Browse code
remove check for infinite loss when there is no data
This should emphatically *not* be the case for the 1D and
2D learners, as the loss enforce an almost-uniform sampling
of the space in the absence of data.
Joseph Weston authored on 16/11/2017 11:53:49
• Bas Nijholt committed on 20/11/2017 15:49:57
Showing 1 changed files
| ... | ... |
@@ -201,10 +201,6 @@ def test_expected_loss_improvement_is_less_than_total_loss(learner_type, f, lear |
| 201 | 201 |
N = random.randint(50, 100) |
| 202 | 202 |
xs, loss_improvements = learner.choose_points(N) |
| 203 | 203 |
|
| 204 |
- # no data -- loss is infinite |
|
| 205 |
- assert all(l == float('inf') for l in loss_improvements)
|
|
| 206 |
- assert learner.loss() == float('inf')
|
|
| 207 |
- |
|
| 208 | 204 |
for x in xs: |
| 209 | 205 |
learner.add_point(x, f(x)) |
| 210 | 206 |
|