MySmartCoPilot
Ask Passes Your pass 7 days left 1 day left Last day

Your country

Tools that support it use your country for local currency, number formats, units and paper size. Your choice is saved only in this browser.

GIS & Geospatial Data (QGIS, GeoPandas, PostGIS) Module 5

Python: Shapely and GeoPandas interview questions

Interview questions on “Python: Shapely and GeoPandas”: warm-ups, then what each harder question tests and how a good answer starts.

26 questions: 6 warm-ups and 20 with model answers · Levels: Fresher to staff

Warm-up questions

Short questions with short answers, each answered in full.

  1. Fresher

    Shapely has no coordinate reference system. What does that mean for point.buffer(1000)?

    Answered in the lesson Shapely 2: geometry operations in Python

  2. Fresher

    What is the difference between within and covered_by for a point on a polygon's edge?

    Answered in the lesson Spatial joins: predicates, nearest and CRS traps

  3. Fresher

    What is the difference between clip and an intersection overlay?

    Answered in the lesson Overlay, dissolve, clip and areal weighting

  4. Fresher

    What does a spatial index store, and why does a query still need an exact test?

    Answered in the lesson Spatial indexes: R-trees, sindex and fast joins

  5. Fresher

    Why can a PIN code not stand in for a household's location?

    Answered in the lesson Geocoding, Indian addresses and DIGIPIN

  6. Fresher

    What is an isochrone, and why is it not a circle?

    Answered in the lesson Street networks, routing and isochrones

Questions with model answers

What each question tests, the first step of a good answer and the follow-ups to expect.

  1. Fresher to mid

    A coverage report counts the households inside the union of 1 km buffers around 8 clinics and finds 2,971. A distance query on the same data finds 2,973 households within 1,000 m of a clinic. Explain the difference, say which number you would publish, and how you would make the pipeline give one answer.

    What it tests
    Knowing that a buffer is an inscribed polygon, choosing an exact method for distance questions, and designing a pipeline whose area and count agree by construction.
    First step
    Ask how the buffers were made. A buffer is a polygon whose corners lie on the circle, 64 sides by default for the geometry's method, so it covers slightly less than the true 1 km disc. Find the two households: they are within 1,000 m of a clinic but outside the polygon, in the slivers between its sides and the circle.
    Follow-ups

    From the lesson Shapely 2: geometry operations in Python

    Opening the model answers with your pass…

    Model answers open with any MySmartCoPilot pass.

    • Pro pass · 1 week₹79$2.99
    • 30 days₹179$5.99
    • 1 year₹1,699$45

    One payment; a pass never renews. The price shown is the total you pay: no fees or charges are added at checkout.

    Choose a passI have a licence key

    It also opens every other in-depth part in Learn and every Pro tool. Sales are final, except where the law gives you a right to withdraw, for example in the EU and the UK (refund policy).

    In the answer:

    • Model answer
    • Follow-up answers
    • Common mistakes
    • Rubric
    • Variants
    • about 3 minutes

  2. Mid to senior

    A regression test compares the WKT of a dissolved zone with a stored string. After a dependency upgrade the test fails in CI, although the maps look identical and nobody changed the code. What do you suspect, how do you confirm it, and how should the test be written?

    What it tests
    Understanding that GEOS does the geometry and its version can change descriptions without changing shapes, and writing geometry tests that compare shapes rather than text.
    First step
    Compare the versions first: print shapely.__version__ and shapely.geos_version_string in both environments. A new Shapely wheel brings a new GEOS, and GEOS may return the same polygon from another start vertex, in the other direction, or with extra vertices on straight edges. Then check shapely.equals(old, new) and the two areas.
    Follow-ups

    From the lesson Shapely 2: geometry operations in Python

    In the answer:

    • Model answer
    • Follow-up answers
    • Common mistakes
    • Rubric
    • Variants
    • about 3 minutes

  3. Fresher to mid

    A partner's GeoPackage of 12 wards says EPSG:32643. Transformed to WGS 84 for a web map, all 12 wards collapse into a speck in the Indian Ocean at about 70.5° E, on the equator. What is wrong with the file, how do you prove it, and how do you fix it?

    What it tests
    Telling a wrong CRS label from wrong coordinates, reading bounds against a CRS's units and area of use, and choosing set_crs or to_crs for the fix.
    First step
    Print the layer's total bounds before any transform. If they read like 75.02 to 75.04 and 18.12 to 18.14, the numbers are degrees, while the label says metres of UTM zone 43N: an easting of 75 m lies about 500 km west of the zone's central meridian, which is where the transform put it.
    Follow-ups

    From the lesson GeoDataFrames: reading, writing and CRS

    In the answer:

    • Model answer
    • Follow-up answers
    • Common mistakes
    • Rubric
    • Variants
    • about 3 minutes

  4. Mid to senior

    A nightly job reads households with read_file and a bounding box per district. Since a new web form started supplying the boxes, one district reports 0 households where it had 696. Nothing raised an error. Explain what happened, and how you would make the job both correct and loud when it goes wrong.

    What it tests
    Reading filters together with the CRS of the data they filter, and designing guards for silent empty results.
    First step
    Compare the box's numbers with the layer's bounds and CRS. A web form most likely sends longitude and latitude, while the households layer is in metres of UTM zone 43N. Then ask what a bounding box knows about its own CRS, and what read_file does with four numbers it cannot place.
    Follow-ups

    From the lesson GeoDataFrames: reading, writing and CRS

    In the answer:

    • Model answer
    • Follow-up answers
    • Common mistakes
    • Rubric
    • Variants
    • about 3 minutes

  5. Fresher to mid

    One wards layer, in metres of UTM zone 43N with int64 household counts, must reach a web map, a desktop GIS team and a data scientist's notebook. Which format and options do you send to each, and what do you check before sending?

    What it tests
    Matching formats to receivers by what each keeps (CRS, types, layers, web readability) and verifying a written file by reading it back.
    First step
    List what each receiver needs. The web map needs a file a browser can parse, in WGS 84: RFC 7946 GeoJSON. The GIS team needs the CRS and types intact, perhaps with other layers: a GeoPackage. The notebook wants fast, typed columns: GeoParquet. Then write each, read it back and compare.
    Follow-ups

    From the lesson GeoDataFrames: reading, writing and CRS

    In the answer:

    • Model answer
    • Follow-up answers
    • Common mistakes
    • Rubric
    • Variants
    • about 3 minutes

  6. Fresher to mid

    A report of households per ward, built with sjoin and predicate="intersects", adds up to 5,037 although the survey has 5,000 households. Nothing failed. What happened, how do you find the households involved, and how do you fix the report so that it cannot happen again?

    What it tests
    Recognising double counts from points on shared boundaries or from overlapping polygons, finding them through duplicate keys, and enforcing a one-match rule with a total check.
    First step
    Look for households that appear in more than one row of the join: joined["household"].duplicated(keep=False). Each one matched two wards. Then ask why: a point exactly on a shared edge intersects both wards, as clinic C2 does, and so does a point where two ward polygons overlap because they were digitised separately.
    Follow-ups

    From the lesson Spatial joins: predicates, nearest and CRS traps

    In the answer:

    • Model answer
    • Follow-up answers
    • Common mistakes
    • Rubric
    • Variants
    • about 3 minutes

  7. Mid to senior

    A coverage report computes the share of households within 1.5 km of a clinic from sjoin_nearest with max_distance=1500, and it says 100 %. Field teams know of households much farther away. Find the bug, give the correct share for the lesson's town, and describe how you would rebuild the report.

    What it tests
    Understanding what a distance limit does to rows without a match, computing shares over the right denominator, and handling ties and limits deliberately.
    First step
    Count the rows the join returned and compare them with the number of households in the survey. If they differ, trust nothing about the share until you know which households are missing from the join, and why a share computed over the rows that remain cannot be anything but 100 %.
    Follow-ups

    From the lesson Spatial joins: predicates, nearest and CRS traps

    In the answer:

    • Model answer
    • Follow-up answers
    • Common mistakes
    • Rubric
    • Variants
    • about 3 minutes

  8. Fresher to mid

    A notebook joins 5,000 survey points to the town's wards and every ward shows 0 households. No error appeared and the code worked last week with another file. How do you find the cause, and what do you add so that the notebook fails loudly next time?

    What it tests
    Diagnosing an empty join from CRS mismatches, swapped axes or wrong layers, reading warnings, and adding guards that turn silent failures into errors.
    First step
    Print the CRS and the total bounds of both layers, and reread every warning the notebook printed. Points around 75° and 18° and wards around 500,000 m east cannot meet, whatever the predicate.
    Follow-ups

    From the lesson Spatial joins: predicates, nearest and CRS traps

    In the answer:

    • Model answer
    • Follow-up answers
    • Common mistakes
    • Rubric
    • Variants
    • about 3 minutes

  9. Fresher to mid

    A coverage report says 85.3 % of ward W12 lies within 1 km of a clinic, but the planning team measured 82.9 % on the same data. The report overlays the wards with each clinic's buffer and sums the pieces per ward. Explain the gap and rebuild the calculation so that it cannot recur.

    What it tests
    Understanding that overlaying overlapping buffers counts shared area once per buffer, and building coverage from a dissolved shape with checks on the shares.
    First step
    Check whether the buffers overlap inside W12. Clinics C7 and C8 are 1,825 m apart, so their 1 km rings share a lens of ground. An overlay with the separate rings returns one piece of W12 per ring, and summing the pieces counts that lens twice. Compare the summed area with the area of the union.
    Follow-ups

    From the lesson Overlay, dissolve, clip and areal weighting

    In the answer:

    • Model answer
    • Follow-up answers
    • Common mistakes
    • Rubric
    • Variants
    • about 2 minutes

  10. Fresher to mid

    Census blocks with households and a density column are dissolved into wards with aggfunc={"households": "sum", "density": "mean"}. A ward of 4 km² holds 1,169 households, yet its density comes out well below what those two numbers give. What is wrong, what is the right density, and how do you set up the dissolve?

    What it tests
    Choosing an aggregation rule per column, telling counts from rates, and knowing dissolve's default.
    First step
    Check the density against its own parts: divide the ward's households by its area and compare. Then ask how a mean of the blocks' densities treats two blocks of very different sizes.
    Follow-ups

    From the lesson Overlay, dissolve, clip and areal weighting

    In the answer:

    • Model answer
    • Follow-up answers
    • Common mistakes
    • Rubric
    • Variants
    • about 2 minutes

  11. Mid to senior

    A health department has households per census block and needs households per clinic catchment, whose boundaries cross the blocks. Describe how you would estimate them, how wrong the estimates can be, and how you would report them honestly.

    What it tests
    Areal weighting built from an overlay, its even-spread assumption, measuring its error where truth exists, and better methods with ancillary data.
    First step
    Overlay the blocks with the catchments, give each piece the block's count times the piece's share of the block's area, and sum the pieces per catchment. Then compare the catchment totals with the census total.
    Follow-ups

    From the lesson Overlay, dissolve, clip and areal weighting

    In the answer:

    • Model answer
    • Follow-up answers
    • Common mistakes
    • Rubric
    • Variants
    • about 2 minutes

  12. Mid to senior

    A notebook assigns 2 million delivery points to 9,000 wards by looping over every pair in Python and calling ward.contains(point). It has run for hours. How do you make it fast, how long should the new version take, and how do you show that its answer did not change?

    What it tests
    Estimating work from counts, replacing a loop over pairs with one indexed join, and proving that a faster rewrite gives the same result.
    First step
    Count the work first: 2 million points times 9,000 wards is 18 billion exact tests, one Python call each. An indexed join tests each point only against the wards whose boxes it falls in, usually one or two, so the exact tests drop to a few million and the calls to one.
    Follow-ups

    From the lesson Spatial indexes: R-trees, sindex and fast joins

    In the answer:

    • Model answer
    • Follow-up answers
    • Common mistakes
    • Rubric
    • Variants
    • about 3 minutes

  13. Mid to senior

    Finding the buildings inside a state's floodplain takes minutes, although the buildings layer has a spatial index. The floodplain is one MultiPolygon of 400,000 vertices along every river of the state. Why is the index not helping, and what do you change?

    What it tests
    Seeing that a large shape's bounding box defeats the first step, and restoring small boxes by exploding and subdividing instead of simplifying.
    First step
    Look at the floodplain's bounding box: one box around every river of a state covers most of the state, so nearly every building becomes a candidate, and each candidate is tested against a polygon of 400,000 vertices. Compare the number of candidates with the number of buildings inside.
    Follow-ups

    From the lesson Spatial indexes: R-trees, sindex and fast joins

    In the answer:

    • Model answer
    • Follow-up answers
    • Common mistakes
    • Rubric
    • Variants
    • about 3 minutes

  14. Senior to staff

    In PostGIS, a query with WHERE ST_Distance(geom, :p) < 500 reads the whole 20-million-row table, although geom has a GiST index. Explain what the planner sees, how you confirm it and how you rewrite the query. After a bulk load, the rewritten query is slow again: what next?

    What it tests
    Knowing which conditions an index can serve, reading EXPLAIN output, and how statistics steer the planner's choice of plan.
    First step
    Run EXPLAIN and read the scan node: a Seq Scan with a filter on st_distance means the index was never searched. Then ask which questions a GiST index of bounding boxes can answer, and whether a distance compared with a number is one of them.
    Follow-ups

    From the lesson Spatial indexes: R-trees, sindex and fast joins

    In the answer:

    • Model answer
    • Follow-up answers
    • Common mistakes
    • Rubric
    • Variants
    • about 3 minutes

  15. Fresher to mid

    A health department geocodes 3,000 clinic addresses for a coverage map, and 412 clinics land on exactly the same point in the middle of the district town. Nothing failed. What happened, how do you find every affected clinic, and what do you do about them?

    What it tests
    Recognising coarse fallback matches, finding them from the rank and from repeated coordinates, and handling addresses without a good match honestly.
    First step
    Look at the answers those 412 rows got. They probably share a low place_rank, 18 or below, the rank of a town or a city: the geocoder could not match the street and fell back to the town's centre. Group the results by coordinate and by rank, and list every answer below street level.
    Follow-ups

    From the lesson Geocoding, Indian addresses and DIGIPIN

    In the answer:

    • Model answer
    • Follow-up answers
    • Common mistakes
    • Rubric
    • Variants
    • about 3 minutes

  16. Mid to senior

    A delivery app stores the DIGIPIN of each customer's door, made from the phone's position at sign-up. Riders report that many parcels go to the neighbour's gate. Explain why, and redesign what the app stores and how it uses it.

    What it tests
    Relating a code's cell size to the error of the position it came from, and designing storage that keeps the accuracy and lets a wrong point be corrected.
    First step
    Compare two sizes. A full DIGIPIN cell is about 4 m across, and a phone outdoors is often several metres off, more between tall buildings. Ask which code a position gets when the fix lands a few metres from the door, and what the rider then sees.
    Follow-ups

    From the lesson Geocoding, Indian addresses and DIGIPIN

    In the answer:

    • Model answer
    • Follow-up answers
    • Common mistakes
    • Rubric
    • Variants
    • about 3 minutes

  17. Mid to senior

    You must geocode 200,000 addresses once, for a public dashboard. A colleague proposes a script that calls the public Nominatim service from ten threads, to finish overnight. What do you tell them, and what plan do you propose instead?

    What it tests
    Knowing a public geocoder's usage policy and licence, estimating a job's time from its rate limit, and choosing a source that suits the volume.
    First step
    Check the plan against the usage policy: one request a second from one thread on one machine, bulk geocoding discouraged, and 4 a minute for scripts that run longer than a day. Ten threads break the first rule at once; then work out how long a single thread would take, and which limit applies to it.
    Follow-ups

    From the lesson Geocoding, Indian addresses and DIGIPIN

    In the answer:

    • Model answer
    • Follow-up answers
    • Common mistakes
    • Rubric
    • Variants
    • about 2 minutes

  18. Fresher to mid

    A district report says 2,973 of the town's 5,000 households live within 1 km of a clinic, from 1 km buffers round the 8 clinics. Residents south of the river say they walk much further. How do you check the claim, and what figure would you report instead?

    What it tests
    Seeing that buffers measure straight lines, computing coverage along the network from all clinics at once, and reporting the method beside the figure.
    First step
    Ask what the 1 km is measured along. A buffer counts every household within 1 km in a straight line, across the river and through blocks. Recompute along the street network: one run of Dijkstra's algorithm from all 8 clinics at once gives every junction its walk to the nearest clinic.
    Follow-ups

    From the lesson Street networks, routing and isochrones

    In the answer:

    • Model answer
    • Follow-up answers
    • Common mistakes
    • Rubric
    • Variants
    • about 2 minutes

  19. Mid to senior

    A delivery app's routes look short on the map, but drivers say they are slow. A colleague finds that the router was called with its default settings. What do you check, and how do you make the routes fast?

    What it tests
    Knowing that routers often minimise length by default, deriving time weights from speeds and tags, and checking travel times against measured ones.
    First step
    Check which weight the router minimised. OSMnx's shortest_path minimises length unless told otherwise, so it returns the shortest route, often through slow streets. Add speeds and travel times to the edges, route by travel time, and compare a sample of routes with drivers' recorded times.
    Follow-ups

    From the lesson Street networks, routing and isochrones

    In the answer:

    • Model answer
    • Follow-up answers
    • Common mistakes
    • Rubric
    • Variants
    • about 2 minutes

  20. Senior to staff

    A state plans ambulance bases so that 90 % of villages are within 30 minutes by road. The first analysis routes on OpenStreetMap roads at posted speeds, at all hours. What would you add before trusting the result, and how would you present its uncertainty?

    What it tests
    Knowing what a simple road graph leaves out, checking a network before routing, and presenting coverage as a range with its assumptions.
    First step
    List what the graph leaves out and what each does to the times: turn restrictions, speeds by hour, road condition, missing roads and ferries. Then ask how you would know whether the road data itself has gaps, before any time is trusted.
    Follow-ups

    From the lesson Street networks, routing and isochrones

    In the answer:

    • Model answer
    • Follow-up answers
    • Common mistakes
    • Rubric
    • Variants
    • about 3 minutes

All GIS & Geospatial Data (QGIS, GeoPandas, PostGIS) interview questions · The GIS & Geospatial Data (QGIS, GeoPandas, PostGIS) track