unreleased

1.7.1

• improved explore-all, adding percentages as text
• upgraded to plotje 0.10

1.7.0

• docstring update
• use codox / metadoc
• added support for model plotting fn via (ml/plot model)
• added plotting with plotje for model type :fastmath/ols (equivalent to all 6 plots of R function 'plot.lm' with default params)
• added explore/explore-all fn to plot all variables of a dataset, using plotje
• deprecated ns scicloj.metamorph.ml.viz

1.6.2

• add inference targets to r-model-matrix

1.6.1

• fixed classloader for r-model-matrix

1.6.0

• allow xjboost4j DMatrix for prediction
• added ml/optimize-hyperparameter
• deprecated ml/evaluate-pipelines
• deprecated ns scicloj.metamorph.ml.metrics,scicloj.metamorph.ml.loss
• added support for :run!,prun!, :pprun! in evaluate-pipelines and optimize-hyperparameter
• added random forrest implementation as model-type: :metamorph.ml/random-forest

1.5.1

• added r-model-matrix/lm helper for one-line linear models
• added ns documentation to most ns

1.5.0

• design-matrix fn expands keyword to identity #41
• design-matrix fn expands no-column name to auto-column name #42
• added support in ns r-model-matrix for model matrix creation using "R-formula syntax" backed up by R itself, with 3 implementations to choose using opencpu,rejine,clojisr

1.4.0

• validate predit-fn results
• new ns scicloj.metamorph.ml.column-metric with metric-fns operating on datasets, not on seq-of-numbers
• TMD's of rdatasets get doc-url as dataset name, ready for clicking #36
• added model names to fastmath/ols #24
• add support for fastmath GLM model 46
• allow ml.dmlc.xgboost4j.java.DMatrix as input to train/test

1.3.2

• removed clj-memory-meter from src-code

1.3.1

• improved docstring with Claude Code
• use TMD 8.016

1.3.0

• upgraded deps

• added missing Java classes from former tmd-smile support

• more clear error message for accuray of categorical maps

• made some internal functions private

  • scicloj.metamorph.ml.rdatasets

    • clean-R-relevant
    • doc-url->md
    • _fetch-dataset

  • scicloj.metamorph.ml.text

    • make-mmap-container
    • make-container
    • ->column--concat-buffers
    • ->column--coalesce-blocks

  • scicloj.metamorph.ml.viz

    • apply-xform-kvs

  • scicloj.metamorph.ml

    • score

• added 3 train/test cache implementatins to scicloj.metamorph.ml.cache

1.2.1

• upgrade deps
• fixed @37

1.2

• removed most toy datasets and forward existing fns to rdatasets/*
  • attention: column nams of some 'toydatasets' have changed, now in kebab-keyword style
• additional validation of consitency of predict / cat maps

1.1.1

• cleaned column names from "." in rdatasets
• better initial doc string for rdatasets functions
• new fn 'dataset-descriptions->doc-strings!' to attach dataset description as docsting to all rdadasets functions

1.1

• added access fn to all datasets from https://vincentarelbundock.github.io/Rdatasets/articles/data.html

1.0

• use malli to describe model options and validate them in ml/train
• renamed :other-metrices -> :other-metrics

0.12

• fixes #30 - dummy classifier does not predict by majority #30
• added dummy regression model
• improved design-matrix feature . Breaking ! columns need to be refered know by "precise name" (string, symbol, keyword)
• made model :fastmath/ols frezzable by nippy

0.11.1

• allow parameters for :fastmath/ols (fixes #27)
• added optional caching for train / predict
• added new evaluation-handler: metrics-and-model-keep-fn
• added option :ppmap with :ppmap-grain-size 10 to ml/eval-pipelines
• added more evaluation-handler fns suitable for model-spec search
• added :probability-distributin to ml/eval-pipelines result
• breaking: move all eval handler to ns scicloj.metamorph.ml.evaluation-handler

0.10.4

• added :target-datatypes in train result and clarified expected 'shape' of prediction

0.10.3

• re-added data.json dependency

0.10.2

• re-added data.csv dependency

0.10.1

• fixed cljdoc build issue

0.10.0

• exclude most hanami deps as we don't need them
• added support for tidy-text and
• added read/write support for TMD<->libsvm files

0.9.0

• added linear models from fastmath
• added tidy output validation
• added support for design-matrix

0.8.2

-fixed metric bug

0.8.1

-fixed bug in ml/tidy

0.8.0

• upgraded deps
• added suport for glance,augment,tidy

0.7.10

• added missing deps

0.7.9

• fixed and documented confusion-map->ds
• added confusin matrix plot
• added generic handling of loglikelyhood
• added mtcars data
• added aic and bic

0.7.8

• fixed default colors of error bands

0.7.7

• fixed test/train color assignments
• adde more docu

0.7.6

• fixed metric.clj filename for ClojureDoc generation

0.7.5

• removed WIP files

0.7.4

• fixed Clojars links and ClojureDoc
• refatored learnining curve

0.7.3

• added dummy classifier
• add checks for matching categorical maps between train and predict
• added titanic toy data
• added ggplot toydaya
• added visualisation for learning curve

0.7.2

• fixed verify ns

0.7.1

• fixed 'verify' ns

0.7

• using tablecloth 7.0
• added uid for grouping of splits
• added learning curve
• added AUC metric (thanks @Prometheus77)

0.6.4

• fix metric calculation in evaluate-pipelines
  • use reverse mappings

0.6.3

• added :as opts to methods signature

0.6.2

• added ensembles

0.5.0

• unified dissoc-in options and handler-fn option

0.4.1

• adding support for unsupervised learning

0.4.0

• changed result of `evaluate-pipelines`` to be more consistent
• added Malli schema to several functios
• added support for experiment tracking on disk
• added experimenta / example code of using nippy to persist evaluation results

0.3.0-beta6

• fixed toy data

0.3.0-beta4

• added sonar data

0.3.0-beta3

• added std-scaler
• added min-max scaler
• added toy data

0.3.0-beta2

0.3.0-beta1 - 06.04.2021

• big change, tech.ml (core) was moved into here