向 Ray Tune 状态添加更多指标 Table(Python,PyTorch)
Add More Metrics to Ray Tune Status Table (Python, PyTorch)
当 运行 tune.run()
对一组配置进行搜索时,是否可以添加更多指标列(即 a
、b
等)状态 table 正在打印出来?
tune.track.log(a=metric1, b=metric2)
将给出以下 table 不包含指标 a
和 b
的列:
== Status ==
Memory usage on this node: 22.1/125.8 GiB
Using FIFO scheduling algorithm.
Resources requested: 1/32 CPUs, 1/4 GPUs, 0.0/65.59 GiB heap, 0.0/22.13 GiB objects
Result logdir: /home/nyxynyx/ray_results/fooba
Number of trials: 4 (3 PENDING, 1 RUNNING)
+--------------+----------+-------+------+-----+
| Trial name | status | loc | lr | x |
|--------------+----------+-------+------+-----|
| fooba_00000 | RUNNING | | 0.01 | 1 |
| fooba_00001 | PENDING | | 0.1 | 1 |
| fooba_00002 | PENDING | | 0.01 | 5 |
| fooba_00003 | PENDING | | 0.1 | 5 |
+--------------+----------+-------+------+-----+
除了 mean_accuracy
之外,我们如何为传递给 tune.track.log()
的每个指标添加一列?
使用 Python 3.7.3 和 Ray 0.8.4
是的!您应该能够使用报告对象执行此操作:https://ray.readthedocs.io/en/latest/tune/api_docs/reporters.html
当 运行 tune.run()
对一组配置进行搜索时,是否可以添加更多指标列(即 a
、b
等)状态 table 正在打印出来?
tune.track.log(a=metric1, b=metric2)
将给出以下 table 不包含指标 a
和 b
的列:
== Status ==
Memory usage on this node: 22.1/125.8 GiB
Using FIFO scheduling algorithm.
Resources requested: 1/32 CPUs, 1/4 GPUs, 0.0/65.59 GiB heap, 0.0/22.13 GiB objects
Result logdir: /home/nyxynyx/ray_results/fooba
Number of trials: 4 (3 PENDING, 1 RUNNING)
+--------------+----------+-------+------+-----+
| Trial name | status | loc | lr | x |
|--------------+----------+-------+------+-----|
| fooba_00000 | RUNNING | | 0.01 | 1 |
| fooba_00001 | PENDING | | 0.1 | 1 |
| fooba_00002 | PENDING | | 0.01 | 5 |
| fooba_00003 | PENDING | | 0.1 | 5 |
+--------------+----------+-------+------+-----+
除了 mean_accuracy
之外,我们如何为传递给 tune.track.log()
的每个指标添加一列?
使用 Python 3.7.3 和 Ray 0.8.4
是的!您应该能够使用报告对象执行此操作:https://ray.readthedocs.io/en/latest/tune/api_docs/reporters.html