Let's show an example of classification using fruit!
How would you describe apples to a computer? How would they differ from oranges?
Remember, computers can only really understand numbers, true false values, and strings within a predefined set
Source: Andrew Rosenberg
Our fruit test shows us everything we need to do a classification machine learning test. For each item with a label (apple, orange, lemon), we use a series of values to try to capture machine-understandable information about the item. These values are a feature representation of the item in question. The features themselves, as we can see above, can be numeric, true/false values, or a string in a set of predefined strings.
Source: Andrew Rosenberg
Our fruit test is an example of a classification task. Classification allows you to predict a categorical value. This is a type of supervised machine learning, meaning we know the labels ahead of time and can give them to the machine learning algorithm so that it can be trained to knows what the categories of our data are. This way, when it comes time to give the algorithm previously unseen data, it knows which categories it's looking for.
We acknowledge that often times we are not trying to divide apples and oranges, but categories of people or personal attributes. While we are going to focus on the mechanics of machine learning, we strongly recommend that this work be paired with a grounding in ethics, such as the DHRI-Ethics workshop.
Let's get to coding!
We are going to classify two different sets of sentences from very different source material in the Brown corpus: one set of sentences from a corpus of news text, and the other set of sentences from a corpus of romance novel text.
from nltk.corpus import brownFor a list of categories in the Brown corpus, use the following code
for cat in brown.categories():
print (cat)adventure
belles_lettres
editorial
fiction
government
hobbies
humor
learned
lore
mystery
news
religion
reviews
romance
science_fiction
news_sent = brown.sents(categories=["news"])
romance_sent = brown.sents(categories=["romance"])print(news_sent[:5])
print()
print(romance_sent[:5])[['The', 'Fulton', 'County', 'Grand', 'Jury', 'said', 'Friday', 'an', 'investigation', 'of', "Atlanta's", 'recent', 'primary', 'election', 'produced', '``', 'no', 'evidence', "''", 'that', 'any', 'irregularities', 'took', 'place', '.'], ['The', 'jury', 'further', 'said', 'in', 'term-end', 'presentments', 'that', 'the', 'City', 'Executive', 'Committee', ',', 'which', 'had', 'over-all', 'charge', 'of', 'the', 'election', ',', '``', 'deserves', 'the', 'praise', 'and', 'thanks', 'of', 'the', 'City', 'of', 'Atlanta', "''", 'for', 'the', 'manner', 'in', 'which', 'the', 'election', 'was', 'conducted', '.'], ['The', 'September-October', 'term', 'jury', 'had', 'been', 'charged', 'by', 'Fulton', 'Superior', 'Court', 'Judge', 'Durwood', 'Pye', 'to', 'investigate', 'reports', 'of', 'possible', '``', 'irregularities', "''", 'in', 'the', 'hard-fought', 'primary', 'which', 'was', 'won', 'by', 'Mayor-nominate', 'Ivan', 'Allen', 'Jr.', '.'], ['``', 'Only', 'a', 'relative', 'handful', 'of', 'such', 'reports', 'was', 'received', "''", ',', 'the', 'jury', 'said', ',', '``', 'considering', 'the', 'widespread', 'interest', 'in', 'the', 'election', ',', 'the', 'number', 'of', 'voters', 'and', 'the', 'size', 'of', 'this', 'city', "''", '.'], ['The', 'jury', 'said', 'it', 'did', 'find', 'that', 'many', 'of', "Georgia's", 'registration', 'and', 'election', 'laws', '``', 'are', 'outmoded', 'or', 'inadequate', 'and', 'often', 'ambiguous', "''", '.']]
[['They', 'neither', 'liked', 'nor', 'disliked', 'the', 'Old', 'Man', '.'], ['To', 'them', 'he', 'could', 'have', 'been', 'the', 'broken', 'bell', 'in', 'the', 'church', 'tower', 'which', 'rang', 'before', 'and', 'after', 'Mass', ',', 'and', 'at', 'noon', ',', 'and', 'at', 'six', 'each', 'evening', '--', 'its', 'tone', ',', 'repetitive', ',', 'monotonous', ',', 'never', 'breaking', 'the', 'boredom', 'of', 'the', 'streets', '.'], ['The', 'Old', 'Man', 'was', 'unimportant', '.'], ['Yet', 'if', 'he', 'were', 'not', 'there', ',', 'they', 'would', 'have', 'missed', 'him', ',', 'as', 'they', 'would', 'have', 'missed', 'the', 'sounds', 'of', 'bees', 'buzzing', 'against', 'the', 'screen', 'door', 'in', 'early', 'June', ';', ';'], ['or', 'the', 'smell', 'of', 'thick', 'tomato', 'paste', '--', 'the', 'ripe', 'smell', 'that', 'was', 'both', 'sweet', 'and', 'sour', '--', 'rising', 'up', 'from', 'aluminum', 'trays', 'wrapped', 'in', 'fly-dotted', 'cheesecloth', '.']]
Each sentence is already tokenized—split into a series of word and punctuation stringes, with whitespace removed. This saves us the time of having to do all of this work ourselves!
To start to organize our data, let's put these sentences into a pandas DataFrame, an object which has a format very similar to an Excel spreadsheet. We will first make two spread sheets (one for news, and one for romance), and then combine them into one. We will also add the category each sentence came from (news or romance), which will be our labels for each sentence and its associated feature representation (which we will build ourselves).
ndf = pd.DataFrame({'sentence': news_sent,
'label':'news'})
rdf = pd.DataFrame({'sentence':romance_sent,
'label':'romance'})# combining two spreadsheets into 1
df = pd.concat([ndf, rdf])Let's see what this DataFrame looks like! Let print out the first 5 rows using .head() and the last 5 using .tail()
df.head()| label | sentence | |
|---|---|---|
| 0 | news | [The, Fulton, County, Grand, Jury, said, Frida... |
| 1 | news | [The, jury, further, said, in, term-end, prese... |
| 2 | news | [The, September-October, term, jury, had, been... |
| 3 | news | [``, Only, a, relative, handful, of, such, rep... |
| 4 | news | [The, jury, said, it, did, find, that, many, o... |
df['label'].value_counts()news 4623
romance 4431
Name: label, dtype: int64
We first create a figure and axes on which to draw our charts using plt.subplots(). Each chart is one axes, and a figure can contain multiple axes. This notation will be explained in detail when we discuss visualization. Our data is encapsulated in df['label'].value_counts(), which is itself a DataFrame. We then tell the Pandas to visualize the DataFrame as a bar chart using .plot.bar(ax=ax, rot=0). The ax keyword tells Pandas which chart in the figure to plot, and the rot keyword controls the rotation of the x axis labels.
fig, ax = plt.subplots()
_ = df['label'].value_counts().plot.bar(ax=ax, rot=0)
fig.savefig("images/categories_counts.png", bbox_inches = 'tight', pad_inches = 0)
We have slightly more news data than romance data, which we should keep in mind as we go ahead with classification.