Exercise: Spam or Not
Write three functions that split Inbox Guard's 1,000 emails into training and test emails with the same share of spam in both, train a LogisticRegression on the training emails, and turn its chances of spam into labels at a threshold, and let pytest check the split, the model and the labels.
Journeyman · AI Basics · Python · Updated
Inbox Guard wants a spam filter, and 1,000 emails that people already marked as spam or not are the examples to learn from. You'll write the three functions behind a first filter: one that splits the emails into emails to learn from and emails to test on, with the same share of spam in both, one that trains a logistic regression, and one that turns its chances of spam into labels at a line you choose. Tests call each function and check what it gives back, and a notebook comes with the exercise for looking at the data while you work.
Get the exercise code
Open this lesson's GitHub repository, github.com/ask-codey/exercise-ml-spam-or-not, also linked from the Exercise box beside the lesson. Clone it into your python-course folder. In VS Code's terminal:
cd ~/python-course
git clone https://github.com/ask-codey/exercise-ml-spam-or-not.git
Git makes a folder, exercise-ml-spam-or-not, with the exercise code inside. Open that folder in VS Code (File > Open Folder), open its terminal with Terminal > New Terminal, and make the exercise's virtual environment. On Windows (PowerShell):
python -m venv .venv
.venv\Scripts\Activate.ps1
pip install -r requirements.txt
On a Mac (zsh):
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
The prompt starts with (.venv) once the virtual environment is on. The exercise's README says what each command does, and what to do if PowerShell won't run the second one.
The folder holds:
emails.csv: the 1,000 emails, our own made-up data, the same file the chapter's lessons use. 299 of them are spam.seed_data.py: the script that made the CSV, with the pattern it planted. You don't need to run it: the committed CSV is what keeps everyone's numbers the same.spam_model.py: the three functions, for you to finish, withFEATURES(the columns the model looks at:num_links,num_exclamations,mentions_free,sender_in_contactsandsubject_all_caps) andLABEL(is_spam), ready to use.explore.ipynb: a notebook for looking at the emails and trying your code. Choose the.venvkernel with Select Kernel (top right) before you run it. The tests don't read it.report.py: splits, trains and labels with your three functions and prints what each step gave. Ready to use.tests/: the tests, which call each function and check what comes back, andconftest.py, which gives each test a fresh copy of the emails. You don't need to change them.requirements.txtandpytest.ini: the packages the exercise needs, at exact versions, and where pytest finds the tests and your code.
What to do
Every function in spam_model.py is already declared with its name and parameters, and for now it stops with a NotImplementedError that says Not built yet. The docstring under each one says what it should do, with examples:
def labels_at(probabilities, threshold):
"""Turn chances of spam into labels at a threshold, and return them as a list of whole numbers.
probabilities is a list (or a NumPy array) of chances of spam, each from 0 to 1,
such as column 1 of a model's predict_proba. An email whose chance is at or
above the threshold is spam (1); below it, not spam (0). A chance exactly on
the line is spam.
labels_at([0.1, 0.75, 0.5, 0.4], 0.5) -> [0, 1, 1, 0]
labels_at([0.1, 0.75, 0.5, 0.4], 0.3) -> [0, 1, 1, 1]
labels_at([], 0.5) -> []
"""
raise NotImplementedError("Not built yet.")
In each function, replace the raise NotImplementedError("Not built yet.") line with your own code:
split_emails(emails, seed)splits the emails into four parts withtrain_test_split: theFEATUREScolumns asX, theLABELcolumn asy, a fifth of the emails held back for testing,seedpassed on asrandom_state, and the split stratified on the label. It returnsX_train, X_test, y_train, y_test. With seed 42: 800 training emails holding 239 spam, 200 test emails holding 60 spam, and no email in both.train_spam_model(X, y)makes aLogisticRegressionwith its default settings, fits it toXandy, and returns the fitted model. Itsclasses_are[0, 1].labels_at(probabilities, threshold)returns a list with a 1 for each chance at or abovethresholdand a 0 for each one below it:[0, 1, 1, 0]for[0.1, 0.75, 0.5, 0.4]at 0.5, and[]for no chances at all.
The split comes first, and train_spam_model is only ever given the training emails: that's what keeps the test emails ones the model never saw.
Check your work
Run the tests from the exercise folder, with the virtual environment on:
python -m pytest
At first all 17 tests fail. That's expected: each function you finish turns its tests green. A failing test's message says what it called, what it expected and what came back, for example labels_at([0.7], 0.7) should return [1], but it returned [0]. No test checks how often the filter is right, which depends on each email's random wobble: the tests check the split, the model and the labels.
To see your functions at work on the whole file, run python report.py.
Hints
- Look before you write. Run the notebook's first cells:
emails.head(10)andemails["is_spam"].value_counts(normalize=True)show the columns and the share of spam. - One more setting. The split is the one from Splitting Data with train_test_split, plus
stratify, which Stratified Splits shows: it takes the labels whose shares both parts should keep. - Make, fit, return.
train_spam_modelis the same three lines as a linear regression's, withLogisticRegressionand no settings of your own. - On the line counts. Probabilities and the 0.5 Line compares chances with a line; here a chance equal to the threshold is spam, so the comparison is
>=. - Plain whole numbers. Append
1or0to a list as you go through the chances, so the answer is a list whateverprobabilitieswas.
When you're done
Every test is green: pytest's last line says 17 passed. And python report.py prints:
1) emails: 1000
2) training emails: 800
3) test emails: 200
4) spam share, training emails: 0.299
5) spam share, test emails: 0.3
6) classes: [0, 1]
7) coefficient for num_links: 0.3
8) coefficient for num_exclamations: 0.24
9) coefficient for mentions_free: 1.49
10) coefficient for sender_in_contacts: -2.41
11) coefficient for subject_all_caps: 1.1
12) intercept: -1.5
13) first five test emails, chance of spam: [0.107, 0.062, 0.025, 0.178, 0.324]
14) first five test emails, labels at 0.5: [0, 0, 0, 0, 0]
15) first five test emails, labels at 0.3: [0, 0, 0, 0, 1]
16) test emails flagged at 0.5: 45
17) test emails flagged at 0.3: 78
18) test emails that really are spam: 60
Now compare lines 7 to 12 with what seed_data.py planted: 0.35 a link, 0.30 an exclamation mark, 1.6 for "free", −2.5 for a sender in your contacts, 1.2 for an all-caps subject, and −1.75 to start. Every sign matches; which weights came out closest? Then look at lines 16 to 18: the line at 0.5 flags 45 emails and the line at 0.3 flags 78, with 60 real spam among the 200. Which line would you choose for a spam folder people rarely check?
Submit your exercise
The exercise is how you finish this lesson. When your work is done, put it on GitHub in a repo of your own, then submit that repo's address here. Hand In Your Work with Git walks through each step.
On GitHub, click +, then New repository. Name it (for example
exercise-ml-spam-or-not), choose Public, and click Create repository. Leave out the README: your folder already has one.In VS Code's terminal, in your exercise folder, send your work to your new repo. Point your copy at your own repo, then commit and push:
git remote set-url origin https://github.com/your-name/exercise-ml-spam-or-not.git git add . git commit -m "My solution" git push -u origin mainPaste your repo's address, such as
https://github.com/your-name/exercise-ml-spam-or-not, into Submit your exercise on this page. The site checks that the repo is public and holds your commits, and then the lesson is complete.
Changed something after you submitted? Push it, and submit again: your latest submission is the one that counts.
So what
A split that keeps the classes' shares, a classifier trained on one side of it, and a line you choose for its probabilities: that's the honest skeleton of every classifier in this course, and each function is small enough to test.