DSMA Admission flashcards

33 cards. Read the term and the definition here. The deck at the top of this page is optional.

Data minimization

A privacy principle requiring that only personal data adequate, relevant, and necessary for a specified purpose be collected and processed, reducing risk and limiting potential harm from breaches.

Data controller

The entity that alone or jointly determines the purposes and means of processing personal data; it bears primary legal responsibility for compliance with data protection law.

Pseudonymization

Replacing direct identifiers with artificial keys so data cannot be attributed to a person without additional information stored separately; the data remains personal data under GDPR.

Breach notification deadline

Under GDPR, a controller must notify the supervisory authority of a personal data breach within 72 hours of becoming aware, unless the breach is unlikely to result in a risk.

Bayes' theorem

P(A|B) = P(B|A)P(A)/P(B). It updates a prior probability P(A) with the likelihood P(B|A) to produce the posterior probability P(A|B).

Central limit theorem

For independent, identically distributed variables with finite variance, the sampling distribution of the sample mean approaches normality as sample size increases, regardless of the population distribution.

Type I error

Rejecting a true null hypothesis; its probability is denoted alpha (α), commonly set at 0.05. It is also called a false positive.

p-value

The probability, assuming the null hypothesis is true, of obtaining a test statistic at least as extreme as the observed one. It is not the probability that the null is true.

Standard error of mean

The standard deviation of the sampling distribution of the sample mean, equal to σ/√n. It decreases as sample size n increases, reflecting greater precision.

Confidence interval

A range of values, computed from sample data, that with a specified confidence level (e.g., 95%) is expected to contain the true population parameter across repeated sampling.

SQL execution order

Logical processing order is FROM, WHERE, GROUP BY, HAVING, SELECT, ORDER BY, LIMIT. This explains why column aliases from SELECT are unavailable in WHERE.

INNER JOIN

A join that returns only rows where the join condition matches in both tables; unmatched rows from either side are excluded from the result set.

WHERE vs HAVING

WHERE filters individual rows before grouping; HAVING filters groups after GROUP BY. HAVING can reference aggregate functions, WHERE cannot.

NULL in SQL

A marker for unknown or missing data. Comparisons with NULL yield UNKNOWN, so use IS NULL or IS NOT NULL; aggregate functions except COUNT(*) ignore NULLs.

Third normal form

A relational schema is in 3NF if it is in 2NF and no non-key attribute is transitively dependent on the primary key, eliminating redundancy that causes update anomalies.

ACID properties

Atomicity, Consistency, Isolation, and Durability: guarantees that database transactions are all-or-nothing, preserve valid states, do not interfere, and persist after commit.

Window function

A SQL function such as ROW_NUMBER() or SUM() OVER (PARTITION BY ...) that computes across a set of rows related to the current row without collapsing them into a single output row.

List comprehension

A concise Python syntax [expr for item in iterable if condition] that builds a new list by evaluating an expression for each item satisfying an optional filter.

loc vs iloc

In pandas, .loc selects rows and columns by label (including boolean arrays); .iloc selects by integer position. Mixing them causes errors or unexpected results.

Pandas groupby

df.groupby(keys) splits data into groups by key values; chaining an aggregation like .agg('mean') or .sum() computes summary statistics per group and returns a combined result.

Vectorization

Performing operations on entire arrays at once, as NumPy and pandas do, rather than looping element by element in Python; this exploits compiled code and is much faster.

Mutable vs immutable

Mutable objects (lists, dicts, sets) can change after creation; immutable objects (tuples, strings, numbers) cannot. Immutability enables safe sharing and use as dictionary keys.

Random seed

A value that initializes a pseudo-random number generator, making stochastic results reproducible. In Python, set numpy.random.seed or pass random_state to scikit-learn functions.

Supervised learning

A machine learning paradigm that trains a model on labeled input-output pairs to predict outputs for new inputs, encompassing classification and regression tasks.

Bias-variance tradeoff

The tension between a model too simple to capture patterns (high bias, underfitting) and one too complex and sensitive to training noise (high variance, overfitting).

Cross-validation

A resampling technique, commonly k-fold, that partitions data into folds, trains on k-1 and validates on the remaining fold, rotating until each fold is used once.

Precision and recall

Precision = TP/(TP+FP), the fraction of predicted positives that are correct. Recall = TP/(TP+FN), the fraction of actual positives that are found. F1 is their harmonic mean.

ROC AUC

The area under the receiver operating characteristic curve, which plots true positive rate against false positive rate across thresholds. A value of 0.5 indicates random ranking; 1.0 is perfect.

Overfitting

When a model learns training-specific noise and performs well on training data but poorly on unseen data; detected by a large gap between training and validation error.

R-squared

The proportion of variance in the dependent variable explained by a regression model: R² = 1 - SS_res/SS_tot. It ranges up to 1 and can be negative for poor models.

Bar chart baseline

Bar charts encode magnitude by length, so the value axis should start at zero; truncating it exaggerates differences and misleads viewers about relative sizes.

Data-ink ratio

Edward Tufte's principle that the proportion of a graphic's ink devoted to actual data should be maximized by removing chartjunk, redundant gridlines, and decorative elements.

Sequential vs diverging palette

Sequential palettes encode ordered low-to-high values with varying lightness; diverging palettes use two hues around a meaningful midpoint to show deviation in either direction.