I85PREP
⌘ K
PCED study materials
PCED study resources

PCED practice questions

One question at a time. Build your PCED understanding.

First 50 of 200 questions. Each one lists the choices, the correct answer, and an explanation. The checker at the top of this page saves progress in this browser when it is available.

Continue in the practice tool above to load more questions, 50 at a time.

Question 1Data collection fundamentals

In a PCED data-collection pipeline, which component is responsible for translating a raw proprietary sensor byte stream into a normalized record before storage?

  1. The dashboard rendering layer
  2. The retention policy engine
  3. The access-control list
  4. The parser or decoder stage Correct answer.

Explanation

Parsing (decoding) is the stage that converts raw input into structured, normalized fields that downstream storage and analytics can use. Retention, visualization, and authorization operate on already-structured data or metadata, not on the byte-level interpretation itself.

Question 2Data collection fundamentals

A collector writes events to disk in fixed-size blocks and tracks a per-block checksum. What is the primary purpose of the checksum in this design?

  1. To encrypt the block at rest
  2. To detect corruption or truncation of a stored block Correct answer.
  3. To compress the block before writing
  4. To order blocks by arrival time

Explanation

A per-block checksum lets the reader verify that the bytes read back match what was written, catching corruption or partial writes. Compression, encryption, and ordering are separate concerns that a checksum does not provide.

Question 3Data collection fundamentals

An engineer must choose between pull-based and push-based collection for thousands of short-lived edge devices that frequently go offline. Which characteristic most favors a push model here?

  1. Devices can deliver data when connectivity exists without the collector polling unreachable hosts Correct answer.
  2. Push eliminates the need for any schema validation
  3. Push always produces smaller payloads than pull
  4. Push guarantees exactly-once delivery by default

Explanation

Push lets intermittently connected devices send data on their own schedule, avoiding wasted polls to offline hosts. Payload size, validation, and delivery guarantees are independent of the pull/push choice and must be designed separately.

Question 4Data collection fundamentals

A source emits timestamps in local time with no offset, and the collector stores them unchanged. Which data-quality problem is most directly introduced?

  1. Loss of numeric precision in counters
  2. Excessive storage cost per record
  3. Ambiguous event ordering across time zones Correct answer.
  4. Reduced network throughput

Explanation

Local timestamps without an offset cannot be reliably converted to a common timeline, so events from different zones may be misordered. Storage cost, numeric precision, and throughput are unrelated to timezone ambiguity.

Question 5Data collection fundamentals

Which practice best supports idempotent ingestion when a collector may retry a batch after a timeout?

  1. Increasing the retry count until the batch succeeds
  2. Disabling acknowledgements to reduce latency
  3. Compressing the batch before retransmission
  4. Attaching a stable deduplication key to each record and rejecting duplicates on write Correct answer.

Explanation

Idempotency requires that a repeated delivery produce the same stored state, which is achieved by deduplicating on a stable key. More retries, missing acknowledgements, and compression do not prevent duplicate records from being stored.

Question 6Data collection fundamentals

A telemetry agent buffers events in memory and flushes every 30 seconds. A host crashes with 20 seconds of unflushed events. What is the most accurate description of this loss?

  1. Data loss caused by an in-memory buffer without durable write-ahead Correct answer.
  2. Data loss caused by schema drift
  3. Data loss caused by clock skew
  4. Data loss caused by access-control misconfiguration

Explanation

Events held only in memory are lost on crash; a durable write-ahead log or more frequent flush would bound the loss window. Schema drift, clock skew, and access control do not explain the disappearance of buffered events.

Question 7Data collection fundamentals

When designing a schema for event collection, which approach best handles sources that add new optional fields over time?

  1. Use a versioned schema that tolerates unknown optional fields while validating required ones Correct answer.
  2. Reject any record whose field set differs from the original
  3. Require every source to emit identical fields forever
  4. Store all records as untyped strings

Explanation

Versioned, forward-tolerant schemas let producers add optional fields without breaking consumers, while still enforcing required fields. Rejecting variation, storing untyped strings, or freezing fields all block legitimate evolution.

Question 8Data collection fundamentals

What does a watermark represent in a streaming collection system that must handle out-of-order events?

  1. The number of partitions assigned to a consumer
  2. The encryption key used for the stream
  3. The maximum size of a single record
  4. A bound on how late events may arrive before a window is considered complete Correct answer.

Explanation

A watermark tracks event-time progress and expresses how far behind the stream may still receive late data. Record size, encryption, and partition assignment are unrelated to event-time windowing.

Question 9Data collection fundamentals

A collector must sample high-volume metrics without losing rare error events. Which sampling strategy best meets that requirement?

  1. Uniform random sampling of all events at a fixed rate
  2. Head-based sampling decided before the event outcome is known
  3. Sampling only the first event of each minute
  4. Tail-based sampling that retains all error traces and samples successful ones Correct answer.

Explanation

Tail-based sampling decides after the outcome is known, so rare errors can be kept while ordinary successes are thinned. Uniform and head-based sampling decide without outcome knowledge and would drop errors proportionally, and time-based sampling ignores rarity entirely.

Question 10Data collection fundamentals

Which statement correctly distinguishes structured from semi-structured data in a collection context?

  1. Structured data conforms to a fixed schema, while semi-structured data carries self-describing tags without a rigid schema Correct answer.
  2. Structured data never requires a parser
  3. Structured data is always stored in memory, while semi-structured data is always on disk
  4. Semi-structured data cannot be validated at all

Explanation

Structured records follow a predefined schema, whereas semi-structured formats such as JSON or XML embed structure in the data itself and tolerate variation. Storage medium, validatability, and the need for parsing do not define the distinction.

Question 11Data collection fundamentals

A pipeline needs to enrich incoming events with reference data from a slowly changing dimension table. Where should this lookup occur to minimize per-event latency?

  1. In a local cache refreshed periodically from the dimension table Correct answer.
  2. By deferring all enrichment to the reporting layer
  3. By embedding the entire dimension table in each event
  4. By querying the remote dimension database for every event

Explanation

Caching the dimension locally avoids a remote round trip per event while staying acceptably fresh through periodic refresh. Per-event remote queries add latency, embedding the table inflates payloads, and deferring enrichment shifts cost rather than removing it.

Question 12Data collection fundamentals

What is the main risk of using a monotonically increasing local counter as the sole identifier for records collected from many hosts?

  1. Records will be rejected for schema drift
  2. Identifiers will exceed the numeric range
  3. Timestamps will be lost
  4. Identifiers will collide across hosts Correct answer.

Explanation

Independent counters on different hosts generate the same values, producing collisions that break deduplication and joins. Numeric range, schema drift, and timestamp loss are separate issues not caused by per-host counters.

Question 13Data collection fundamentals

Which encoding is most appropriate when a collector must transmit many repetitive numeric records with minimal bandwidth and the consumer controls both ends?

  1. Pretty-printed JSON with full field names
  2. A compact binary encoding with a shared schema Correct answer.
  3. XML with verbose element names
  4. Plain-text CSV with quoted headers on every row

Explanation

A shared-schema binary encoding removes repeated field names and text overhead, minimizing bandwidth when both ends agree on the format. Pretty JSON, verbose XML, and repeated CSV headers all add substantial redundant bytes.

Question 14Data collection fundamentals

A regulatory requirement states that collected personal data must be deleted after 90 days. Which pipeline capability directly enforces this?

  1. A sampling rate applied at ingestion
  2. A compression codec applied at write time
  3. A retention policy with automated expiry and deletion Correct answer.
  4. A checksum stored with each block

Explanation

Retention policies define lifetimes and trigger automated deletion when the period elapses. Compression, sampling, and checksums affect size, volume, and integrity but never remove data by age.

Question 15Data collection fundamentals

During collection, a numeric field occasionally arrives as an empty string. Which handling best preserves data quality while keeping the pipeline running?

  1. Halt the pipeline until the source is fixed
  2. Coerce the empty string to a null value and flag it in a quality metric Correct answer.
  3. Drop the entire record without logging
  4. Silently convert the empty string to zero

Explanation

Treating the empty string as null preserves the distinction between missing and zero while a quality metric surfaces the issue. Converting to zero fabricates data, silent dropping loses information, and halting reduces availability.

Question 16Data collection fundamentals

Which mechanism best protects the confidentiality of data in transit between an edge collector and a central ingestion endpoint?

  1. A retention schedule
  2. Mutual TLS between the collector and the endpoint Correct answer.
  3. A per-block checksum
  4. A compression codec

Explanation

Mutual TLS encrypts the channel and authenticates both ends, protecting confidentiality in transit. Checksums, compression, and retention address integrity, size, and lifetime rather than confidentiality.

Question 17Data collection fundamentals

A collector aggregates counters over one-minute windows but the reporting layer expects cumulative totals since process start. What transformation must occur?

  1. Resample the windows to hourly granularity
  2. Convert per-window deltas into a running cumulative sum Correct answer.
  3. Apply a checksum to each window
  4. Convert cumulative totals into per-window deltas

Explanation

The stored data is in delta form, so producing cumulative totals requires accumulating deltas over time. The reverse transformation, resampling, and checksums do not yield the expected cumulative view.

Question 18Data collection fundamentals

Which factor most directly determines the maximum sustainable ingestion throughput of a single collector process?

  1. The rate at which it can parse, validate, and persist records Correct answer.
  2. The length of the retention period
  3. The number of dashboards built on the data
  4. The color scheme of the monitoring UI

Explanation

Throughput is bounded by the work done per record in parsing, validation, and persistence. Dashboards, UI styling, and retention length do not limit how fast records can be processed.

Question 19Data collection fundamentals

An engineer observes that late-arriving events are being assigned to the wrong hourly bucket. Which collection design change most directly addresses this?

  1. Window on event time and allow a bounded lateness period before finalizing Correct answer.
  2. Reduce the number of partitions
  3. Increase the batch size of each write
  4. Window on processing time with no lateness allowance

Explanation

Event-time windows with an allowed lateness bound place records in the correct bucket even when they arrive out of order. Processing-time windows ignore the event timestamp, and batch size or partition count do not fix bucket assignment.

Question 20Data collection fundamentals

What is the role of a dead-letter queue in a data-collection pipeline?

  1. To compress records for long-term archiving
  2. To store the most recent successful records for fast replay
  3. To hold records that repeatedly fail processing so they can be inspected and reprocessed Correct answer.
  4. To encrypt records before transmission

Explanation

A dead-letter queue isolates records that fail validation or processing, preventing them from blocking the stream while preserving them for diagnosis and replay. It is not a cache of successes, an encryption layer, or an archive codec.

Question 21Data collection fundamentals

A source sends duplicate events because its upstream system retries on ambiguous acknowledgements. Which combination best bounds the impact on downstream aggregates?

  1. Aggregate first and deduplicate afterward
  2. Increase the aggregation window to one day
  3. Store duplicates and rely on dashboards to hide them
  4. Deduplicate on a stable key before aggregation Correct answer.

Explanation

Deduplicating before aggregation prevents duplicate contributions to sums and counts. Aggregating first would already have double-counted, a longer window does not remove duplicates, and hiding them in dashboards leaves stored data wrong.

Question 22Data collection fundamentals

Which statement best describes the trade-off of increasing batch size in a collector that sends records to a remote endpoint?

  1. Larger batches improve throughput but increase latency and the amount of data at risk per failure Correct answer.
  2. Larger batches improve both throughput and latency with no downside
  3. Larger batches reduce throughput but improve durability
  4. Larger batches have no effect on latency or failure exposure

Explanation

Batching amortizes per-request overhead for higher throughput, but records wait longer before transmission and a failed batch risks more data at once. There is no free improvement, and batch size clearly affects both latency and failure exposure.

Question 23Data collection fundamentals

A collector must record the origin of each event so analysts can trace it back to a specific device and software build. Which metadata is most directly required?

  1. The compression codec used for storage
  2. The retention period of the dataset
  3. A device identifier and a software version tag attached to each record Correct answer.
  4. A checksum of the record payload

Explanation

Provenance requires identifying which device and build produced an event, so device ID and version tags must be attached at collection time. Checksums, retention, and codecs describe integrity, lifetime, and storage format, not origin.

Question 24Data collection fundamentals

An engineer needs to replay six months of collected events into a rebuilt analytics system. Which property of the stored data is most essential?

  1. The collector used push rather than pull
  2. The raw events were retained durably in an immutable, replayable form Correct answer.
  3. The events were compressed with the fastest codec
  4. The dashboards were rendered in real time

Explanation

Replay requires that the original events still exist in a durable, immutable store that can be read again in order. Codec speed, dashboard rendering, and transport direction do not determine whether historical data can be reprocessed.

Question 25Data collection fundamentals

Which scenario best illustrates schema drift in a collection pipeline?

  1. A consumer reads the same schema version for a year
  2. A collector compresses records before writing
  3. A producer renames a field and changes its type from integer to string without notice Correct answer.
  4. A dashboard adds a new chart to an existing panel

Explanation

Schema drift occurs when producers change field names or types outside an agreed contract, breaking consumers that expect the old shape. Stable schemas, compression, and dashboard edits do not alter the data contract.

Question 26Data collection fundamentals

A team is choosing between pull-based and push-based collection for metrics emitted by thousands of short-lived containers. Which characteristic of push-based collection makes it better suited to this environment?

  1. The collector can enforce a fixed polling interval for every container.
  2. The collector can guarantee delivery ordering without acknowledgements.
  3. The collector does not need to discover and poll each container before it terminates. Correct answer.
  4. The collector can deduplicate samples by querying each container twice.

Explanation

Push-based collection has the agent send data to an endpoint, so ephemeral producers never need to be individually discovered or polled before they disappear. Pull-based scraping requires the collector to know and reach each target, which fails when containers live for only seconds.

Question 27Data collection fundamentals

A collector receives JSON events where an optional field named `region` is absent in about 40% of records. Which collection-side handling keeps the dataset usable for downstream aggregation without inventing information?

  1. Store the absent field as an explicit null and record that the source omitted it. Correct answer.
  2. Fill the absent field with the most frequent region observed so far.
  3. Drop the entire event because it does not conform to the full schema.
  4. Replace the absent field with the string "unknown" so every record has a value.

Explanation

Representing an omitted optional field as null preserves the distinction between a missing value and a real value, and downstream aggregation can then choose how to treat it. Substituting strings or imputed values fabricates data that analysts cannot later separate from observed values.

Question 28Data collection fundamentals

Which statement best describes the purpose of a schema registry in a data-collection pipeline?

  1. It stores versioned schemas so producers and consumers can validate and evolve record formats compatibly. Correct answer.
  2. It encrypts records at rest using keys rotated on a fixed schedule.
  3. It routes records to partitions based on a hash of the record key.
  4. It buffers records in memory so slow consumers do not block fast producers.

Explanation

A schema registry holds versioned definitions of record formats and lets producers and consumers check compatibility as schemas evolve. Buffering, encryption, and partitioning are handled by other components of the pipeline.

Question 29Data collection fundamentals

A collector reads rows from a relational table into a data lake. The table has no reliable updated-at column, and rows are occasionally modified in place. Which extraction strategy best limits the chance of missing modified rows?

  1. Extract only rows whose primary key is greater than the last key seen.
  2. Extract a random sample of rows on each run to detect changes.
  3. Extract only rows created after the last extraction timestamp.
  4. Periodically extract the full table and reconcile it against previously loaded rows. Correct answer.

Explanation

Without a reliable change timestamp or version column, incremental filters on key or creation time cannot detect in-place updates. Periodic full extraction with reconciliation is the safe fallback because every current row is observed.

Question 30Data collection fundamentals

A collector writes each event to a local append-only file before acknowledging it to the source, and later ships the file to central storage. What property does this design primarily provide?

  1. Automatic schema validation of each event.
  2. Durability of accepted events across a collector restart. Correct answer.
  3. Strict global ordering of events from all sources.
  4. Lower storage cost than writing directly to central storage.

Explanation

Persisting to a local append-only log before acknowledgement means an accepted event survives a process restart and can be re-shipped. It does not by itself reduce cost, impose global ordering, or validate schemas.

Question 31Data collection fundamentals

A collection pipeline ingests clickstream events whose timestamps are set by client devices with unreliable clocks. Which approach produces the most trustworthy event-time ordering?

  1. Record both device time and collector receipt time, and use receipt time as the authoritative ordering key. Correct answer.
  2. Use device time alone because it reflects when the user actually acted.
  3. Use the order in which events reach the broker, ignoring all timestamps.
  4. Overwrite device time with the current server time before storage.

Explanation

Capturing both timestamps preserves the device's claimed event time while providing a server-side reference that is not subject to client clock skew. Using receipt time as the authoritative key keeps ordering trustworthy while still allowing later analysis of device time.

Question 32Data collection fundamentals

An ingestion endpoint must accept bursts that occasionally exceed steady-state capacity by a factor of five. Which design element most directly absorbs those bursts without dropping data?

  1. A durable queue between the endpoint and downstream processing sized for expected burst volume. Correct answer.
  2. A larger thread pool in the endpoint with no queue in front of it.
  3. A shorter client retry timeout so producers resend sooner.
  4. Synchronous writes to the downstream database on every request.

Explanation

A durable buffer decouples arrival rate from processing rate, letting the endpoint accept bursts and downstream consumers drain them later. Retry tuning, synchronous writes, and extra threads do not create a place for excess data to wait.

Question 33Data collection fundamentals

A team must collect audit events that regulators require to be tamper-evident and retained for seven years. Which combination of collection properties best satisfies both requirements?

  1. Append-only storage with cryptographic chaining of records and an enforced retention policy. Correct answer.
  2. Mutable storage with access controls and quarterly manual review.
  3. Compressed storage with a rolling seven-day window and daily backups.
  4. Replication across three regions with automatic deletion after one year.

Explanation

Tamper evidence requires that records cannot be silently altered, which append-only storage plus cryptographic chaining provides, while retention policy enforces the seven-year horizon. Access controls alone do not make records tamper-evident, and shorter windows fail the retention requirement.

Question 34Data collection fundamentals

A collector batches records and sends them to a remote endpoint. Increasing batch size improves throughput but raises the maximum data loss if the collector process dies before a flush. Which design change best reduces that loss without giving up most of the throughput gain?

  1. Flush on a size or time threshold, whichever comes first, and persist the batch before sending. Correct answer.
  2. Remove batching entirely and send one record per request.
  3. Disable acknowledgement handling so sends complete faster.
  4. Increase batch size further so flushes occur less often.

Explanation

Combining a size threshold with a time threshold bounds how long records wait, and persisting the batch before sending means an in-flight batch survives a crash. This retains most batching efficiency while capping exposure.

Question 35Organizing and cleaning data

In a tidy rectangular dataset, what does each row represent?

  1. A single variable
  2. A single summary statistic
  3. A single data type
  4. A single observational unit Correct answer.

Explanation

Tidy data follows the convention that each row is one observation of one observational unit, while each column is one variable. Grouping summaries or variable definitions into rows breaks that convention and complicates downstream analysis.

Question 36Organizing and cleaning data

A table stores one column per month (jan_sales, feb_sales, mar_sales). Which reshaping converts this wide layout into a long layout with month and sales columns?

  1. Deduplicating rows by month
  2. Joining the table to itself on month
  3. Pivoting wider on the month columns
  4. Melting or unpivoting the month columns Correct answer.

Explanation

Unpivoting (melting) collapses multiple measurement columns into a key column naming the source column and a value column holding the measurement. The result is one row per entity-month pair, which is the long, tidy form.

Question 37Organizing and cleaning data

Which value most appropriately represents a missing numeric measurement in a database column typed as integer?

  1. The string 'N/A'
  2. A NULL marker Correct answer.
  3. The number 0
  4. The number -1

Explanation

NULL is the standard relational marker for an absent value and propagates correctly through most aggregation and comparison logic. Substituting sentinel strings or numbers silently blends missingness with real values.

Question 38Organizing and cleaning data

A dataset of customer emails contains values like ' [email protected] ' and '[email protected]'. Which cleaning step best standardizes these before deduplication?

  1. Sorting the column alphabetically
  2. Removing all rows containing uppercase letters
  3. Trimming whitespace and lowercasing the domain and local part Correct answer.
  4. Hashing each value with a random salt

Explanation

Trimming surrounding whitespace and normalizing case produces comparable canonical values so that duplicates collapse. Email local parts are technically case-sensitive, but in practice most providers treat them case-insensitively, so lowercasing is a common normalization.

Question 39Organizing and cleaning data

A column of dates is stored as text in several formats such as '2023-03-05', '05/03/2023', and 'March 5, 2023'. What is the most reliable first step to make them analyzable?

  1. Concatenate the values into one string
  2. Sort the column lexicographically
  3. Parse with an explicit format per pattern and cast to a date type Correct answer.
  4. Compute the mean of the text values

Explanation

Ambiguous text dates must be parsed with explicit format patterns, then cast to a native date or timestamp type. Without explicit formats, day-month ambiguity like 05/03 versus 03/05 can silently swap days and months.

Question 40Organizing and cleaning data

Which statement correctly describes the difference between filtering and trimming in data cleaning?

  1. Filtering removes columns; trimming removes rows
  2. Filtering removes whitespace; trimming removes rows by condition
  3. Filtering and trimming both remove duplicate rows
  4. Filtering removes rows by condition; trimming removes leading and trailing whitespace from strings Correct answer.

Explanation

Filtering selects which rows remain based on a predicate, whereas trimming is a string operation that strips surrounding whitespace. They operate on different axes of the data and are not interchangeable.

Question 41Organizing and cleaning data

A survey records age as text with entries like '34', 'thirty-four', and 'unknown'. Which approach best prepares this column for numeric analysis?

  1. Cast values to numeric, coercing non-numeric entries to missing, then handle missingness Correct answer.
  2. Delete the entire column because it contains text
  3. Replace every value with the column mode
  4. Average the raw strings directly

Explanation

Coercing to numeric turns unparseable entries into missing values, which can then be imputed, flagged, or excluded explicitly. This preserves the valid numeric measurements instead of discarding the whole variable.

Question 42Organizing and cleaning data

What is the primary purpose of assigning a stable, unique identifier to each record before merging datasets?

  1. To convert all columns to text
  2. To make join results deterministic and allow tracing back to source records Correct answer.
  3. To sort the rows automatically
  4. To reduce the file size on disk

Explanation

A stable key ensures joins match the intended records and lets you audit which source row produced each output row. Without it, joins on mutable or non-unique fields risk duplicate matches and silent row multiplication.

Question 43Organizing and cleaning data

After a left join of orders (10,000 rows) to a customers table, the result has 12,400 rows. What is the most likely cause?

  1. The customers table has duplicate keys, so some orders matched multiple customer rows Correct answer.
  2. The left join removed unmatched orders
  3. The orders table was sorted incorrectly
  4. The join converted numeric keys to text

Explanation

A left join preserves every left row but can multiply rows when the right table has multiple rows per key. Duplicate keys in the customers table therefore inflate the output beyond the left row count.

Question 44Organizing and cleaning data

Which operation removes exact duplicate records while keeping one copy of each distinct record?

  1. Normalization
  2. Deduplication Correct answer.
  3. Sampling
  4. Imputation

Explanation

Deduplication identifies rows that are identical across the relevant key or full record and retains a single representative. Sampling, imputation, and normalization address different cleaning concerns.

Question 45Organizing and cleaning data

Two customer records share the same email but differ in name spelling ('Jon' versus 'Jonathan'). Which technique is designed to group such near-duplicate records?

  1. Random sampling
  2. Exact string equality matching
  3. Fuzzy matching or record linkage Correct answer.
  4. Column-wise min-max scaling

Explanation

Fuzzy matching and record linkage compare fields with similarity measures so that records differing slightly can be grouped as the same entity. Exact equality would treat the variants as distinct.

Question 46Organizing and cleaning data

A numeric column has values ranging from 1 to 1,000,000 and a few extreme outliers. Which cleaning action is most defensible before modeling?

  1. Multiply every value by 1,000
  2. Silently delete every value above the 90th percentile
  3. Replace all values with the column mean
  4. Investigate outliers, then cap, transform, or flag them with documented rules Correct answer.

Explanation

Outliers should be examined to determine whether they are errors or genuine extreme values, then handled with a documented rule such as winsorizing, log transform, or flagging. Silent deletion biases the data and is not reproducible.

Question 47Organizing and cleaning data

What does min-max normalization do to a numeric column?

  1. It rescales values linearly so the minimum maps to 0 and the maximum maps to 1 Correct answer.
  2. It replaces values with their rank order
  3. It removes values outside the interquartile range
  4. It converts values to z-scores with mean 0 and standard deviation 1

Explanation

Min-max normalization applies a linear transform using the observed minimum and maximum, mapping the range onto [0, 1]. It preserves relative distances but is sensitive to outliers that stretch the range.

Question 48Organizing and cleaning data

A column contains inconsistent category labels: 'USA', 'U.S.A.', 'United States', and 'us'. Which cleaning approach standardizes them most reliably?

  1. Drop all rows with any of these labels
  2. Treat each spelling as a separate category
  3. Encode each label as a distinct integer by first appearance
  4. Map variants to a canonical value using a lookup or normalization rules Correct answer.

Explanation

A mapping table or rule set converts known variants to one canonical category, making counts and joins meaningful. Encoding by first appearance would assign different integers to what is really the same country.

Question 49Organizing and cleaning data

Which statement best characterizes the role of a data dictionary during cleaning?

  1. It stores the raw backup of the dataset
  2. It automatically imputes missing values
  3. It compresses the dataset for storage
  4. It documents each field's meaning, units, allowed values, and provenance Correct answer.

Explanation

A data dictionary records field definitions, units, valid ranges, and provenance so cleaning decisions can be validated and reproduced. It is documentation, not a processing engine or storage mechanism.

Question 50Organizing and cleaning data

A pipeline must be reproducible. Which practice best supports that goal for cleaning steps?

  1. Relying on memory to repeat the same manual edits
  2. Deleting the raw data after the first clean pass
  3. Manually editing cells in a spreadsheet and saving over the raw file
  4. Encoding cleaning steps as versioned scripts or configuration with logged parameters Correct answer.

Explanation

Versioned code or configuration captures each transformation and its parameters so the output can be regenerated exactly. Manual edits overwrite evidence and cannot be rerun consistently.