationship between input and output features in the model are easy to explain.

Interpretable

A model is ___ if the outputs produced by the model match the actual outputs with new data.

Predictive

K-Nearest Neighbors

A supervised learning algorithm that predicts the output of a new instance using instances with similar inputs.

Metric

A method of determining the distance between two instances.

Confusion Matrix

A table that summarizes the combinations of predicted and actual values.

Accuracy =

\frac{\text{TP} + \text{TN}}{\text{TP}+\text{FP}+\text{TN}+\text{FN}}

Precision =

\frac{\text{TP}}{\text{TP} + \text{FP}}

Recall =

\frac{\text{TP}}{\text{TP}+\text{FN}}

Receiver Operating Characteristic Curve (ROC Curve)

Measures how well a classification model distinguishes between classes at various probabilties.

Area Under The ROC Curve (AUC)

A metric used to compare the performance between two classification models.

Naive Bayes Classification

A supervised learning classifier that uses the number of times a category occurs in a class to eastimate the likelihood of an instance being in that class.

P(\text{class}|\text{data}) indicates the probability that ___.

The probability of an instance being in \text{class} given \text{data}.

Laplace Smoothing

Adds one ficitonal instance to a class if none exist.

Naive Bayes Classification assumes all categories are ___.

Equally important

Support Vector Machine

A supervised learning algorithm that uses hyperplanes to divide data into different classes.

Hyperplane

A flat surface that is one dimension lower than the input feature space.

A dataset is ___ if a hyperplane can divide the dataset so that all instances of one class fall on one side and everything else falls on the other.

Well-Seperated

Margin

The space between a hyperplane and its supporting vectors.

Support Vectors

The closest instances to a hyperplane.

Vectors on the wrong side of a hyperplane are often given a ___.

Penalty

Hinge Function

Takes the distance from the margin as input, returns a 0 if vector is on the right side and a linear penalty if on the wrong side.

Sensitivity/Recall

The True-Positive rate.

Specificity

The True-Negative rate.

Accuracy

The ratio of the number of correct labels to the total labels.

Missclassification Rate

The ratio of the number of incorrect labels to the total labels.

Missclassification Rate =

1 - \text{Accuracy}

F1 Score

A number between 0 and 1 that represents the harmonic mean of precision and recall.

F1 Score =

2 \frac{\text{Precision} * \text{Recall}}{\text{Precision} + \text{Recall}}

Sensitivity =

\frac{\text{TP}}{\text{TP}+\text{FN}}

Specificity =

\frac{\text{TN}}{\text{TN}+\text{FP}}

Entropy

Describes the number of ways a situation could diverge.

Steps to make a decision tree:

Calculate entropy of decision, split decision’s attributes into subtables and calculate their entropy, choose the attribute with the largest entropy, then repeat the process.

Information Gain

Entropy before split compared to entropy after split.

Heuristic

The attribute that produces the purest node.

Entropy / Expected Information needed to classify tuple D=

\text{Info}(D)=-\sum_{m}^{i=l}p_{i}\log_{2}(p_{i})

Information needed to classify D after using A to split D into v partitions=

\text{Info}_{A}(D)=\sum^{v}_{j=l}\frac{|D_{j}|}{|D|}*I(D_j)

Information gained by branching on attribute A=

\text{Gain}(A)=\text{Info}(D)-\text{Info}_{A}(D)

When picking a distance metric for kNN, the metric doesn’t have to be the ___ on a graph.

Physical distance

The ___ set is used to train the model before testing it.

Training

The ___ set is used to test the model’s abilities after training it.

Testing

Picking an ___ is the 3rd step in creating a kNN model.

Evaluation Metric

The k in kNN represents the ___.

Distance Metric

Unsupervised Learning

Teaching a model to categorize data where no labels are available.

kMeans

An unsupervised learning technique that groups different tuples together based on known attributes.

Centroids

The center points in a cluster for kMeans.

Each cluster in kMeans represents an individual ___.

Attribute

Step 3 of kMeans is to ___.

Move the centroids to the average location of the data points

kMeans should repeat until ___.

The centroids move either very little or not at all.

kMeans has the possible to fall into an ___ or give a ___ answer.

Infinite loop, Useless