ClearML/clearml-docs
1
0
Fork 0
mirror of https://github.com/clearml/clearml-docs synced 2025-06-26 18:17:44 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Add metric encoding information (#578)

Browse Source
This commit is contained in:
pollfly
2023-05-31 12:37:20 +03:00
committed by GitHub
parent 8d46338267
commit 7cd3a46930
2 changed files with 31 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
docs/webapp/webapp_reports.md
View File

@@ -94,7 +94,7 @@ resources will be displayed. See [Dynamic Queries](#dynamic-queries) below.
* `single` (single-scalar values table)
* `sample` (debug sample)
* `parcoords` (hyperparameter comparison plots) - for this option, you need to also specify the following parameters:
* `metrics` - Unique metric/variant ID formatted like `metric_id.variant_id` (find with your browser's inspect. See note [below](#event_id))
* `metrics` - Unique metric/variant ID formatted like `metric_id.variant_id` (see note [below](#event_id))
* `variants` - Parameters to include in the plot (write in following format `<section_name>.<parameter_1>&<section_name>.<parameter_2>`)
* `value_type` - Specify which metric values to use. The options are:
* `min_value`
@@ -144,22 +144,18 @@ used.
```
A list of tags that the experiment should contain is specified in the `tags` argument. You can also specify tags that
exclude experiments. See tag filter syntax examples [here](../clearml_sdk/task_sdk.md#tag-filters).
* Request the `training/accuracy` scalar plot of the 5 experiments with the best accuracy scores
* Request the `training/accuracy` scalar plot of the 5 experiments with the best accuracy scores (see Metric/Variant IDs note [below](#event_id)):
```
src="<web_server>?objectType=task&type=scalar&metrics=training&variants=accuracy&project=4043a1657f374e9298649c6ba72ad233&page_size=5&page=0&order_by[]=-last_metrics.<metric_event_id>.<variant_event_id>.value"
src="<web_server>/widgets/?objectType=task&type=scalar&metrics=training&variants=accuracy&project=4043a1657f374e9298649c6ba72ad233&page_size=5&page=0&order_by[]=-last_metrics.<metric_id>.<variant_id>.value"
```
<a id="event_id"></a>
:::tip Event IDs
The `tasks.get_all` and `models.get_all` API calls' parameters sometimes need event IDs, instead of names. To find event IDs:
1. Go to the relevant Experiments/Model table > Open the **Developer Tools** window (inspect) > click **Network**.
1. Execute the action you want the embed code to do (e.g. sort by update time, sort by accuracy).
1. Click on the API call `tasks.get_all_ex`/`models.get_all_ex` that appears in the **Network** tab.
1. Click on the **Payload** panel.
1. Click on the relevant parameter to see the relevant event's ID. For example, if you sorted by experiment accuracy,
you will see the metric's event ID under the `order_by` parameter.
:::tip Metric/Variant IDs
Metric names need to be MD5 encoded for parallel coordinate plots and for ordering query results by metric
performance. You can encode the strings in Python with `hashlib.md5(str("<metric_string>").encode("utf-8")).hexdigest()`,
and use the returned MD5 hash in your query.
:::