eh train
Train a new eigenspace model from a corpus directory.
Usage
eh train CORPUS_DIR --language LANG -o OUTPUT [OPTIONS]
Arguments
| Argument |
Description |
CORPUS_DIR |
Root directory of the code corpus (positional, required) |
Options
| Option |
Default |
Description |
-o, --output PATH |
required |
Output .npz model path |
--language LANG |
required |
Target language key (e.g., python, go, multi) |
--corpus-class {A,B} |
A |
Corpus quality class |
--n-components INT |
(auto) |
Explicit number of principal components |
--variance-threshold FLOAT |
0.90 |
Min cumulative explained variance for auto-select |
--version TEXT |
(package version) |
Model version string |
--force |
off |
Overwrite existing output file |
Example
# Sync a corpus from manifest
eh corpus sync corpora/my-corpus.toml corpus/my-corpus
# Train
eh train corpus/my-corpus --language python -o models/my-model.npz
# Verify
eh inspect models/my-model.npz
Corpus classes
| Class |
Description |
| A |
Single-language — curated, reviewed, exemplary code in one language |
| B |
Cross-language pattern — structural patterns across multiple languages |