Command line¶

erdr (or python -m erdr). Global options: --version, -v/-vv (more log output). Errors are reported in one line with exit code 2; failed validations exit with 1.

On a terminal the output uses the project's colours (ADR-019): column headers by what they hold, the classes K1, K2 and the samples D, F in their colours, ⚠ lines orange, ✓ green and ✗ red. Redirected output has no colour codes, and NO_COLOR=1 turns colour off.

Command	What it does
`erdr run [DATASET] [-c FILE]`	runs ERDR; prints Step 1, every iteration, D, F, the routes and the cases (`--brief` for the result only, `--all` for every row)
`erdr spheres [DATASET] [-s OBJECT \\| -t ITERATION] [-f FIGURE]`	Steps 1–5 as hyperspheres: every object's spheres, one object's, or one iteration's; saves the figure
`erdr readings [DATASET] [--both-z]`	every combination of the readings A, B, C (and the own-class z)
`erdr sweep [DATASET] [--k 3-41] [-r 111 -r 121:own]`	\|D\| and its class composition for many k and readings
`erdr validate [--against WORKBOOK] [--fast] [--details]`	the package against every computed sheet of the ERDR workbook
`erdr runs` · `erdr export [RUN] -f FORMATS`	list run folders; repeat a saved run and export it (`--readings`, `--sweep`)
`erdr data list` · `check [DATASET…]` · `info [DATASET] [-c FILE]`	the dataset folder by type, its checks (type folders, file conventions), one dataset's summary (type, metric, memory)
`erdr config show` · `init FILE` · `check FILE`	presets and configuration files
`erdr plugins`	normalizers, metrics, dataset formats, export formats, built-in datasets
`erdr gui [DATASET] [-c FILE]`	the desktop application

erdr run¶

erdr run                                         # the workbook's experiment
erdr run heart-disease-270 --brief -k 5
erdr run -p closest heart-disease-270            # preset readings
erdr run -r 211 --z-scope own-class -k 1 --k-rule warn
erdr run --metric minkowski --normalizer robust  # other Step 0 plug-ins (defaults for their parameters)
erdr run uzb-kor-100 --brief                     # a nominal dataset: metric hamming (auto)
erdr run -c configs\heart-disease-270.yaml --save --export all

--save writes a run folder (runs/<time>_<config hash>/: manifest, configuration, the dataset, Step 1 and the trace, the training and validation samples); --export FORMATS writes the chosen formats into <run>/export or --out. Overrides are applied to the configuration file's values and validated.

erdr data¶

> erdr data list
datasets folder: D:\PhD\CODING\erdr\datasets
  default                    default                            (7, 13, 2)       .dat, .csv
Quantitative/  every feature quantitative · metric (auto): manhattan
  cancer-589                 Cancer (589, 44, 2)                (589, 44, 2)     .dat, .csv
  …
  tcga-luad-589              TCGA_LUAD (589, 57772, 2)          (589, 57772, 2)  .dat, .csv  401 MB
Nominal/  every feature nominal · metric (auto): hamming
  …
Heterogeneous/  quantitative and nominal features · metric (auto): zhuravlev
  …

> erdr data info uzb-kor-100
Uzb-Kor (100, 24, 2)  (100, 24, 2)
  type          nominal — every feature nominal
  classes       K1 = 1: 50 · K2 = 2: 50
  quantitative  0: —
  nominal       24: x₁, x₂, …, x₂₄
  metric        hamming (auto · nominal dataset)
  suited        zhuravlev, weighted-zhuravlev, heom, gower, hamming
  memory        about 1 MB for a run
  permitted k   3, 5, …, 97  (odd k from 3 to 2·min|Kᵢ| − 3 = 97)

erdr data check checks every dataset of the folder (or only the ones named): header against name, .dat against .csv, feature types against the type folder, the author's file conventions line by line; it exits with 1 when a problem is found. erdr data info -c FILE shows the metric that configuration runs on the dataset, or ⚠ why it cannot. erdr run warns before a run that needs more memory than is free (TCGA_LUAD: about 790 MB).

erdr spheres¶

erdr spheres                              # every object's enemy sphere and k-sphere, the checks,
                                          # and the spheres around Sᵢ in every iteration
erdr spheres --object S2 -f s2.svg        # the two spheres of S₂, and their diagram
erdr spheres -t 2 -f iteration-2.png      # iteration 2: every object's sphere, L, the moves
erdr spheres heart-disease-270 -p closest -t 1

Without --object or --iteration, --figure saves the spheres of every object (up to 12 panels). The settings options are those of erdr run; a non-literal reading is flagged with ⚠.

erdr readings and erdr sweep¶

$ erdr readings --both-z
…
6 different training samples from 16 readings

$ erdr sweep heart-disease-270 --k 3,5,7,13,17 -r 111 -r 121:own

A sweep without --k covers every permitted k; ranges like 3-41 step over odd k. Readings are codes of A, B, C; :own takes z over the own class. Pairs for which z(Sᵢ, k) does not exist are skipped.

erdr validate¶

$ erdr validate
ERDR algorithm [Heart-Disease (7, 13, 2)].xlsx  ·  tolerance 1e-09
  Parameters                     5 checks  ✓
  …
  Hyperspheres (Step 1)        235 checks  ✓
  …
  Validation                    53 checks  ✓
  2927 checks
validation passed

The workbook is found under resources/Experiments/ from the working directory upwards; --fast skips the 270-object runs of Article Table 1 Check.