πͺ spaCy Project: Detecting Commands, Arguments and Options in CLI Help messages (Named Entity Recognition)
This project deals with the backbone model that powers helpner.
The project.yml
defines the data assets required by the
project, as well as the available commands and workflows. For details, see the
spaCy projects documentation.
The following commands are defined by the project. They
can be executed using spacy project run [name]
.
Commands are only re-run if their inputs have changed.
Command | Description |
---|---|
create-dataset |
Create a dataset using cli-help-maker |
split |
Split the .jsonl dataset file contents in train/dev |
convert |
Convert .jsonl files to .spacy format |
train |
Train a named entity recognition model on cli help messages |
evaluate |
Evaluate the model and export metrics |
package |
Package the trained model so it can be installed |
release |
WIP: Upload the packaged model to github releases |
readme |
Auto-generate README via spacy. |
The following workflows are defined by the project. They
can be executed using spacy project run [name]
and will run the specified commands in order. Commands are only re-run if their
inputs have changed.
Workflow | Steps |
---|---|
all |
split β convert β train β evaluate |
The following assets are defined by the project. They can
be fetched by running spacy project assets
in the project directory.
File | Source | Description |
---|---|---|
assets/dataset.jsonl |
Local | JSONL-formatted training data obtained from cli-help-maker |
The following metrics are obtained from the spacy benchmark accuracy
command:
Type | Score |
---|---|
token_acc | 100.00 |
token_p | 100.00 |
token_r | 100.00 |
token_f | 100.00 |
speed | 3228586.60 |
P | R | F | |
---|---|---|---|
CMD | 98.25 | 99.96 | 99.10 |
ARG | 94.79 | 89.97 | 92.32 |
OPT | 98.88 | 98.96 | 98.92 |
The naming conventions and versioning scheme follows the conventions used in spacy-models (in spirit), but so much simplified.
See for the model-versioning section, replace spaCy
with helpner
, and that's the idea.
The trained models are uploaded as release assets in GitHub, so they can be installed via pip, and loaded in spacy via spacy.load
.
A total of 1000 help messages were used.
mean | std | min | 25% | 50% | 75% | max | |
---|---|---|---|---|---|---|---|
indent_spaces | 3.012 | 1.00043 | 2 | 2 | 4 | 4 | 4 |
total_width | 89.344 | 16.2353 | 70 | 78 | 80 | 100 | 120 |
prob_name_capitalized | 0.503812 | 0.287329 | 0.00189856 | 0.254035 | 0.523925 | 0.74786 | 0.998479 |
description_before | 0.472 | 0.499465 | 0 | 0 | 0 | 1 | 1 |
description_after | 0.498 | 0.500246 | 0 | 0 | 0 | 1 | 1 |
program_description_prob | 0.490548 | 0.28722 | 0.00167024 | 0.242151 | 0.477199 | 0.74083 | 0.99586 |
usage_section | 0.516 | 0.499994 | 0 | 0 | 1 | 1 | 1 |
usage_pattern_capitalized | 0.484 | 0.499994 | 0 | 0 | 0 | 1 | 1 |
commands_section | 0.815 | 0.388492 | 0 | 1 | 1 | 1 | 1 |
commands_header | 1 | 0 | 1 | 1 | 1 | 1 | 1 |
commands_capitalized | 0.49958 | 0.284029 | 0.0022922 | 0.25243 | 0.501338 | 0.737582 | 0.999426 |
commands_documented_prob | 0.505911 | 0.283441 | 0.00123438 | 0.27576 | 0.509805 | 0.742631 | 0.99995 |
arguments_section | 0.492 | 0.500186 | 0 | 0 | 0 | 1 | 1 |
arguments_header | 1 | 0 | 1 | 1 | 1 | 1 | 1 |
argument_repeated | 0.497881 | 0.285127 | 0.00101773 | 0.270007 | 0.488525 | 0.746068 | 0.999867 |
argument_documented_prob | 0.486583 | 0.289539 | 0.000206963 | 0.228881 | 0.472501 | 0.738158 | 0.998935 |
arguments_pattern_capitalized | 0.512 | 0.500106 | 0 | 0 | 1 | 1 | 1 |
argument_capitalized_prob | 0.492618 | 0.287912 | 0.0060436 | 0.246821 | 0.489228 | 0.747413 | 0.999545 |
argument_optional_prob | 0.505243 | 0.295162 | 7.63351e-05 | 0.235492 | 0.507031 | 0.770653 | 0.999283 |
argument_any_number_prob | 0.503465 | 0.295311 | 0.000182664 | 0.247908 | 0.511808 | 0.770671 | 0.999878 |
argument_nested_prob | 0.5016 | 0.290744 | 0.00096628 | 0.244742 | 0.51879 | 0.751403 | 0.998248 |
options_section | 0.506 | 0.500214 | 0 | 0 | 1 | 1 | 1 |
options_header | 1 | 0 | 1 | 1 | 1 | 1 | 1 |
option_documented_prob | 0.507132 | 0.291259 | 0.000381441 | 0.251737 | 0.517262 | 0.761607 | 0.999397 |
options_pattern_capitalized | 0.484 | 0.499994 | 0 | 0 | 0 | 1 | 1 |
options_shortcut | 0.510277 | 0.286645 | 0.00107218 | 0.272586 | 0.521389 | 0.750301 | 0.998364 |
options_shortcut_capitalized_prob | 0.499165 | 0.289056 | 0.000546323 | 0.253128 | 0.495272 | 0.760074 | 0.999867 |
options_shortcut_all_caps | 0.524 | 0.499674 | 0 | 0 | 1 | 1 | 1 |
exclusive_group_optional_prob | 0.50469 | 0.290709 | 0.00185361 | 0.254987 | 0.514023 | 0.755868 | 0.995541 |
options_mutually_exclusive_prob | 0.209 | 0.406798 | 0 | 0 | 0 | 0 | 1 |
option_set_size | 1.977 | 1.42496 | 0 | 1 | 2 | 3 | 4 |
option_set_size_prob | 0.506309 | 0.28327 | 0.000160774 | 0.26604 | 0.504979 | 0.742098 | 0.99916 |
number_of_commands | 4.52 | 3.4571 | 1 | 2 | 4 | 6 | 15 |
number_of_arguments | 4.63 | 3.74688 | 1 | 2 | 3 | 7 | 15 |
number_of_options | 4.543 | 3.72505 | 1 | 1 | 3 | 7 | 15 |
exclusive_programs | 1.573 | 1.00682 | 1 | 1 | 1 | 2 | 8 |