Update README with results

This commit is contained in:
Jack Turner 2020-06-05 09:30:13 +01:00
parent d1bbe9ff2e
commit af12bfb58f

View File

@ -18,7 +18,7 @@ Each command will finish by calling `process_results.py`, which will print a tab
| Method | Search time (s) | CIFAR-10 (val) | CIFAR-10 (test) | CIFAR-100 (val) | CIFAR-100 (test) | ImageNet16-120 (val) | ImageNet16-120 (test) |
|:-------------|------------------:|:-----------------|:------------------|:------------------|:-------------------|:-----------------------|:------------------------|
| Ours (N=10) | 1.73435 | 88.99 +- 0.24 | 92.42 +- 0.33 | 67.86 +- 0.49 | 67.54 +- 0.75 | 41.16 +- 2.31 | 40.98 +- 2.72 |
| Ours (N=10) | 1.73435 | 88.47 +- 1.33 | 91.53 +- 1.62 | 66.49 +- 3.08 | 66.63 +- 3.14 | 38.33 +- 4.98 | 38.33 +- 5.22 |
| Ours (N=100) | 17.4139 | 89.18 +- 0.29 | 91.76 +- 1.28 | 67.17 +- 2.79 | 67.27 +- 2.68 | 40.84 +- 5.36 | 41.33 +- 5.74
`./reproduce 500` will produce the following table (which is the same as what we report in the paper):
@ -31,4 +31,6 @@ Each command will finish by calling `process_results.py`, which will print a tab
To try different sample sizes, simply change the `--n_samples` argument in the call to `search.py`, and update the list of sample sizes on line 51 of `process_results.py`.
Note that search times may vary from the reported result owing to hardware setup.
The code is licensed under the MIT licence.