General Discussions
Discuss any topics that are not product-specific here.
In a last few days, I didn't hold the time to do the challenge because of my work and year-end party, but today I did it! Certainly, today's challenge is easier before a few days.
The easiest way to do this was, as expected, to explode the number of records, so I had to do it different way.
So,it took about 4 minutes in part 2.
P1 was straightforward. P2 required some ugly brute force.
P2: Generate the min/max for each path by row. Then do a calc to determine if y_min <=[row-1:y_max] THEN 1 for the 14 rows possible. Filtered where THEN =0 and plug magic row back into original logic to determine x,y coordinate then final logic
Managed to catch up on Day 15! Part 2 was really fun to try and figure out/optimise after realising I couldn't scaffold a 4m*4m grid to check against. Following a couple of ideas that still lead to huge blowups and cancelling a couple of workflows after 15-20 minutes, I decided to try and use spatial for this which actually ended up running in 0.2 seconds! Entire flow for P1+P2 takes ~17s.
It was difficult to resolve the performance issue for Part2.
Process took 12 mins for part2 and most of them are for multi-Row formula to process 31M records.
So - took a completely different approach to part 1 vs. part 2.
Very quickly realised (i.e. after running for an hour) that solving part 2 by using a brute-force array would not work.
Part 1 I took a very brute-force approach.
- For each sensor - find the manhattan distance
- Then create a diamond shape of points based on manhattan distance of blocked spots
- Then combine all these
... and filter for the focus row.
This is almost all done in the input parser:
And then the outer call looks like this.
Part 2:
I originally did part 2 in the same way - it worked for the example test but didn't scale to the full data.
To make this scale - switched to Spatial since this is really an intersection of squares exercise.
To explain this.
The top piece creates a combined polygon of the diamonds for the coverage for every sensor.
The bottom piece creates the bounding box.
Then you subtract the top from the bottom and that tells you what pieces remain.
Split them into polygons
- The example test had thin sliver-type polygons - so I did a bit more work to see if I could fit in a 1x1 beacon - but this wasn't needed on the final data.
Things to watch out for:
- Use X & Y as floating point long & lat
- Scale these values up / down to fit within the range of -2;2 (this allows you to use Alteryx spatial tools which only allow Earth coords and staying within the range of -2;2 reduces issues of curvature since spatial tools are not flat-plain geometry but geography (spherical earth).
Step 1: Overlapping sensor areas:
Step 2: Merge into one poly:
Subtract from Bounding Box:
Final Resul (tiny box with the centroid in the middle)
although get answer via spatial, but still want to find another method. take less than 10 seconds in total.
remember to take negative... stuck for hours due to the negative...
identify whether square touch the y = 2000000,
then merge the range instead of generating every position.
and deduct the Beacon.
part2:
using the equation method.
assume all the side is 4 equation
upper_upward = y = x + c + distance
lower_upward = y = x + c - distance
upper_downward = y = -x + c + distance
lower_downward = y = -x + c - distance
as only one position, the point must attached to 4 lines above. (2 lines in x and 2 lines in y)
hence there must have variance of 2 for each pair above.
get the average of the C (just -1 or +1 as they are 1 position above or bottom), that is the equation for x and y.
intersect the equation, we can get the point x and y
and multiple the required 4000000 + [y]
