Class: LLMExperiment::CLI::MetricsCommand
- Inherits:
-
Command
- Object
- Command
- LLMExperiment::CLI::MetricsCommand
show all
- Defined in:
- lib/llm_experiment/cli/metrics_command.rb
Constant Summary
collapse
- NAME =
"metrics"
- SUMMARY =
"medians and exact Mann-Whitney U, never pooled"
Instance Method Summary
collapse
Methods inherited from Command
#call, #initialize, summary
Instance Method Details
#add_options(parser) ⇒ Object
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
|
# File 'lib/llm_experiment/cli/metrics_command.rb', line 9
def add_options(parser)
parser.banner = <<~BANNER
Usage: llmx metrics [options]
Reads every metrics.json written by `llmx parse` and reports one cell
per app and agent. Nothing is pooled across those: two agents do not
count tool calls the same way, and two apps are different sizes of
haystack, so a number averaged over them measures nothing.
p-values are exact. With a handful of tasks per cell the normal
approximation to the Mann-Whitney U is not trustworthy, so every
split of the pooled values is enumerated instead.
BANNER
end
|
#run(_argv) ⇒ Object
25
26
27
|
# File 'lib/llm_experiment/cli/metrics_command.rb', line 25
def run(_argv)
MetricsReport.new(experiment: experiment).render
end
|