opam-version: "2.0"
authors: "Francois Berenger"
maintainer: "unixjunkie@sdf.org"
homepage: "https://github.com/UnixJunkie/linwrap"
bug-reports: "https://github.com/UnixJunkie/linwrap/issues"
dev-repo: "git+https://github.com/UnixJunkie/linwrap.git"
license: "BSD-3-Clause"
build: ["dune" "build" "-p" name "-j" jobs]
depends: [
  "base-unix"
  "batteries"
  "cpm"
  "dolog" {>= "4.0.0" & < "5.0.0"}
  "dune" {>= "1.10"}
  "minicli"
  "parany" {>= "6.0.0" & < "10.0.0"}
  "conf-liblinear-tools"
]
synopsis: "Wrapper around liblinear-tools"
description: """
Only L2-regularized logistic regression is supported currently.
Each model is trained on balanced bootstraps from the training set
(one bootstrap for the positive class, one for the negative class).
The size of the bootstrap is the size of the smallest (under-represented)
class. Bagging is supported and allows to obtain better models.

usage: linwrap
  -i <filename>: training set or DB to screen
  [-o <filename>]: predictions output file
  [-np <int>]: ncores
  [-c <float>]: fix C
  [-w <float>]: fix w1
  [-k <int>]: number of bags for bagging (default=off)
  [-n <int>]: folds of cross validation
  [--seed <int>]: fix random seed
  [-p <float>]: training set portion (in [0.0:1.0])
  [{-l|--load} <filename>]: prod. mode; use trained models
  [{-s|--save} <filename>]: train. mode; save trained models
  [--scan-c]: scan for best C
  [--scan-w]: scan weight to counter class imbalance
  [--scan-k]: scan number of bags (advice: optim. k rather than w)
"""
url {
  src: "https://github.com/UnixJunkie/linwrap/archive/v0.0.1.tar.gz"
  checksum: [
    "sha256=460bbd11012e1d081497a03a42737562aa3333f0ec7913d4aea601ad866afb9f"
    "md5=4456917240f47526a681d1230d53ebf8"
  ]
}