summaryrefslogtreecommitdiff
path: root/lib
AgeCommit message (Collapse)AuthorLines
2024-02-09from_json: Support models without argCount and valueBirte Kristina Friesel-3/+5
2024-02-08json export: leave out undefined valueError / functionError itemsBirte Kristina Friesel-7/+4
2024-02-08model export/import: rename paramDecisionValue to thresholdBirte Kristina Friesel-11/+9
2024-02-06Add ScalarSplitFunction support for manual scalar splits in RMTBirte Kristina Friesel-10/+135
2024-02-02... add cli.py with --load-json definitionBirte Kristina Friesel-0/+6
2024-02-02ModelAttribute.from_json: Set mean/median from static modelBirte Kristina Friesel-0/+2
2024-02-02AnalyticModel: Add from_json constructorBirte Kristina Friesel-6/+32
2024-02-02Model: do not store distinct_values; they're part of stats anywaysBirte Kristina Friesel-10/+6
2024-02-01plotter: remove redundant color specifierBirte Kristina Friesel-1/+1
2024-01-29LMT: Fix DFATOOL_LMT_MIN_SAMPLES_LEAF default valueBirte Kristina Friesel-1/+3
2024-01-26plot_param: Handle low numeric rangesBirte Kristina Friesel-0/+2
2024-01-25Use LMT algorithm defaults for max depth and min samples leaf.Birte Kristina Friesel-3/+3
max bins remains at 120 (≠25), but that one should only affect run-time and not accuracy/complexity
2024-01-25XGB: switch to XGBoost defaultsBirte Kristina Friesel-9/+68
2024-01-25add an optional XV progress barBirte Kristina Friesel-0/+19
2024-01-25LMT: Tailor hyper-parameters towards higher accuracy (and longer training time)Birte Kristina Friesel-1/+54
Also, allow users to override them
2024-01-25add hyper-parameter export for CART and LMTBirte Kristina Friesel-0/+20
2024-01-24Use separate /hyper/ hierarchy for hyper-parameter exportBirte Kristina Friesel-13/+20
2024-01-23--function-override: Use : for splitting to support spaces in key/attr namesBirte Kristina Friesel-1/+1
Also, document it in doc/modeling-methods
2024-01-23Allow --skip-param-stats to be used in FIT_FOL modeBirte Kristina Friesel-5/+22
2024-01-22Support LUT model independent of --skip-param-statsBirte Kristina Friesel-6/+7
2024-01-22Add --skip-param-stats support to analyze-{archive,log}Birte Kristina Friesel-0/+8
2024-01-19RMT: Set DFATOOL_RMT_WEIGHTED_AVG=1 for weighted average in queriesBirte Kristina Friesel-0/+6
2024-01-19Store n_samples in all relevant ModelFunction instancesBirte Kristina Friesel-7/+14
2024-01-19functions: explicitly mention _xgb_ hyper-parameters in dref exportBirte Kristina Friesel-7/+7
2024-01-19dataref export: include XGB hyper-parametersBirte Kristina Friesel-1/+15
2024-01-19optionally store n_samples in ModelFunctionBirte Kristina Friesel-17/+29
2024-01-19Remove unused ModelAttribute.split code; splits are handled via SplitFunctionBirte Kristina Friesel-80/+34
2024-01-19parameters: nitsBirte Kristina Friesel-4/+6
2024-01-19assertionsBirte Kristina Friesel-1/+4
2024-01-18Rename DFATOOL_{'' → ULS_}SKIP_CODEPENDENT_CHECKBirte Kristina Friesel-1/+1
2024-01-18kconfig: ignore strings by default (implicit DFATOOL_KCONF_IGNORE_STRING=1)Birte Kristina Friesel-1/+1
2024-01-18build_dtree: do not ignore codependent parametersBirte Kristina Friesel-2/+7
Ignore non-numeric function arguments instead. Strings arguments are often paired with a length indicater and we are only interested in the latter.
2024-01-18parameters: ModelAttribute: remane .ignore_param to .ignore_codependent_paramBirte Kristina Friesel-12/+14
This should help avoid confusion.
2024-01-17paramfit: use sum(ssr) rather than mean(rmsd) for best-fit selectionBirte Kristina Friesel-9/+21
2024-01-17cli: fix energy_Pt model checkBirte Kristina Friesel-2/+2
2024-01-17_try_fits: remove unused results_by_paramBirte Kristina Friesel-10/+0
2024-01-17Disable DFATOOL_DTREE_IGNORE_IRRELEVANT_PARAMS by defaultBirte Kristina Friesel-2/+2
It is not working properly in 95% of cases
2024-01-17dtree irrelevant param detection nits; still not working properlyBirte Kristina Friesel-5/+5
2024-01-17Stop splitting if there is no split that reduces lossBirte Kristina Friesel-1/+3
2024-01-16build_dtree: add another assertion for good measureBirte Kristina Friesel-4/+2
2024-01-16model: Correctly set threshold from compute_statsBirte Kristina Friesel-1/+1
2024-01-16model: ModelAttribute is defined in .parameters, not .automataBirte Kristina Friesel-1/+2
2024-01-16Add LMT support for --export-json / ModelAttribute.to_jsonBirte Kristina Friesel-1/+5
2024-01-16plotter: remove useless 'red' colour specifierBirte Kristina Friesel-1/+1
2024-01-16Implement --show-model=param for LMTBirte Kristina Friesel-0/+55
2024-01-15improve(?) loss_ignore_scalar implementation: more opportunities for FFSBirte Kristina Friesel-6/+44
2024-01-15dtraa loss_ignore_scalar: calculate unsuitable params for each nodeBirte Kristina Friesel-2/+20
The global list may no longer be helpful after the observations have been partitioned by parent nodes
2024-01-15dtree: replace condition by assertionBirte Kristina Friesel-11/+11
2024-01-15decrease non-debug verbosityBirte Kristina Friesel-6/+2
2024-01-15PTAModel.asses: support transitions with partial dataBirte Kristina Friesel-1/+1
This can happen when --filter-observation is in use