Machine Learning I
In Simple Terms, Define Classification
A supervised learning technique that teaches a computer model to assign a categorical label or class to new, unseen data based on patterns learned from labeled training data.
Explain two differences between a Bar Chart & a Histogram
Data Type:
Bar Chart displays categorical or discrete groups (e.g., Cities, Departments).
Histogram displays continuous numerical distributions (e.g., Age, Income ranges).
-------------------------------
Gaps Between Bars:
Bar Chart has visible spaces between bars to indicate distinct categories.
Histogram bars are touching (no spaces) to show a continuous range.
Who was the whale the Straw Hats met at the Grand Line?
Laboon
In Simple Terms, Define Linear Regression
Linear Regression is used to draw a straight / linear line through data points to estimate how one variable changes when another variable changes.
List Five Data Cleaning Steps
1. Handling Missing Values
2. Removing Duplicates
3. Correcting Data Types
4. Handling Outliers
5. Standardizing Text & Time Features
6. Validating the data / values
An ensemble learning technique designed to improve model accuracy by correcting the mistakes of previous models
Boosting
Chart used to visualize the relationship between two numerical variables
Scatter Plot
Who taught Naruto the Rasengan?
Master Jiraiya
What risk do you face if you evaluate model performance on training data alone?
Evaluating a model on training data alone creates a major risk of overfitting.
The model may memorize the specific patterns of the training data instead of learning general underlying rules. The performance metrics on the training set will look unrealistically high but the model will perform poorly when exposed to new, unseen real-world data (poor generalization).
You are defining a function where you're doing some kind of division. But when you run the code, you see a ZeroDivision Error. What changes will you make in the code so this error is not raised and your code runs successfully?
Add conditional logic (an if/else statement) or input validation to check whether the denominator is zero before attempting division. Alternatively, use exception handling (try/except).
Define stacking (An ensemble technique)
An ensemble machine learning technique that combines the predictions of multiple base models to produce a final, more accurate prediction
Combine multiple interactive charts and graphs into a single real-time viewing interface
Dashboards
A bird that cannot fly but can swim
Penguin
Allows you to selectively apply different preprocessing transformations to different columns of a dataset within a single, unified step
Column Transformer
The Difference Between iloc and loc
loc : Selects rows using row labels.
iloc : Selects rows using the integer positions
DOUBLE REWARD
What is Curse of Dimensionality and How does it impact the model performance?
The Curse of Dimensionality occurs when a dataset has too many columns (features).
Causes Severe Overfitting.
Fails with Distance-Based Models like KNN.
Slower Computation.
What is Data-Ink Ratio?
A principle that states to remove unnecessary visual distractions to display core elements needed to show the data.
The Hardest Natural Substance on Earth
Diamond
Why is it important to scale values in the model building process?
Feature scaling ensures that all numerical inputs contribute equally to model training regardless of their original measurement units.
Algorithms that calculate distances (like KNN) will mistakenly treat features with larger ranges (e.g., Salary from 1,000–50,000) as far more important than features with smaller ranges (e.g., Age from 18–65).
The Difference Between pd.concat() and pd.merge()
pd.concat() (Stacking Data): Used to bind or glue DataFrames together along an axis (rows or columns). It matches records by index positions or column names rather than matching key values.
pd.merge() (Relational Join): It combines DataFrames side-by-side based on matching shared key columns (e.g., on='customer_id'), using join types such as inner, left, right, or outer.
You are using KNN to classify retail customers based on two numerical features: Age (range: 18 to 75 years) and Annual Spend (range: 500 to 50,000 BHD).
What will happen if you train the KNN model without scaling these features first?
KNN relies on calculated distance measurements (such as Euclidean distance) between data points.
Because Annual Spend operates on a scale of thousands while Age operates on a scale of tens, calculated distances will be almost entirely dominated by spend differences. The model will treat Age as practically irrelevant simply due to its smaller numerical scale.
In terms of the mean and median, how can we determine if a distribution is skewed?
With reference to the mean / median, when can we expect to get a right / left skew?
If the difference between them is small, then its closer to a normal distribution. When it's large it indicates that the chart is skewed.
BONUS:
A distribution is left-skewed if the mean is less than the median.
A distribution is right-skewed if the mean is greater than the median.
Who painted Mona Lisa?
Leonardo da Vinci
DOUBLE REWARD
Let’s say you’re building a linear regression model to predict property prices (in BHD) based on property size in square meters m2. Your model yields the following results:
Intercept: 25,000
Coefficient: 15.5
MAE: 3,200
RMSE: 4,500
How would you explain what each of these four outputs means in plain English to a non-technical business stakeholder?
1. Intercept (25,000): This is the estimated baseline / starting price of a property before factoring in its size.
2. Coefficient of 15.5: For every additional 1 m2 increase in property size, the estimated property price increases by 15.5BHD.
3. MAE: On average, our price estimates are off by 3,200 BHD per property. If a house is actually worth 100,000 BHD, our prediction typically falls between 96,800 BHD and 103,200 BHD.
4. RMSE (4,500 BHD): This is also an average error measurement, but it places extra penalty on big mistakes. Because our RMSE (4,500 BHD) is higher than our MAE (3,200 BHD), it warns us that while most predictions are close, the model occasionally makes a very large pricing error on certain properties.
What is the output of the following:
print([' '.join(['yadot', 'ylrae']+['evael', 'nac', 'uoy'])][0][::-1])
You can leave early today