Bayesian Models look at

Prob. something happening, and prob. of it happening given evidence

The derivative of the error with respect to a model parameter tells you...

How much of the error that parameter was responsible for, How much that model parameter effects changes in the error

How do we think of an error surface?

F(model parameters) = model error

Stochastic gradient descent

Makes a change to model parameters after each training example

The derivative of the model error is different from the learning rate because...

derivative assigns responsibility, learning rate assigns how much change

A machine learning parameter is...

Something the engineer may set, but the model changes during training

The prior distribution is....

How likely the hypothesis would occur in the 1st place given what you know

A Random Forest classifier

is multiple decision trees working together, is an ensemble classifier

A Decision Tree chooses a split based on

Maximizing information gain, Minimizing entropy

Support Vector Machines

Find a hyperplane that separates the data into classes

Train subsets of the data on multiple models

A Naiive Bayes classifier in sci-kit learn

Estimates the prior probability from the data distribution

What part of the machine learning model DOES NOT GET CHANGED?

Your training data (unless you collect some later)

Splits the features based on information theory

A machine learning hyperparameter is...

Something engineer sets, but model doesn't change during training

Sets how much change to make to a model parameter, Is a hyperparmeter of model training

Iteratively train models to fit cases initial models did not fit well

Batch gradient descent

Makes a change to model parameters after a subset of training examples

studbud

ML Basics and Tabular Models Review

29 termsStudBud Administration14 plays

We will be reviewing basic Machine Learning parameters and models

Flashcards29 cardsFPractice quizPHost liveL

Questions 29

Answers

1
Bayesian Models look at
Prob. something happening, and prob. of it happening given evidence
Everything BUT the prior distribution of a variable
Only the probability of something not occuring
The split of the model based on information theory
2
The derivative of the error with respect to a model parameter tells you...
How you should choose your next hyper-parameter
How big of a change you should make to the error
How much of the error that parameter was responsible for
How much that model parameter effects changes in the error
3
How do we think of an error surface?
F(model parameters) = model error
F(hyper-parameters) = Model parameters
F(model error) = model parameters
F(hyper-parameters) = Model best fit
4
We calculate entropy using the _____ function
logarithm (log)
Compound Interest
Taylor Series
polynomial
5
Stochastic gradient descent
Makes a change to model parameters after each training example
Makes a change to model parameters after all training examples
Makes a change to model parameters after a subset of training examples
Does not make changes to model parameters
6
The derivative of the model error is different from the learning rate because...
Derivative is magnitude, learning rate is direction
There is no difference between the two
derivative assigns how much change, learning rate assigns responsibility
derivative assigns responsibility, learning rate assigns how much change
7
A machine learning parameter is...
Something that doesn't change during training
Something the engineer may set, but the model changes during training
Something engineer sets, but model doesn't change during training
Something engineer sets, but models changes during training
8
The prior distribution is....
not a part of Bayesian models
How likely something is given the evidence
How likely the hypothesis would occur in the 1st place given what you know
The ratio of evidence to hypothesis values
9
What part of the ML model changes BASED ON THE MODEL PARAMETERS?
Your training data (unless you collect some later)
The model hyperparameters
The model error
The model parameters
10
A Random Forest classifier
Is multiple bayesian models working together
Always trains on 100% of the training data for each model
is multiple decision trees working together
is an ensemble classifier
11
What part of the ML model is changed by the ENGINEER?
Your training data (unless you collect some later)
The model hyperparameters
The model error
The model parameters
12
A Decision Tree chooses a split based on
Maximizing entropy
Maximizing information gain
Minimizing entropy
Minimizing information gain
13
What part of the ML model is changed by the TRAINING ALGORITHM, not the engineer?
The model parameters
Your training data (unless you collect some later)
The model error
The model hyperparameters
14
Support Vector Machines
Find a hyperplane that separates the data into classes
Are used primarily to replace neural networks
Use hydroplaning techniques
Are difficult to use with multivariate data
15
Bagging Models
Use gradient descent in most implementation
Are NOT ensemble methods
Use Bag's Theorem to train models
Train subsets of the data on multiple models
16
A Naiive Bayes classifier in sci-kit learn
Always estimates the prior from engineer input as a hyperparameter
Estimates the prior probability from the data distribution
Cannot be implemented with continuous values
Doesn't not estimate the prior distribution
17
Linear Regression: y=mx + b. What are parameters?
x
m
y
b
18
What part of the machine learning model DOES NOT GET CHANGED?
Your training data (unless you collect some later)
The model hyperparameters
The model parameters
The model error
19
A Decision Tree
Does not use the gini index or information gain properties
Is based off Bayesian statistics
Cannot be used in ensemble classifiers
Splits the features based on information theory
20
Entropy is...
Certainty
Information Gain
Uncertainty
Surprise
21
Height is a ______ variable, favorite band is a ______ variable
continuous, categorical
categorical, discrete
continuous, continous
categorical, continuous
22
A machine learning hyperparameter is...
Something engineer sets, but model doesn't change during training
Something engineer sets, but models changes during training
Something the engineer may set, but the model changes during training
Something that doesn't change during training
23
The learning rate
Sets how much change to make to a model parameter
Is a hyperparmeter of model training
Assigns responsibility to a model parameter
Describes the direction a model parameter should move
24
Boosting Models
Iteratively train models on larger and larger subsets of the data
Cannot use ensemble techniques
Iteratively train models to fit cases initial models did not fit well
Only are implemented with AdaBoost
25
SVM stands for
Support Vector Machine
Standard Vector Model
Support Vile Monsters
So Very Machine
26
The goal of model training is to find the best _________ to produce the lowest _______
model hyper-parameters, model hyper-parameters
model error, model parameters
model type, model parameters
model parameters, model error
27
Batch gradient descent
Makes a change to model parameters after a subset of training examples
Makes a change to model parameters after all training examples
Makes a change to model parameters after each training example
Does not make changes to model parameters
28
We want model error to be _______.
transmorgified
minimized
quantized
maximized
29
Support Vector Machines
Use the popcorn trick
Use gradient boosting inherently
Add dimensions to data
Reduce data dimensions

ML Basics and Tabular Models Review Flashcards | StudBud