IBM Data Science Professional Certificate flashcards

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

CRISP-DM

A six-phase process model for data science projects: business understanding, data understanding, data preparation, modeling, evaluation, and deployment. Phases are iterative, and projects often loop back after evaluation.

Type I vs Type II error

Type I error rejects a true null hypothesis and occurs with probability alpha. Type II error fails to reject a false null hypothesis and occurs with probability beta, so statistical power equals one minus beta.

Nominal vs ordinal data

Nominal data are unordered categories such as country or color. Ordinal data are categories with a meaningful order, such as satisfaction levels, but the distances between levels need not be equal.

Bias-variance tradeoff

Model error decomposes into squared bias, variance, and irreducible noise. Simple models have high bias and underfit; complex models have high variance and overfit training noise, harming generalization.

List vs tuple vs set

A list is ordered and mutable, a tuple is ordered and immutable, and a set is unordered with unique elements. Choose tuples for fixed records and sets for membership tests and deduplication.

Python slice semantics

The expression seq[start:stop:step] includes start and excludes stop. Negative indices count from the end, and a step of minus one reverses the sequence. Slicing a list returns a new shallow copy.

Mutable default argument pitfall

Default argument values are evaluated once when the function is defined, not on each call. A default list or dict therefore persists state across calls; use None as the default and build inside.

List comprehension syntax

A list comprehension [expr for item in iterable if condition] builds a list in one expression. Replacing brackets with parentheses yields a lazy generator that saves memory on large inputs.

WHERE vs HAVING

WHERE filters individual rows before grouping and cannot reference aggregate functions. HAVING filters groups after GROUP BY and is where aggregate conditions such as COUNT(*) > 5 belong.

LEFT JOIN behavior

A LEFT JOIN returns every row from the left table and matching rows from the right. Unmatched right-side columns are filled with NULL, so downstream filters must handle those nulls explicitly.

SQL NULL comparison

NULL represents an unknown value, so comparisons like col = NULL evaluate to unknown rather than true. Use IS NULL or IS NOT NULL, and COALESCE to substitute a default value.

Window function PARTITION BY

A window function computes across related rows without collapsing them. PARTITION BY defines the groups and ORDER BY defines sequence, enabling running totals and ranks alongside detail rows.

NumPy broadcasting

Broadcasting lets NumPy operate on arrays of different shapes by stretching size-one dimensions. A column vector of shape (3,1) combined with a row of shape (1,4) produces a (3,4) result.

loc vs iloc

In pandas, .loc selects rows and columns by label and includes the stop label in slices. .iloc selects by integer position and excludes the stop position, matching standard Python slicing.

pandas groupby transform

groupby().transform() returns a result aligned to the original row index rather than one row per group. It is used to add group means or z-scores back as new columns for comparison.

IQR outlier rule

Compute the interquartile range as Q3 minus Q1. Values below Q1 minus 1.5 times the IQR, or above Q3 plus 1.5 times the IQR, are commonly flagged as outliers in box plots.

Pivot vs melt

pivot_table reshapes long data into a wide summary grid with aggregation. melt does the reverse, turning wide columns into identifier, variable, and value columns suitable for plotting or modeling.

Bar chart zero baseline

Bars encode value by length, so the axis must start at zero or differences are visually exaggerated. Line charts encode trend by position and may use a truncated axis if clearly labeled.

Sequential vs diverging palette

A sequential palette varies lightness for ordered magnitude such as counts. A diverging palette uses two hues around a meaningful midpoint, suited to deviations like profit versus loss.

Matplotlib Figure and Axes

A Figure is the overall canvas and can contain multiple Axes, each an individual plot with its own data, limits, and labels. plt.subplots creates a figure and array of axes together.

Histogram bin width

Bin width controls a histogram's apparent shape: too few bins hide structure and too many create noise. Always report the binning choice, and compare with a box or violin plot for distribution shape.

Precision vs recall

Precision is true positives divided by predicted positives and measures how often a positive prediction is correct. Recall is true positives divided by actual positives and measures coverage of real cases.

Cross-validation purpose

k-fold cross-validation splits training data into k parts, training on k-1 and validating on the held-out fold, then averaging. It gives a lower-variance estimate of generalization than one split.

L1 vs L2 regularization

L1 regularization adds the absolute coefficient sum to the loss and can shrink coefficients exactly to zero, performing feature selection. L2 adds squared coefficients and shrinks them smoothly toward zero.

Data leakage in pipelines

Leakage occurs when information from validation or test data influences training, such as fitting a scaler on the full dataset. Fit preprocessing inside a pipeline on training folds only.

k-means assumptions

k-means partitions points into k clusters by minimizing within-cluster squared distance. It requires k in advance, is sensitive to initialization and scaling, and works best with roughly spherical clusters.

Geocoding with geopy

geopy's Nominatim geocoder converts a street address into latitude and longitude. Public services impose rate limits, so cache results and supply a descriptive user agent identifying your application.

Folium choropleth map

Folium builds interactive Leaflet maps in Python. A choropleth shades geographic areas by a data value, requiring a GeoJSON boundary file and a key linking regions to the data.

Jupyter kernel state

The kernel holds all variables and imports for the session. Cells run out of order can leave stale state, so restart and run all before trusting or sharing results.

Jupyter magic commands

Lines starting with a percent sign are magics. %matplotlib inline renders plots inside the notebook, %timeit benchmarks a statement, and the exclamation prefix runs shell commands.

Git commit and push

git add stages changes, git commit -m records a snapshot with a message, and git push uploads commits to a remote such as GitHub. Branches isolate work until a pull request merges it.

Watson Studio deployment

A trained model is stored in the Watson Machine Learning repository, deployed as an online endpoint, and scored by sending records to that endpoint with an authentication token.

SCQA narrative structure

Situation, Complication, Question, Answer is a presentation structure that states context, the problem, the decision needed, and the recommendation before supporting evidence, suited to executive audiences.