Overview
This module sets the baseline for the Enterprise Big Data Scientist (EBDS) course. It defines data science as the end-to-end practice of turning a business question into a model or insight, separates the scientist's job from the analyst's and the engineer's, and places predictive and prescriptive work above the descriptive reporting covered at Enterprise Big Data Analyst (EBDA) level. It then reviews the Python you will read throughout the course: pandas DataFrames, NumPy arrays, indexing and slicing rules, and the habits that make notebook code reproducible.
The last part revisits the Enterprise Big Data Framework (EBDF) itself, its six elements (big data strategy, architecture, algorithms, processes, functions and artificial intelligence), so you can place every later technique inside an organizational context. Treat it as the map for modules 2 through 9: statistics and machine learning sit in the algorithms capability, Spark in architecture, and ethics in governance and functions.
How it's tested
Expect short scenario and recognition items: classify an activity as descriptive, diagnostic, predictive or prescriptive; decide which role or EBDF capability owns a task; or read a few lines of Python and predict the output. If the exam is taken open book, the Guide helps with definitions but not with code tracing, so practice reading snippets quickly.
-
Classify the analytics type
A scenario describes what a team produces, such as a churn probability or a recommended discount, and you name the type of analytics. The verb in the deliverable (what happened, why, what will, what should) decides it.
-
Who owns this task
A task such as building an ingestion pipeline or designing an experiment is described, and you assign it to the data engineer, analyst or scientist role, or to an EBDF capability.
-
Trace the code
A short pandas or NumPy snippet is shown and you predict the shape, value or error it produces, which tests indexing, slicing and copy semantics.
A scenario says a scientist's churn model is stale because the nightly job that loads customer events keeps failing, and asks who should fix it. The failing component is ingestion, which the data engineer owns. The scientist's responsibility starts once data arrives: retraining, revalidating and monitoring the model. An option assigning the pipeline repair to the scientist sounds helpful but mixes the roles the module separates.
Topics 3
Official topics from the Enterprise Big Data Framework Alliance outline, in outline order.
1.1 Introduction to Data ScienceWhat data science is, the analytics maturity ladder, and how the scientist role differs from analyst and engineer.
- What it means
- Data science combines statistics, programming and domain knowledge to answer questions that need inference or prediction from data. It runs as an iterative process: frame the question, acquire and prepare data, explore, model, evaluate, and communicate or deploy.
- What to understand
- Know the four analytics types: descriptive (what happened), diagnostic (why), predictive (what will happen) and prescriptive (what should we do). The data engineer builds and runs the pipelines, the analyst explains existing data, and the scientist builds and validates models that generalize to new data.
- How it shows up
- A scenario lists deliverables or job tasks and asks which analytics type or role each belongs to, or which process step a team has skipped.
- Common confusion
- Calling any model output prescriptive. A risk score predicts; only an output that selects or optimizes an action is prescriptive.
1.2 Python ReviewThe pandas, NumPy and core Python behaviors that course notebooks and exam snippets rely on.
- What it means
- The course is taught mainly in Python notebooks, so you must read DataFrame and array code fluently. That includes boolean masking, groupby-aggregate, merges, vectorized arithmetic and broadcasting.
- What to understand
- Python slices exclude the stop index, but pandas .loc label slices include both ends while .iloc position slices exclude the end. NumPy basic slicing returns a view, so editing it changes the original array, whereas boolean or fancy indexing returns a copy. Vectorized operations replace explicit loops for speed.
- How it shows up
- A snippet such as df.loc[2:4] versus df.iloc[2:4] is shown and you choose how many rows come back, three versus two.
- Common confusion
- Assuming .loc and .iloc slice the same way, or that a NumPy slice is an independent copy.
1.3 EBDF ReviewThe Enterprise Big Data Framework capabilities and vocabulary carried over from the Professional and Analyst levels.
- What it means
- The Enterprise Big Data Framework (EBDF) is a vendor-neutral body of knowledge that organizes enterprise big data work into six elements: big data strategy, big data architecture, big data algorithms, big data processes, big data functions and artificial intelligence.
- What to understand
- Know which capability a problem belongs to: platform and storage choices are architecture, statistical and machine learning methods are algorithms, repeatable analysis and governance workflows are processes, and roles and organizational structures such as a center of excellence are functions. Also know the defining characteristics of big data: volume, velocity, variety and veracity.
- How it shows up
- A scenario describes an organizational problem, such as models built but never governed or reused, and asks which capability is weak.
- Common confusion
- Treating every big data problem as an architecture problem when the scenario actually describes a missing process or role.
High-yield concepts
The terms, rules and formulas this domain turns into questions most often.
-
Four types of analytics
Descriptive summarizes what happened, diagnostic explains why, predictive estimates what will happen, and prescriptive recommends or optimizes an action.
- Remember
- Classify by the question the output answers, not by how sophisticated the technique looks.
- Tested as
- Scenario items describe a deliverable and ask you to label it, which is the most direct way this module is tested.
- Trap
- Labeling a machine learning forecast prescriptive simply because it uses a model, when it recommends no action.
-
Data science process
An iterative cycle of problem framing, data acquisition, preparation, exploration, modeling, evaluation and communication or deployment.
- Remember
- Framing the business question comes first, and evaluation can send you back to preparation or framing.
- Tested as
- Questions ask which step a failing project skipped or what the team should do FIRST.
- Trap
- Jumping to model selection before the target variable and success metric are defined.
-
Data scientist vs. analyst vs. engineer
Engineers build and operate data pipelines and platforms, analysts query and explain existing data, and scientists design experiments and build predictive models.
- Remember
- The scientist's distinctive output is a validated model or inference that generalizes beyond the data in hand.
- Tested as
- Role-assignment scenarios appear in this module and again in Applied Data Science case studies.
- Trap
- Assigning pipeline reliability or ingestion work to the data scientist because they consume the data.
-
Characteristics of big data
Volume (size), velocity (arrival speed), variety (structured, semi-structured and unstructured formats) and veracity (trustworthiness), with value often added.
- Remember
- Each characteristic points to a different remedy: distributed storage for volume, streaming for velocity, schema-on-read for variety, quality checks for veracity.
- Tested as
- Scenarios name a symptom, such as sensor readings arriving every millisecond, and expect you to identify the characteristic driving it.
- Trap
- Calling a dataset big data only because it is large, ignoring velocity and variety.
-
EBDF six elements
The framework groups enterprise big data work into six elements: strategy, architecture, algorithms, processes, functions and artificial intelligence.
- Remember
- Methods live in algorithms, platforms in architecture, repeatable workflows and governance in processes, roles and structures in functions.
- Tested as
- Organizational scenarios ask which capability is missing or weak.
- Trap
- Fixing a governance or role gap by buying technology.
-
pandas .loc vs. .iloc
.loc selects by label and includes both endpoints of a slice, while .iloc selects by integer position and excludes the stop index like core Python.
- Remember
- df.loc[2:4] on a default integer index returns three rows; df.iloc[2:4] returns two.
- Tested as
- Code-reading items use exactly this difference.
- Trap
- Applying Python's exclusive-stop rule to label slicing.
-
NumPy view vs. copy
Basic slicing like a[1:3] returns a view that shares memory with the original, while boolean masks and integer-array indexing return copies.
- Remember
- Changing a view changes the source array; call .copy() when you need independence.
- Tested as
- Snippets that modify a slice and then print the original test this behavior.
- Trap
- Expecting the original array to stay unchanged after editing a slice.
-
Vectorization and broadcasting
Operations applied to whole arrays run in compiled code, and broadcasting stretches compatible shapes, such as (3,1) with (1,4) giving (3,4), without copying data.
- Remember
- Shapes are compared from the trailing dimension and must be equal or 1.
- Tested as
- Questions ask for the result shape or why a loop is slow on large data.
- Trap
- Believing (3,) and (4,) arrays broadcast together; they raise an error.
-
Reproducibility in notebooks
Fixed random seeds, recorded package versions and cells run top to bottom make a notebook's results repeatable by others.
- Remember
- Out-of-order cell execution can produce results that a clean rerun cannot reproduce.
- Tested as
- Process questions ask why a colleague cannot reproduce a model's accuracy.
- Trap
- Blaming the algorithm when an unset random seed or hidden notebook state explains the difference.
Side by side
Concepts candidates mix up, compared on the point that decides the answer.
Descriptive vs. diagnostic vs. predictive vs. prescriptive analytics
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Descriptive | Summarizes historical data, such as monthly sales by region | Answers what happened | Thinking a sophisticated dashboard moves it up the ladder |
| Diagnostic | Explains causes, for example drill-down or root-cause analysis | Answers why it happened | Confusing correlation found in a drill-down with proven cause |
| Predictive | Estimates future or unknown values, such as churn probability | Answers what will happen | Calling it prescriptive because a model is involved |
| Prescriptive | Recommends or optimizes a decision, such as which offer to send | Answers what we should do | Missing that it usually builds on a predictive model |
Data engineer vs. data analyst vs. data scientist
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Data engineer | Builds and operates ingestion, storage and processing pipelines | Owns data availability and reliability | Assigning model validation to this role |
| Data analyst | Queries, summarizes and visualizes existing data to answer business questions | Mostly descriptive and diagnostic work | Assuming analysts never use statistics |
| Data scientist | Designs experiments and builds models that generalize to new data | Owns predictive and prescriptive models and their evaluation | Assigning pipeline uptime to the scientist |
pandas .loc vs. .iloc
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| .loc | Label-based selection of rows and columns | Slice end label is included | Expecting df.loc[0:2] to return two rows |
| .iloc | Integer position-based selection | Slice end position is excluded | Using labels with .iloc on a non-integer index |
Exam traps
Each trap is an answer that looks right. Learn the clue that should change your mind.
-
Model equals prescriptive
- What candidates pick
- Prescriptive analytics for a churn-probability score.
- Why it looks right
- It uses machine learning, which feels like the most advanced category.
- Why it's wrong
- A probability forecasts an outcome; it becomes prescriptive only when something selects or optimizes an action from it.
Clue to watch for Check whether the output recommends what to do or only estimates what will happen.
-
Scientist owns the pipeline
- What candidates pick
- The data scientist should fix the failing nightly ingestion job.
- Why it looks right
- The scientist is the one blocked by the missing data.
- Why it's wrong
- Ingestion reliability belongs to data engineering; the scientist consumes the data and validates models.
Clue to watch for Words like ingestion, scheduling, uptime or pipeline point to the engineer.
-
Inclusive versus exclusive slicing
- What candidates pick
- Two rows for df.loc[2:4] on a default integer index.
- Why it looks right
- Python slices exclude the stop value everywhere else.
- Why it's wrong
- Label slicing with .loc includes both endpoints, so it returns rows 2, 3 and 4.
Clue to watch for Notice whether the snippet uses .loc or .iloc before counting.
-
Size alone makes big data
- What candidates pick
- A 50 GB static table is the clearest big data case in the options.
- Why it looks right
- Volume is the best-known characteristic.
- Why it's wrong
- A continuous high-frequency stream of mixed formats can be the better big data example because velocity and variety drive the architecture choice.
Clue to watch for Look for arrival rate and format mix, not only size.
-
Technology for an organizational gap
- What candidates pick
- Buy a new analytics platform when models are built but never reused or governed.
- Why it looks right
- Platforms promise to solve reuse and sharing.
- Why it's wrong
- The scenario describes missing processes and roles, which the EBDF treats as process and function capabilities, not architecture.
Clue to watch for Symptoms like duplicated effort, no ownership or no review step signal a process or function gap.
-
Skipping problem framing
- What candidates pick
- Start by comparing algorithms when a project's results do not satisfy the business.
- Why it looks right
- Model choice is the visible technical lever.
- Why it's wrong
- If the target and success metric were never agreed, a better algorithm optimizes the wrong thing.
Clue to watch for Stakeholders disagree about what success means, or the target variable is vague.
Answer framework
Use this sequence to place any introductory scenario before you answer.
- Identify the business question the scenario is really asking, and confirm whether a target variable and an agreed success metric already exist before judging anything else.
- Classify the deliverable by the question it answers: what happened, why, what will happen, or what to do.
- Assign the task to the right role by its verb: build and operate pipelines, explain existing data, or model and validate.
- Map organizational symptoms to an EBDF capability before choosing a fix, separating technology gaps from process or role gaps.
- For code snippets, check .loc versus .iloc, view versus copy and array shapes before computing any value, then trace the code line by line.
Practice question
Want more like this? The EBDS practice test has questions for every domain.
Checklist
0 of 10
Tick each one you can do without notes. Your ticks stay in this browser.
Study strategy
Start with topic 1.1 and make the analytics ladder automatic: for any deliverable, ask which question it answers. Then rehearse the role split until you can assign a task by its verb alone. These two habits pay off across the exam, because later case studies open with exactly this kind of framing.
Next, do the Python review hands-on rather than by reading. Open a notebook and test .loc against .iloc, edit a NumPy slice and print the original, and broadcast arrays of mismatched shapes to see which ones fail. Candidates usually underestimate this part because they write Python daily, yet exam snippets target the edge cases that everyday code rarely hits.
Finish with the EBDF review using your Analyst-level notes: list each capability with two example problems that belong to it, and practice separating a technology gap from a process or role gap. Afterwards, take a short timed set mixing classification and code-tracing items, then move to Module 2, where the statistics behind every later model begins.
FAQ
Is Module 1 worth studying if I already passed the Enterprise Big Data Analyst exam?
Yes. The framework review is familiar, but the data science framing and the Python details are new emphases, and they are the cheapest marks available.
How many exam questions come from this module?
The Alliance does not publish module weights. An even split of the 80 questions suggests about nine, but treat that as a planning estimate only.
How deep does the Python knowledge need to be?
Deep enough to read pandas and NumPy code and predict its result, including indexing, slicing, view-versus-copy and broadcasting rules. Writing long programs from memory is not the focus.
What is the difference between predictive and prescriptive analytics?
Predictive analytics estimates an outcome, such as a churn probability. Prescriptive analytics goes a step further and selects or optimizes the action to take, often using a predictive model as an input.
Is the EBDS exam open book?
The Alliance's exam voucher page describes it as open book with the Enterprise Big Data Scientist's Guide; confirm the current rule with APMG-International when you book.
How does this module connect to the rest of the course?
It supplies the process and vocabulary that later modules plug into: statistics and machine learning sit in the algorithms capability, Spark in architecture, and ethics in governance.
Why does the EBDF review matter for a technical exam?
Scenario questions are set inside organizations, so you often need to identify whether the problem is technical, procedural or organizational before choosing a method.
Other Data & Analytics certifications
Where the same subject shows up in other exams. Cards marked "Same subject" open the matching domain.
Test your Domain 1 knowledge
You have covered the analytics ladder, the scientist's role, the Python edge cases and the EBDF capabilities. Check whether you can classify scenarios and trace code at exam speed.