Expert mode

Resume

The profiling expert mode is intended for users who want to measure the performance of Pandora2D on their personal computer. In the output folder, they can obtain a number of charts that calculate averages and other metrics for each step throughout the executions.

How to profile more functions ?

This option requires the user to be familiar with the pandora2d code.

First, when they activate the expert_mode key in the configuration, they have access by default to performance information related to each stage of the state machine. All data is stored in the code in a pandas.DataFrame and locally in a CSV file, then presented as a graph in a PDF file.

If the user wants to analyze the performance of another function, they can add the decorator @mem_time_profile_profile(name=”Function name”) above that function. If they want to obtain more metrics, they need to add them to the “metrics_list” in the profiling.py file.

The graphs are handled by the generate_figure function.

Note

Profiling certain functions can significantly increase execution times.

Parameters and configuration:

Expert mode profiling section is composed of the following keys:

Table 10 Expert mode section

Name	Description	Type	Default value	Required
folder_name	path where to save profiling informations	string	None	Yes

Example

{
    "input":
    {
        // inputs' content
    }
    ,
    "pipeline" :
    {
        // pipeline content
    },
    "expert_mode":
    {
        "profiling":
        {
            "folder_name": "profiling_output"
    },
    "output": {
        "path": "expert_mode_output"
    }
}