Feature Engineering for Machine Learning in Python
| Course Details | ||||||||
|---|---|---|---|---|---|---|---|---|
|
||||||||
| Languages | ||||||||
|
||||||||
| Outcomes | ||||||||
|
||||||||
| Price & Terms | ||||||||
|
||||||||
| Technical | ||||||||
|
||||||||
| Specification | ||||||||
|
||||||||
Learn to encode categories, handle missing values, rescale skewed data and build text features in Python so raw data is ready for machine learning.
Before a machine learning model can be used, its data usually needs a lot of munging and feature engineering. Columns have to be cleaned, encoded and rescaled into features a model can learn from. Here you prepare a Stack Overflow developer survey and a set of historic US presidential inauguration addresses with pandas in Python. You work on categorical and continuous columns, gaps in the records, skewed distributions and outliers, and free text, with hands-on practice for preparing data for your own models.
What You Will Achieve
- Turn categorical and numeric columns in a pandas DataFrame into model-ready features
- Locate missing values and choose between deletion, constants and imputation to fill them
- Check a column’s distribution, then rescale, log-transform or trim outliers before modeling
- Convert a text corpus into word count or tf-idf features while keeping the feature count in check
Program Overview
Categorical and Numeric Features
Explores a survey dataset and its data types, and shows why those types shape the features you build with pandas.
- Selecting columns by data type
- One-hot encoding, dummy variables and uncommon categories
- Binarizing and binning numeric values
Messy Data and Statistical Assumptions
Finds and handles gaps and stray characters, checks distributions, and applies the same transformations to training and test data.
- Finding missing values and measuring sparsity
- Listwise deletion, constant fills and imputation
- Cleaning stray characters with method chaining
- Normalization, standardization, log transforms and outlier removal
Features from Text Data
Turns inauguration addresses into columnar features and weighs how much context each method keeps against how many columns it adds.
- Text cleanup and high-level text features
- Word counts and limiting features
- Tf-idf on seen and unseen data
- N-grams and the most common words
Certificate Awarded (Paid)
Job Outlook
These preparation skills are used wherever tabular or text data feeds machine learning models, including data science, analytics and machine learning engineering work.
Good for: Learners building machine learning models in Python who want guided practice preparing survey data and text before training.
Keep in mind: The practice runs on two datasets, a developer survey for the tabular columns and inauguration speeches for text, and the text chapter centers on word counts, tf-idf and n-grams.
Bottom line: A practical run through the cleanup and encoding steps that sit between raw survey or text data and a model that can learn from it.
Frequently Asked Questions
Which datasets are used in the exercises?
A Stack Overflow developer survey supports the categorical, numeric and missing-value work. Historic US presidential inauguration addresses are the corpus for the chapter on text features.
How are scaling steps handled for new data?
Lessons on scaling and transforming new data show how to apply the same transformations to training and test data. The text chapter repeats the idea for unseen documents with tf-idf.
Which outlier removal approaches are taught?
Two approaches are covered: percentage-based removal and statistical outlier removal. Both come after lessons on reading a column's distribution and judging when a transformation is not needed at all.
How does the text chapter keep feature counts manageable?
It compares methods by how much context they extract from a text and limits the features kept from word counts. It also shows how longer n-grams add context while creating more columns.
User Reviews
Write your review
You must be logged in to post a review.
Reviews can only be posted by registered learners. Log in or create a free account, then you will return here to post.
There are no reviews yet.