Details marks hw02

How did I mark?

Interpolations

I had a pytest unit test to test the specific cases with a simple hill samples.xyz. Download test_all.py to try:

pip install pytest
put test_all.py in the same folder as your my_code_hw02.py
in the console, type pytest

I just compared your values to those obtained with startinpy interpolate(), including for outside the CH and directly on vertices.

The most common mistake was Laplace interpolation directly at a location where a vertex exists… This meant that you inserted no vertex and then deleted the vertex at the end of the code?! Which meant havoc, your TIN is “compromised”.

GFTIN

I didn’t use the dataset I gave you (as previously announced), but found a random place in Limburg with a few trees and buildings and some “hills”, and tested with only that dataset (download d1.laz).

I couldn’t test exactly the GFTIN because you all handled the bbox/convexhull differently, so I ran your function with --input_thinning 1000 and then calculated the differences between your terrain and the one obtained with classification and linear TIN:

  print("=== DTM differences ===")
  grid3 = grid_gftin - grid_c_linear
  write_grid_to_disk("dtm_diff.tiff", grid3, (bbox[0], bbox[3]), 1.0) 

and then I visualised the differences to see if they are generally the same as a correct implementation.

I also verified that your number of vertices in the DT was similar to what I obtain (~5900); I was very lax here, if it was around 5900 it was fine.

Then I ran again with --input_thinning 1000 --max_dist 5.0, and verified that the error patterns are similar to mine and the number of vertices ~6400.

Some remarks in no particular order of importance

Using ChatGPT to help you code is allowed and encourage (I do that all the time…), but for several of you I reckon that ChatGPT was perhaps of too much help, if I may say. I saw code that is very complex and using methods that were not per se in the book or standard (eg for the circumcircle of a triangle), which means that I am not sure you understood what the code precisely does. Some of you had Laplace values that were a bit off, but more than just rounding off errors that accumulate. To mark Laplace I used a tolerance of 0.001 and for most of you it worked, but if it didn’t it’s complex for me to say where it went wrong to be honest. So I would strongly encourage you to minimise the amount of code you copy from ChatGPT.
I didn’t ask for comments in the code this time (an error…) but I highly recommend you comment your code for future-you. Some parts were quite tricky, and I can guarantee you that in a few weeks if you open that code you won’t remember what you did at line XYZ.
Enjoy the holidays ☃️ ❄️ (or perhaps more accurately: 🌧️ ☔ 🥶)