ClearML/clearml-docs
1
0
Fork 0
mirror of https://github.com/clearml/clearml-docs synced 2025-04-20 22:35:32 +00:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity

Update FAQ about hyperparameter v. metric comparison (#794)

Browse Source
This commit is contained in:
pollfly 2024-03-11 19:24:03 +02:00 committed by GitHub
parent 1a94010023
commit 12482948d3
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 13 additions and 47 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

58
docs/faq.md
View File

@ -556,55 +556,21 @@ tab, click `Download full log`.
#### How do I create a graph comparing hyperparameters vs. model accuracy? <a id="compare-graph-parameters"></a> #### How do I create a graph comparing hyperparameters vs. model accuracy? <a id="compare-graph-parameters"></a>
You can use the UI's [experiment comparison features](webapp/webapp_exp_comparing.md) to compare the logged hyperparameter You can use the UI's [experiment comparison features](webapp/webapp_exp_comparing.md) to compare the logged hyperparameter
and accuracy values of several experiments. and accuracy values of several experiments. In the experiment comparison page, under the **HYPERPARAMETERS** tab
you can visualize experiments' hyperparameter values in relation to performance metrics in a scatter plot or parallel
coordinates plot:
* [Scatter plot](webapp/webapp_exp_comparing.md#scatter-plot): View the correlation between a selected hyperparameter and
metric. For example, the image below shows a scatter plot that displays the values of a performance metric (`epoch_accuracy`)
and a hyperparameter (`epochs`) of a few experiments:
In an experiment comparison page, under the **HYPERPARAMETERS** tab, you can view the experiments' hyperparameter values ![Scatter plot comparison](img/faq_compare_scatter.png)
in relation to a specific metric (e.g. accuracy) in a parallel coordinates plot.
The image below shows a parallel coordinates plot which displays the values of selected hyperparameters (`base_lr`, * [Parallel coordinates plot](webapp/webapp_exp_comparing.md#parallel-coordinates-mode): View the impact of hyperparameters
`batch_size`, and `number_of_epochs`) and a performance metric (`accuracy`) of three experiments. on selected metric(s). For example, the image below shows
a parallel coordinates plot which displays the values of selected hyperparameters (`base_lr`, `batch_size`, and
`number_of_epochs`) and a performance metric (`accuracy`) of three experiments:
![Parallel Coordinates](img/compare_parallel_coordinates.png) ![Parallel Coordinates](img/compare_parallel_coordinates.png)
You can also visualize the differences in a scatter plot. In each experiment whose values will be compared, report a plot
with a single point, x-axis for the hyperparameter value, and Y-axis for the accuracy.
In the code below, the task reports a single-point scatter plot with `number_layers` as the x-axis and
`accuracy` as the Y-axis:
```python
number_layers = 10
accuracy = 0.95
Task.current_task().get_logger().report_scatter2d(
title="performance",
series="accuracy",
iteration=0,
mode='markers',
scatter=[(number_layers, accuracy)]
)
```
When these experiments are compared in the UI's experiment comparison, all the reported `performance/accuracy` values
are displayed in a single plot.
![Comparison plot](img/compare_plots.png)
Another option is a histogram chart:
```python
number_layers = 10
accuracy = 0.95
Task.current_task().get_logger().report_histogram(
title="performance",
series="accuracy",
iteration=0,
labels=['accuracy'],
values=[accuracy],
xlabels=['number_layers %d' % number_layers]
)
```
![Comparison plot histogram](img/compare_plots_hist.png)
<br/> <br/>

BIN
docs/img/faq_compare_scatter.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 133 KiB