Notebook 8

Question 1

What are Categorical Variables?

Answer 1

Question 2

import statement need for OneHotEncoder?

Answer 2

Question 3

What is Ordinal encoding?

Answer 3

Question 4

What is one-hot and dummy encoding?

Answer 4

Question 5

How do we use apndas to apply one-hot or dummy encoding to a dataframe?

Answer 5

Question 6

How do we change this one-hot encoding to dummy encoding?

Answer 6

Question 7

How do we use scikit-learn for one-hot encoding for a pandas df? How is it different from pandas?

Answer 7

This is what is read is sparse_output = False

<Compressed Sparse Row sparse matrix of dtype ‘float64’
with 5 stored elements and shape (5, 3)>

Question 8

How do we use sckit-learn’s encoder to get dummy encoding?

Answer 8

Question 9

When do you use dummy encoding vs. one-hot encoding?

Answer 9

Question 10

Say if you have use scikit-learn to use onehot encoding how do we convert it back into a dataframe? What do we need to be mindful of?

Answer 10

Question 11

What is mixed or heterogenous data types?

Answer 11

datasets that have both nuerical and categorical features, that is mixed data types, also called multiple variables types or heterogenous data.

Question 12

how would you code to perform one-hot and dummy encoding on a mixed data set laptop_price.csv?

Answer 12

Question 13

For a mixed data set how could you code to explicity specify the categorical features for one-hot coding?

How could you separate hot the numerical and categorical columns?

Answer 13

Question 14

How is inconsistent preprocessing a common pitfall?

Answer 14

Question 15

Common pitfalls: What is Data leakage?

Answer 15

Question 16

How can we avoid data leakage when pre-processing?

Answer 16

Question 17

What can lead to preprocessing could we perform he that would lead to data leakage?

Answer 17

Question 18

How can we correct for this preprocessing error that is causing data leakage?

Answer 18

Question 19

What property do some scikit learn object have inherently? How does this end up leading to prepreocessing errors?

Answer 19

Question 20

What does the random_state prameter determine?

Answer 20

Question 21

What happens to our estimators if we pass instances to our random_state?

Answer 21

MIGHT NEED ADJUSTING

Question 22

What are CV Splitters?

Answer 22

Question 23

When do we want to pass an integer vs. random_state to an estimator?

Answer 23

Question 24

When RandomState are passed to CV splitters what occurs?

Answer 24

Question 25

General recommendation: Getting reproducible result across multiple execution?

Answer 25

Question 26

General recommendation: Robustness of cross-validation results?

Answer 26

Question 27

For the mixed dataset laptop_price, how would I create test and training data for target y for column Price (Euro) of test size =0.2? How would I pull out from the design matrix XAS the categorical and numerical columns?

Answer 27

Question 28

For our laptop_price dataset, what function can I use to pre-process our data? Generally what will it do to our different data types and how do I got this?

Answer 28

Question 29

How can we make the code more readable and slightly more sophisticated?

Answer 29

Question 30

Can you use this information then apply a test and training data for design matrix X for the laptop_price dataset?, Then after preprocess the data before we can then test it noting this is a mixed dataset?

Answer 30

Question 31

What is imputation? What does the SimpleImputer function do? Apply it to this dataset?

Answer 31

can change the 'strategy' to 'median' for for that to replace the results

Question 32

HOw does the SimpleImputer work (Scikit learn)?

Answer 32

Question 33

What are some more sophisticated scaling methods?

Answer 33

Question 34

What are both the MixMaxScaler function and MaxAbsScaler functions?

Answer 34

Scaling features to lie between a given minimum and maximum value, often between zero and one, or so that the maximum absolute value of each feature is scaled to unit size. This can be achieved using MinMaxScaler or MaxAbsScaler, respectively. This can be useful if you know that your data is bounded in some range and is not Gaussian, and for preserving zero entries in sparse data.

Question 35

What does the RobustScaler function do?

Answer 35

your data contains many outliers, scaling using the mean and variance of the data is likely to not work very well. For such data you want to use RobustScaler, which scales data using the median and interquartile range (IQR), making it less sensitive to outliers. Typical examples of such data are housing prices and income data where outliers are common.

Question 36

Example of MinMaxScaler on a dataset?

Answer 36

Question 37

Why are duplicated rows a problem and how do we deal with them?

Answer 37

Question 38

What are the import statmenets for all our scoring and performance measures including all their prequisites that would probably be needed in Machine Learning modelling?

Answer 38

Add what which corresponds to which

Question 39

for the label vector on 'Outcome' for the diabetes dataset: Load the dataset, train_test_split the data and scale the data?

Answer 39

Ho

Question 40

How do we find the best hyperparameter for the SVM CLassifierfor this dataset?

Answer 40

Question 41

Once my SVM Classifier has been created, and the best parameters are found, how do we comput our performance measures?

Answer 41

check accuracy by head and verify that the precision and recall for the 1class(positive class) are the values printed here

Question 42

What is an important message to keep note about our performmance measures?

Answer 42

Question 43

How do we greate a hyperparameter grid search for a logistic regression?

Answer 43

Question 44

How do we test the performance of this regression?

Answer 44

Question 45

How do we generate ROC Curve and AUC?

Answer 45

What does this mean though?

Question 46

DO YOU ADD THE SECTION 3.4.3 SCORING STUFF??

Question 47

How do we create a 2D grid?

Answer 47

Question 48

When creating a 2-D why do we separate the arrays? How do we evaluate a function on this grid?

Answer 48

We now have three 2D arrays: the `X` and `Y` arrays with the grid coordinates, and the array `Z` containing the value of the function at each grid point.

Question 49

Answer 49

Question 50

How do you plot a contour lines for this function on our 2-D grid?

Answer 50

Question 51

How do you plot a filled contour regions for this function on our 2-D grid?

Answer 51

LOOK UP PROPER WAY TO FIND WAY TO FIND THE CRITICAL POINTS

Question 52

How do we add more contour plots to plt.contourf and plt.contour functions?

Answer 52

use the kwarg: levels = int

Question 53

How do specify the levels of counters to plot What a bout the linestyles themselves?

Answer 53

Question 54

What does the cmap Kwarg do in our contour plots?

Answer 54

other cmaps = 'magma', 'plasma', 'Reds', 'jet' adding alpha = 0 to 1 will change the opacity with the closer the to 1 the more brighter it will be

Question 55

How do we code the decision boundaries here?

Answer 55

Question 56

How do you plot the contourf and contour on the sample graph?

Answer 56

Question 57

Answer 57

Question 58

Answer 58

Question 59

Walk me through on a high level what each line of this code is doing?

Answer 59

--- - A Python list with classifiers and hyperparameter settings is created. - Three datasets are created and put into a list. - The lists are looped over to generate the comparison grid. - Within the outer loop (dataset loop) you should recognise where the data is scaled and the train_test split is made. - There is then code to plot data. - Within the inner loop (classifier loop) you should recognise the fitting of the training data and the computation of score from the test data. - There is then code to plot the results. You would recognise the functions contourf and scatter. While more involved and sophisticated, this code more-or-less corresponds to the example in the Contour_plot notebook.

Question 60

Locate where the code creates the linearly separable dataset? Where can you set the added noise to zero? What about for when it is 20 What happens to the score when you do this?

Answer 60

X += 2 * rng.uniform(size=X.shape) The score decreases by about 0.5 when increasing the noise

Question 61

What do these lines of code generate on a graph?

Answer 61

make_moon --> creates two interleaving half circles * Shuffle --> wheather to shuffle the samples * Noise --> Standard evision of Gaussian noise added to the data make_circles -->makes a larger circle containing the smaller circle * Shuffle --> wheather to shuffle the samples * Noise --> Standard evision of Gaussian noise added to the data * Factor --> the Scale factor between inner and out circles in the range [0,1) More noise generally makes the score worse for make_circles score in low for linear.svm as data is not linearly separable even with soft margins

Question 62

For binary classification problems, what will always result in a score of about 0.5?

Answer 62

Either randomly guessing the class of the binary classification problem or settle all the test data to one of the classes . (in our case we just can set y_test = 0 * y_test)

Question 63

What is the difference between settling this to linear of RBF?

Answer 63

* Linear is the dot plot of the input samples which creates our usual hyperplane spit with straight lines * Radial Basis function (RBF) also know as the Gausian Kernel,easure the similarity between two data points in infinite dimension and then approaches classificaiton by majority vote The kernel function is: K(x₁,₂) = exp (-γ * IIx₁ - ₂II²

Question 64

What values does gamme take for RBF SVM classification?

Answer 64

Question 65

How would you update this to add a Poly SVM?

Answer 65

Setting degree =1 for the poly kernel should reproduce the linear case. Note that the circles data is approximately rotationally symmetric, so linear classifiers will have a hard time argeeing on an orientiation. As already discussed the scores are bad even for noise free circle data with linear classifiers.

Question 66

import libraries for 3D plotting?

Answer 66

Question 67

How do we plot this using 3D plotting?

Answer 67

You should see a graph of the function z=f(x,y) as a surface plot in three dimensions. Surface plots make local and global minima very obvious. You should also see faint lines on the surface showing the grid. Before discussing the Python code, it is worth changing the colouring of the surface. can add cmap='plasma' or 'viridis' or 'magma'

Question 68

Why do we need these when making £D plots?

Answer 68

Question 69

What does this function do?

Answer 69

elev = rotates the surface in the z-axis e.g. elev=90 looks at the plot for above whereas -30 is looking at it from 30 degrees below change the azim (azimuth) angle rotations the view of the surface around the z-axis

Question 70

What does adding edgecolor='k' to the ax.plot_surface function do?

Answer 70

Question 71

Plotting a scatter plot in 3D?

Answer 71

Question 72

What is changing the point size to 50*(z_point) do? What about 500*np.abs(z) WHat abou tchanging the colours to colours = y_points + 10? what about to -y_points?

Answer 72

500* np.ab(z) just makes the scatter plot larger