Help
Options
Didn't know it?
click below
click below
Knew it?
click below
click below
Don't Know
Remaining cards (0)
Know
retry
shuffle
restart
0:00
Embed Code - If you would like this activity on your web page, copy the script below and paste it into your web page.
Normal Size Small Size show me how
Normal Size Small Size show me how
MIS 665 Final
| Question | Answer |
|---|---|
| the role of machine learning within the broader AI hierarchy | Machine learning is a subset of AI that learns patterns from data |
| Example of unsupervised learning | Grouping customers into segments based on purchasing behavior (no predicting) |
| R-Squared (coefficient of determination) measures: | the proportion of variance in Y explained by the model |
| Multicollinearity in regression refers to | High correlation among the predictor variables themselves |
| When applying a preprocessing pipeline to scoring (deployment) data, you should use | scaler.transform(X_score) to apply the scale learned from training data |
| The primary reason for splitting data into training and test sets is to | Evaluate how well the model generalizes to data it has not seen before |
| What does Lasso regularization do to coefficients of predictors that contribute little to the model | It shrinks them toward zero, potentially to exactly zero |
| In healthcare classification, which metric is most critical to minimize when missing a positive case (e.g., a patient a risk) carries severe consequences | False Negative Rate |
| In the lab's clustering dataset (Weight, Cholesterol, Gender), which feature dominated cluster assignments when the data was NOT standardized | Cholesterol, because it had the highest variance |
| The Elbow method for selecting k plots which quantity on the y-axis? | Within-Cluster Sum of Squares (WCSS / inertia) |
| In Lab 1 (Boston Housing), how many PCA components were needed to retain at least 90% of the variance in the 13-feature dataset? | 7 |
| The explained variance ratio of a principal component represents | the proportion of total variance in the original data captured by that component |
| In TF-IDF vectorization, each column in the resulting feature matrix represents | A unique word (term) from the vocabulary, weighted by frequency and rarity |
| Cosine similarity between two vectors measures | the angle between the two vectors, with 1.0 meaning identical direction |
| The all-MiniLM-L6-v2 embedding model used in Lab 3 produces vectors with how many dimensions | 384 |
| Why does a plain LLM fail when asked about a private company's Q3 revenue | The document is not in the model's training data, so the model either refuses or fabricates a confident-sounding answer |
| Place the six stages of a RAG pipeline in the correct roder | Load --> Chunk --> Embed --> Store --> Retrieve --> Generate |
| The embedding model (gemini-embedding-001) used in this lab returns | A 3,072-number vector that captures the meaning of the input text |
| Why does cosine similarity on embeddings reliably retrieve the right document even when the query phrase does not appear verbatim in the corpus | Because embeddings encode meaning, so synonyms and paraphrases land near each other in vector space |
| The two main types of supervised learning problems covered in Lab 1 are ____ (predicting a number) and _____ (predicting a category) | regression, classification |
| In k-Means clustering, the algorithm repeats two steps until convergence: assigning each point to the nearest ___, and then recalculating the ___ of each cluster | centroid, centroid |
| In the AI hierarchy, ____ AI extends generative AI by taking actions and autonomously completing multi-step goals | Agentic |
| Deep learning is preferred over traditional ML when working with ________ data such as images, audio, and raw text | unstructured |
| ColumnTransformer allows you to apply ______ to numeric features and ____ to categorical features within a single preprocessing step | transformers/scalers, encoders |
| A false negative in the heart attack classifier means a patient who ___ have a second heart attack was predicted ____ | did, healthy |
| Logistic Regression outputs a value between o and 1 that represents the _____ of the positive class, which can be thresholded at 0.5 to produce a class label | probability |
| The Euclidean distance from (5,5) to (0,0) as the square root of (5^2 + 5^2), equals approximately ___ | 7.07 |
| To profile the clusters and understand what each group represents, we calculate the ____ of each feature within each cluster label | mean/centroid |
| PCA components are ordered so that the first component explains the _____ amount of variance, the second explains the next most, and so on | largest |
| TF-IDF stands for Term Frequency - ___ Document Frequency | Inverse |
| Unlike TF-IDF, embedding vectors are ____ -- every dimension carries meaning and no entries are zero | dense |
| Cosine similarity measures the ____ between two vectors rather than their absolute distance, making it well-suited for comparing text embeddings of different lengths | angle |
| The six stages of every RAG pipeline are: Load, ____, Embed, Store, Retrieve, and Generate | Chunk |
| The embedding model gemini-embedding-001 produces vectors with _______ dimensions, so embedding 100 chunks yields a matrix of shape (100,____) | 3072, 3072 |
| ML model workflow | define x y variables, split validation, initialize, fit(), predict(), compare actual y with predicted y, deploy model |
| PCA is used for | numerical columns |
| Embedding is used for | text data |
| RAG stands for | Retrieval Augmented Generation |
| RAGs are used for | private data, uses embedding |
| LLM vs RAG | LLM trained on public internet data from memory, RAG uses provided knowledge base and is a smart search system |
| What are the 3 main types of ML problems? | Regression (numerical output), Classification (categorical output), Clustering (no labeled output). |
| What is regression? | A supervised learning method that predicts continuous numerical values. |
| What is classification? | A supervised learning method that predicts categories or labels. |
| What is clustering? | An unsupervised learning method that groups similar data points without labels. |
| What is the relationship between AI, ML, and DL? | AI ⟶ ML ⟶ DL (DL is a subset of ML, ML is a subset of AI). |
| What is ML best at? | Structured/tabular data. |
| What is Deep Learning best at? | Unstructured data like images, text, and video. |
| What is the order of AI evolution? | ML → DL → Generative AI → RAG → Agentic AI |
| What is the goal of regression? | Minimize sum of squared errors between actual and predicted values. |
| Why do we split validation data? | To test model performance on unseen data and prevent overfitting |
| Why do we standardize numerical columns? | To put all features on the same scale so no variable dominates. |
| What is multicollinearity? | When features are highly correlated with each other. |
| How do you handle multicollinearity? | Remove variables, use Lasso regression, or feature selection. |
| What is Lasso regression? | A regression method that penalizes large coefficients and performs feature selection |
| What is f_regression used for? | Feature selection by measuring relationship between features and target variable. |
| What is ColumnTransformer? | A tool to apply different preprocessing steps to different column types. |
| Why use ColumnTransformer? | To scale numerical data and encode categorical data correctly in one pipeline. |
| What is a confusion matrix? | A table showing TP, TN, FP, FN results of a classification model. |
| ML workflow steps? | 1) Initialize model → 2) Fit → 3) Predict |
| fit() vs transform()? | fit learns patterns; transform applies learned transformation. |
| What is fit_transform()? | Fits data and then immediately transforms it. |
| What is a pipeline in ML? | A sequence of preprocessing + modeling steps combined into one workflow. |
| Name common classification models. | Decision Tree, KNN, Logistic Regression, Random Forest. |
| What is the goal of clustering? | To group similar data points for profiling or pattern discovery. |
| What is Euclidean distance? | The straight-line distance between two points. |
| What is the elbow method? | A technique to choose the optimal number of clusters (K). |
| What is PCA (Principal Component Analysis)? | A method that reduces features while keeping most variance |
| Why standardize before PCA? | Because PCA is sensitive to scale. |
| Why is TF-IDF a sparse matrix? | Because most word positions are zero (most words don’t appear). |
| What is chunking in RAG? | Splitting documents into smaller sections (like paragraphs). |
| What does retrieval do in RAG? | Finds the most relevant chunks (top-k) based on similarity. |
| What does the LLM do in RAG? | Generates a final response using retrieved information. |
| **Phase 1** | Data Capture |
| **Phase 2** | Data Preparation & Transformation |
| **Phase 3** | Descriptive Analytics / Predictive Analytics |
| ML uses unstructured/structured data? | Structured |
| DL uses unstructures/structured data? | Unstructured |
Created by:
lexi.welte
Popular Business sets