[ Py DS ] Ch5 - Machine Learning (Part4)

Feature Engineering
The previous sections outline the fundamental ideas of machine learning, but all of the examples assume that you have numerical data in a tidy, [n_samples, n_features] format. In the real world, data rarely comes in such a form. With this in mind, one of the more important steps in using machine learning in practice is feature engineering— that is, taking whatever information you have about your problem and turning it into numbers that you can use to build your feature matrix.

In this section, we will cover a few common examples of feature engineering tasks: features for representing categorical data, features for representing text, and features for representing images. Additionally, we will discuss derived features for increasing model complexity and imputation of missing data. Often this process is known as vectorization, as it involves converting arbitrary data into well-behaved vectors.

Categorical Features
One common type of non-numerical data is categorical data. For example, imagine you are exploring some data on housing prices, and along with numerical features like “price” and “rooms,” you also have “neighborhood” information. For example, your data might look something like this:

view plaincopy to clipboardprint?

1. data = [  
2.     {'price': 850000, 'rooms': 4, 'neighborhood': 'Queen Anne'},  
3.     {'price': 700000, 'rooms': 3, 'neighborhood': 'Fremont'},  
4.     {'price': 650000, 'rooms': 3, 'neighborhood': 'Wallingford'},  
5.     {'price': 600000, 'rooms': 2, 'neighborhood': 'Fremont'}  
6. ]  

You might be tempted to encode this data with a straightforward numerical mapping:

view plaincopy to clipboardprint?

1. {'Queen Anne': 1, 'Fremont': 2, 'Wallingford': 3};  

It turns out that this is not generally a useful approach in Scikit-Learn: the package’s models make the fundamental assumption that numerical features reflect algebraic quantities. Thus such a mapping would imply, for example, that Queen Anne < Fremont < Wallingford, or even that Wallingford - Queen Anne = Fremont, which (niche demographic jokes aside) does not make much sense.

In this case, one proven technique is to use one-hot encoding, which effectively creates extra columns indicating the presence or absence of a category with a value of 1 or 0, respectively. When your data comes as a list of dictionaries, Scikit-Learn’s DictVectorizer will do this for you:

view plaincopy to clipboardprint?

1. from sklearn.feature_extraction import DictVectorizer  
2.   
3. data = [  
4.     {'price': 850000, 'rooms': 4, 'neighborhood': 'Queen Anne'},  
5.     {'price': 700000, 'rooms': 3, 'neighborhood': 'Fremont'},  
6.     {'price': 650000, 'rooms': 3, 'neighborhood': 'Wallingford'},  
7.     {'price': 600000, 'rooms': 2, 'neighborhood': 'Fremont'}  
8. ]  
9.   
10. vec = DictVectorizer(sparse=False, dtype=int)  
11. vec.fit_transform(data)  

Output:

view plaincopy to clipboardprint?

1. array([[     0,      1,      0, 850000,      4],  
2.        [     1,      0,      0, 700000,      3],  
3.        [     0,      0,      1, 650000,      3],  
4.        [     1,      0,      0, 600000,      2]], dtype=int32)  

Notice that the neighborhood column has been expanded into three separate columns, representing the three neighborhood labels, and that each row has a 1 in the column associated with its neighborhood. With these categorical features thus encoded, you can proceed as normal with fitting a Scikit-Learn model.

To see the meaning of each column, you can inspect the feature names:

view plaincopy to clipboardprint?

1. vec.get_feature_names()  

Output:

view plaincopy to clipboardprint?

1. ['neighborhood=Fremont',  
2. 'neighborhood=Queen Anne',  
3. 'neighborhood=Wallingford',  
4. 'price',  
5. 'rooms']  

There is one clear disadvantage of this approach: if your category has many possible values, this can greatly increase the size of your dataset. However, because the encoded data contains mostly zeros, a sparse output can be a very efficient solution:

view plaincopy to clipboardprint?

1. vec = DictVectorizer(sparse=True, dtype=int)  
2. vec.fit_transform(data)  

Many (though not yet all) of the Scikit-Learn estimators accept such sparse inputs when fitting and evaluating models. sklearn.preprocessing.OneHotEncoder and sklearn.feature_extraction.FeatureHasher are two additional tools that Scikit-Learn includes to support this type of encoding.

Text Features
Another common need in feature engineering is to convert text to a set of representative numerical values. For example, most automatic mining of social media data relies on some form of encoding the text as numbers. One of the simplest methods of encoding data is by word counts: you take each snippet of text, count the occurrences of each word within it, and put the results in a table.

For example, consider the following set of three phrases:

view plaincopy to clipboardprint?

1. sample = ['problem of evil',  
2.           'evil queen',  
3.           'horizon problem']  

For a vectorization of this data based on word count, we could construct a column representing the word “problem,” the word “evil,” the word “horizon,” and so on. While doing this by hand would be possible, we can avoid the tedium by using Scikit-Learn’s CountVectorizer:

view plaincopy to clipboardprint?

1. from sklearn.feature_extraction.text import CountVectorizer  
2.   
3. vec = CountVectorizer()  
4. X = vec.fit_transform(sample)  

The result is a sparse matrix recording the number of times each word appears; it is easier to inspect if we convert this to a DataFrame with labeled columns:

view plaincopy to clipboardprint?

1. import pandas as pd  
2. pd.DataFrame(X.toarray(), columns=vec.get_feature_names())  

There are some issues with this approach, however: the raw word counts lead to features that put too much weight on words that appear very frequently, and this can be suboptimal in some classification algorithms. One approach to fix this is known as term frequency–inverse document frequency (TF–IDF), which weights the word counts by a measure of how often they appear in the documents. The syntax for computing these features is similar to the previous example:

view plaincopy to clipboardprint?

1. from sklearn.feature_extraction.text import TfidfVectorizer  
2.   
3. vec = TfidfVectorizer()  
4. X = vec.fit_transform(sample)  
5. pd.DataFrame(X.toarray(), columns=vec.get_feature_names())  

For an example of using TF–IDF in a classification problem, see “In Depth: Naive Bayes Classification” .

Image Features
Another common need is to suitably encode images for machine learning analysis. The simplest approach is what we used for the digits data in “Introducing Scikit-Learn” on page 343: simply using the pixel values themselves. But depending on the application, such approaches may not be optimal.

A comprehensive summary of feature extraction techniques for images is well beyondthe scope of this section, but you can find excellent implementations of many of the standard approaches in the Scikit-Image project. For one example of using Scikit-Learn and Scikit-Image together, see “Application: A Face Detection Pipeline”.

Derived Features
Another useful type of feature is one that is mathematically derived from some input features. We saw an example of this in “Hyperparameters and Model Validation” when we constructed polynomial features from our input data. We saw that we could convert a linear regression into a polynomial regression not by changing the model, but by transforming the input! This is sometimes known as basis function regression, and is explored further in “In Depth: Linear Regression”.

For example, this data clearly cannot be well described by a straight line (Figure 5-35):

view plaincopy to clipboardprint?

1. %matplotlib inline  
2. import numpy as np  
3. import matplotlib.pyplot as plt  
4. x = np.array([1, 2, 3, 4, 5])  
5. y = np.array([4, 2, 1, 3, 7])  
6. plt.scatter(x, y);  

Figure 5-35. Data that is not well described by a straight line

Still, we can fit a line to the data using LinearRegression and get the optimal result (Figure 5-36):

view plaincopy to clipboardprint?

1. from sklearn.linear_model import LinearRegression  
2. X = x[:, np.newaxis]  
3. model = LinearRegression().fit(X, y)  
4. yfit = model.predict(X)  
5. plt.scatter(x, y)  
6. plt.plot(x, yfit);  

Figure 5-36. A poor straight-line fit

It’s clear that we need a more sophisticated model to describe the relationship between x and y. We can do this by transforming the data, adding extra columns of features to drive more flexibility in the model. For example, we can add polynomial features to the data this way:

view plaincopy to clipboardprint?

1. from sklearn.preprocessing import PolynomialFeatures  
2.   
3. poly = PolynomialFeatures(degree=3, include_bias=False)  
4. X2 = poly.fit_transform(X)  
5. print(X2)  

The derived feature matrix has one column representing x, and a second column representing x^2, and a third column representing x^3. Computing a linear regression on this expanded input gives a much closer fit to our data (Figure 5-37):

view plaincopy to clipboardprint?

1. model = LinearRegression().fit(X2, y)  
2. yfit = model.predict(X2)  
3. plt.scatter(x, y)  
4. plt.plot(x, yfit);  

Figure 5-37. A linear fit to polynomial features derived from the data

This idea of improving a model not by changing the model, but by transforming the inputs, is fundamental to many of the more powerful machine learning methods. We explore this idea further in “In Depth: Linear Regression” in the context of basis function regression. More generally, this is one motivational path to the powerful set of techniques known as kernel methods, which we will explore in “In-Depth: Support Vector Machines” on page 405.

Imputation of Missing Data
Another common need in feature engineering is handling missing data. We discussed the handling of missing data in DataFrames in “Handling Missing Data” , and saw that often the NaN value is used to mark missing values. For example, we might have a dataset that looks like this:

view plaincopy to clipboardprint?

1. from numpy import nan  
2. X = np.array([  
3.     [ nan, 0, 3 ],  
4.     [ 3, 7, 9 ],  
5.     [ 3, 5, 2 ],  
6.     [ 4, nan, 6 ],  
7.     [ 8, 8, 1 ]])  
8.   
9. y = np.array([14, 16, -1, 8, -5])  

When applying a typical machine learning model to such data, we will need to first replace such missing data with some appropriate fill value. This is known as imputation of missing values, and strategies range from simple (e.g., replacing missing values with the mean of the column) to sophisticated (e.g., using matrix completion or a robust model to handle such data).

The sophisticated approaches tend to be very application-specific, and we won’t dive into them here. For a baseline imputation approach, using the mean, median, or most frequent value, Scikit-Learn provides the Impute package:

view plaincopy to clipboardprint?

1. from sklearn.impute import SimpleImputer  
2.   
3. imp = SimpleImputer(strategy='mean')  
4. X2 = imp.fit_transform(X)  
5. print(X2)  

Output:

view plaincopy to clipboardprint?

1. array([[4.5, 0. , 3. ],  
2.        [3. , 7. , 9. ],  
3.        [3. , 5. , 2. ],  
4.        [4. , 5. , 6. ],  
5.        [8. , 8. , 1. ]])  

We see that in the resulting data, the two missing values have been replaced with the mean of the remaining values in the column. This imputed data can then be fed directly into, for example, a LinearRegression estimator:

view plaincopy to clipboardprint?

1. model = LinearRegression().fit(X2, y)  
2. model.predict(X2)  

Feature Pipelines
With any of the preceding examples, it can quickly become tedious to do the transformations by hand, especially if you wish to string together multiple steps. For example, we might want a processing pipeline that looks something like this:

1. Impute missing values using the mean
2. Transform features to quadratic
3. Fit a linear regression

To streamline this type of processing pipeline, Scikit-Learn provides a pipeline package, which can be used as follows:

view plaincopy to clipboardprint?

1. from sklearn.pipeline import make_pipeline  
2.   
3. model = make_pipeline(SimpleImputer(strategy='mean'),  
4.                       PolynomialFeatures(degree=2),  
5.                       LinearRegression())  

This pipeline looks and acts like a standard Scikit-Learn object, and will apply all the specified steps to any input data.

view plaincopy to clipboardprint?

1. model.fit(X, y) # X with missing values, from above  
2.   
3. print(y)  
4. print(model.predict(X))  

Output:

[14 16 -1 8 -5]
[14. 16. -1. 8. -5.]

All the steps of the model are applied automatically. Notice that for the simplicity of this demonstration, we’ve applied the model to the data it was trained on; this is why it was able to perfectly predict the result (refer back to “Hyperparameters and Model Validation” for further discussion of this).

Supplement
* [ Py DS ] Ch5 - Machine Learning (Part5)
* Medium - A Simple Guide to Scikit-learn Pipelines

Learn how to use pipelines in a scikit-learn machine learning workflow...

This message was edited 42 times. Last update was at 22/10/2019 12:31:22