jnavarro86 / mltext

Machine Learning with Text in Python - Assessments solutions

Machine Learning with Text in Python

This repo contains the solutions for the assessments proposed in the following course -

• Course information and FAQs
• Instructor: Kevin Markham
• Teaching Assistant: Alex Egorenkov

Course Schedule

• Before the Course
• Week 1: Working with Text Data in scikit-learn
• Week 2: Basic Natural Language Processing (NLP)
• Week 3: Intermediate NLP and Basic Regular Expressions
• Week 4: Intermediate Regular Expressions
• Week 5: Working a Text-Based Data Science Problem
• Week 6: Advanced Machine Learning Techniques
• After the Course

---

Before the Course

• Make sure that scikit-learn, pandas, and matplotlib (and their dependencies) are installed on your system. The easiest way to accomplish this is by downloading the Anaconda distribution of Python. Both Python 2 and 3 are welcome.
• If you are not familiar with Git and GitHub, watch my quick introduction to Git and GitHub (8 minutes). Note that the repository shown in the video is from a previous iteration of the course, and the GitHub interface has also changed slightly.
  • For a longer introduction to Git and GitHub, watch my 11-video series (36 minutes).
• If you are not familiar with the Jupyter notebook, watch my introductory video (8 minute segment). Note that the Jupyter notebook was previously called the "IPython notebook", and the interface has also changed slightly. (Here is the notebook shown in the video.)
• If you are not yet comfortable with scikit-learn, review the notebooks and/or videos from my scikit-learn video series, focusing specifically on the following topics:
  • Machine learning terminology, and working with data in scikit-learn (video 3, notebook)
  • scikit-learn's 4-step modeling pattern (video 4, notebook)
  • Train/test split (video 5, notebook)
  • Accuracy, confusion matrix, and AUC (video 9, notebook)
• If you are not yet comfortable with pandas, review the notebook and/or videos from my pandas video series. Alternatively, review another one of my recommended pandas resources.

---

Week 1: Working with Text Data in scikit-learn

Topics covered:

• Model building in scikit-learn (refresher)
• Representing text as numerical data
• Reading the SMS data
• Vectorizing the SMS data
• Building a Naive Bayes model
• Comparing Naive Bayes with logistic regression
• Calculating the "spamminess" of each token
• Creating a DataFrame from individual text files

---

Week 2: Basic Natural Language Processing (NLP)

Topics covered:

• What is NLP?
• Reading in the Yelp reviews corpus
• Tokenizing the text
• Comparing the accuracy of different approaches
• Removing frequent terms (stop words)
• Removing infrequent terms
• Handling Unicode errors

---

Week 3: Intermediate NLP and Basic Regular Expressions

Topics covered:

• Intermediate NLP:
  • Reading in the Yelp reviews corpus
  • Term Frequency-Inverse Document Frequency (TF-IDF)
  • Using TF-IDF to summarize a Yelp review
  • Sentiment analysis using TextBlob
• Basic Regular Expressions:
  • Why learn regular expressions?
  • Rules for searching
  • Metacharacters
  • Quantifiers
  • Using regular expressions in Python
  • Match groups
  • Character classes
  • Finding multiple matches

---

Week 4: Intermediate Regular Expressions

Topics covered:

• Week 3 homework review
• Greedy or lazy quantifiers
• Alternatives
• Substitution
• Anchors
• Option flags
• Assorted functionality

---

Week 5: Working a Text-Based Data Science Problem

Topics covered:

• Reading in and exploring the data
• Feature engineering
• Model evaluation using train_test_split and cross_val_score
• Making predictions for new data
• Searching for optimal tuning parameters using GridSearchCV
• Extracting features from text using CountVectorizer
• Chaining steps into a Pipeline

---

Week 6: Advanced Machine Learning Techniques

Topics covered:

• Reading in the Kaggle data and adding features
• Using a Pipeline for proper cross-validation
• Combining GridSearchCV with Pipeline
• Efficiently searching for tuning parameters using RandomizedSearchCV
• Adding features to a document-term matrix (using SciPy)
• Adding features to a document-term matrix (using FeatureUnion)
• Ensembling models
• Locating groups of similar cuisines
• Model stacking

---