Hi,
I notice that in the best combination of parameters founded by grid_search, the max_leaf_nodes value is 30 whereas it was 10 in the previous exercice. I think that the remark ‘The accuracy and the best parameters of the grid-searched pipeline are similar to the ones we found in the previous exercise, where we searched the best parameters “by hand” through a double for loop.’ should be adjusted.
Hi,
I agree that at the moment when we mention that the results are similar it does not seem obvious as indeed max_leaf_nodes=30 is higher than max_leaf_nodes=10.
But it is also true that the impact of max_leaf_nodes is rather small for that given range of values once learning_rate=0.1 is fixed:
| learning_rate | max_leaf_nodes | mean_test_score | std_test_score | rank_test_score | |
|---|---|---|---|---|---|
| 3 | 0.1 | 25 | 0.868827 | 0.001068 | 1 |
| 2 | 0.1 | 20 | 0.868690 | 0.000550 | 2 |
| 1 | 0.1 | 15 | 0.868281 | 0.000358 | 3 |
| 4 | 0.1 | 30 | 0.868063 | 0.000850 | 4 |
| 0 | 0.1 | 10 | 0.866425 | 0.000359 | 5 |
In this sense max_leaf_nodes=[10, 30] are “similar” as in “same order of magnitude” and “same impact”.
In any case, any suggested wording to make this clearer is welcome.