Changelog

[0.2.4] - 2026-05-05

Features

• Added _ManagerMultiPropertyProxy.fit_per(group_property, model, ...) — fits one regression per unique value of a discrete grouping property and returns a {label: RegressionFit} dict. Mirrors Crossplot.add_regression_per at the proxy level so users can produce per-group fits for tables and reports without going through a plot. Each group gets a fresh deep-copy of model so locked params and degree settings persist across groups; subsets smaller than min_samples (default 5) are skipped with a warning. Also available through manager.filter(where=...) views via _ValueFilteringProxy.fit_per. (#proxy-fit-per)

• Added pool=True to _ManagerPropertyProxy.stats(). When set, returns a DataFrame with one row per filter-group combination instead of one row per (well, group) — mean, std, min, max, and percentiles are computed across the pooled long-form data, so dispersion statistics are correct (not the mean of per-well point estimates). The new samples method counts data points per group. Percentile methods accept both long form (percentile_10) and short form (p10). (#stats-pool)

Bug Fixes

• Fixed: manager.filter(where={"Zone": "Reservoir"}).PHIE.data() now actually filters by Zone. Previously the post-filter on _ValueFilteringProxy.data() silently no-op’d when the filter column was missing from the underlying proxy’s chain — the column wasn’t in the DataFrame and the col in df.columns guard skipped. The wrapper now auto-adds proxy.filter(col) for any where= key whose column the proxy doesn’t already emit, so the where clause is applied consistently across view.PROP.data(), view.properties([...]).data(), and view.PROP_PAIR.fit() paths. (#where-actually-filters)

[0.2.3] - 2026-05-05

Features

• Crossplot.add_regression gained equation_format= and decimals= arguments that propagate to the legend label. equation_format="petrel" renders an exponential fit as pow(10, c1*x + c0) — copy-paste ready for Petrel calculator workflows. decimals overrides the default 4-decimal precision (e.g. decimals=2 for shorter legend text). Default behavior is unchanged for back-compat. Closes Story 4 on the Crossplot side. (#add-regression-equation-format)

• Crossplot.add_regression gained legend_loc= (matplotlib loc string or coords) and legend_decimals= (alias for decimals=) arguments. legend_loc= overrides the auto-placement algorithm for the regression legend — pass "upper left" once and stop hand-rolling mlegend.Legend reconstructions. The latest legend_loc= value across multiple add_regression calls is used since the regression legend is rebuilt as a single block. (#add-regression-legend-loc)

• Crossplot.add_regression(where=...) now reuses Property.colors for the regression line when where is a single-key, single-value filter on a discrete property whose palette is set. Set the palette once on the manager, fit per-facies regressions, and the lines pick up the matching color automatically. Explicit line_color= still overrides; where filters spanning multiple values fall back to the existing "red" default. (#add-regression-palette-reuse)

• Added Crossplot.add_regression_per(group_property, regression_type, ...) — fits one regression per unique value of a discrete grouping property and adds them all in a single call. Names default to the property labels; line colors come from Property.colors when set; all other add_regression kwargs (equation_format, legend_loc, min_samples, …) are forwarded. Replaces the typical for loop that enumerates facies/zones by hand. (#add-regression-per)

• Crossplot.add_regression gained where= and min_samples= arguments. where accepts a dict mapping public column names (the same as x, y, color, shape, size on the Crossplot) to allowed values, or a callable returning a boolean mask over the prepared data. The fit uses only the matching rows; R² and equation reflect that subset. If the filtered subset has fewer than min_samples (default 5) points, a warning is emitted and the regression is skipped. Users no longer need to monkey-patch _data to fit per-zone or per-facies trends. (#add-regression-where)

• Added the Artifact protocol (logsuite.Artifact) — a base class for self-contained result objects that consumers (Crossplot, Table, …) accept via .add(artifact). Subclasses implement _render_in_<consumer>(...) for the consumers they support; unsupported consumers raise TypeError. (#artifact-protocol)

• Crossplot now accepts a ManagerView or WellDataManager as its first argument, matching the architectural direction of consumers reading from the manager substrate. The recommended call shape becomes Crossplot(manager.filter(wells=[...]), x="PHIE", y="PERM") — single source of truth for which wells are included. Existing Crossplot(well_or_list, ...) calls continue to work unchanged. (#crossplot-accepts-view)

• Added Crossplot.add(artifact) — single composition entry point for any object implementing _render_in_crossplot(ax). The plot is rendered first if needed, then the artifact draws onto the axis and the legend is refreshed. Coexists with the existing kwargs/methods; the artifact protocol is the path forward for new visualization extensions. (#crossplot-add)

• Added Crossplot.column_for(public_name) — returns the prepared-data column name for a property bound to x, y, color, shape, or size on the crossplot, or None if no binding matches. Useful when writing a callable mask for add_regression(where=...): instead of hardcoding df["color_val"], ask the crossplot which column the property maps to. The internal logic (previously _resolve_where_column) is now exposed as a stable public method. (#crossplot-column-for)

• Crossplot constructor now accepts equation_format= and decimals= as defaults that propagate to every add_regression call. Set Crossplot(equation_format="petrel", decimals=3) once and all regression legends render in Petrel calculator syntax with three-decimal precision. Per-call kwargs on add_regression still override. (#crossplot-equation-format-default)

• Added Crossplot.add_table_panel(df, position="bottom"|"right", title=None, formatters=None, table_fraction=0.30). Attaches a DataFrame as a matplotlib-rendered table inside the same figure; the figure grows along the panel axis so the scatter is not squished. xplot.save("file.svg") then produces the combined deliverable in one file — the standard SCAL/DG3 shape.

  Defaults handle the common deliverable cases out of the box: NaN values render as "N/A", MultiIndex column labels flatten via " | ", and MultiIndex row labels visually merge repeated outer levels (the outer cell is blanked when it matches the previous row, mimicking the merged-cell look without needing matplotlib cell merging). formatters={"PHIE": ".4f"} accepts callables or Python format specs per column.

  Implementation lives in a new module logsuite.visualization.table_panel whose functions are reusable by future composite/Template consumers. Closes Story 7 and friction #11. (#crossplot-table-panel)

• Added flat_columns= argument to _ManagerPropertyProxy.stats(return_df=True). When True, grouping columns use the actual filter property names (Zone, NTG, …) instead of the generic Group/Group1/Group2. Default remains False for backward compatibility; this is expected to flip in a later major version. (#flat-columns)

• WellDataManager and ManagerView now support dict-like access: manager["Well_A"] returns the Well by original name, "Well_A" in manager checks membership, len(manager) reports the well count, and iterating the manager yields Well objects in insertion order. Original names containing characters that aren’t valid Python attributes (slashes, hyphens, spaces) work as keys without sanitisation: manager["12/3-2 B"]. (#manager-dict-access)

• Added WellDataManager.filter(wells=...) returning a ManagerView. The view is a read-only subset of the manager: property proxies, statistics, and .data() calls operate only on the selected wells. Views are immutable; view.filter(wells=...) further narrows. ManagerView is now exported from logsuite. (#manager-view)

• Extended WellDataManager.filter() and ManagerView.filter() with a where= argument for value-subset filtering. where={"Zone": "Reservoir"} post-filters .data() outputs to matching rows; the special key where={"well": [...]} selects wells. Stat methods (mean, stats, sums_avg, …) raise NotImplementedError while value filters are active — use .data() and compute externally, or .filter("Zone") for grouped stats. (#manager-view-where)

• Added .data() method to _ManagerPropertyProxy and _ManagerMultiPropertyProxy. Returns a long-format DataFrame across all wells with a leading well column and one column per active filter (named after the filter property). Replaces the manual loop-and-concat pattern users previously needed for pooled raw data, and gives Crossplot/regression artifacts a stable substrate going forward. (#proxy-data)

• _ManagerPropertyProxy.data() gained warn_missing=True (default). Set warn_missing=False for batch / scripted workflows where wells lacking the property are expected and the per-call UserWarning is noise. Closes Story 6 — pooled data extraction is now fully covered. (#proxy-data-warn-missing)

• Added weighted= argument to _ManagerPropertyProxy.data() and _ManagerMultiPropertyProxy.data(). When True, appends a Weight column with the per-row depth interval (computed via compute_intervals, edge-corrected per well). Users can replicate stats()-style depth-weighted percentiles externally with e.g. np.average(df["PHIE"], weights=df["Weight"]). (#proxy-data-weighted)

• Added .fit(model, ...) on multi-property proxies. Pools data across the (sub-)view, fits the regression with the first property as x and the second as y, and returns a RegressionFit. Works through manager.filter(where=...) so users can fit on subsets in one expression: manager.filter(where={"Zone": "Reservoir"}).properties(["PHIE", "PERM"]).fit(ExponentialRegression(), name="Reservoir"). (#proxy-fit)

• Added RegressionFit (logsuite.RegressionFit) — a renderable artifact wrapping a fitted regression model. Provides equation(format="natural"|"log10"|"petrel", decimals=4) for Petrel-calculator-syntax exports of exponential fits, configurable decimal precision in legend labels via label(decimals=), and _render_in_crossplot / _render_in_table dispatch. ExponentialRegression natively supports all three formats; other models fall back to natural form for log10/petrel. (#regression-fit)

• Added logsuite.set_quiet(True) to silence library-emitted informational status messages (✓ Set colors for …, ✓ Loaded N properties …). Defaults stay verbose so existing notebooks see the same output, but scripted / batch users can call set_quiet(True) once at startup for clean stdout. Warnings and errors are unaffected. Implemented via a tiny emit_status helper in logsuite.utils; the broadcast prints in WellDataManager.load_properties and the property-proxy setters route through it. (#set-quiet)

• Added visualization/style.py with resolve_discrete_palette — a single resolver for categorical colors that honors Property.colors first and falls back to a default palette only for codes the user has not assigned. Crossplot now consumes this resolver in both single-group and multi-group plotting paths, matching WellView’s existing behavior. Setting manager.<discrete_property>.colors = {...} now propagates to every visualization consistently. (#style-resolver)

Bug Fixes

• Fixed TypeError: 'NoneType' object is not callable when calling Crossplot.regression(). The constructor parameter regression= was being assigned to self.regression, shadowing the method of the same name. The constructor now stores it on self._initial_regression internally; the public regression= kwarg name is unchanged. (#regression-method-shadow-fix)

Deprecations

• Passing regression, regression_by_color, regression_by_group, regression_by_color_and_shape, or regression_by_shape_and_color to the Crossplot constructor is now deprecated. They will be removed in a future minor release. Use crossplot.add_regression(kind, where=...) for subsetted fits, or build a RegressionFit and call crossplot.add(fit) for full control over the artifact. (#deprecate-regression-kwargs)

• Well.Crossplot(), Well.WellView(), and WellDataManager.Crossplot() now emit DeprecationWarning when called. They violate the layered-dependency rule (lower layer importing from higher) and will be removed in a future minor release. Construct visualizations directly instead: Crossplot(manager.filter(wells=[...]), x=..., y=...) or WellView(well, ...). Existing call sites continue to function during the deprecation window. (#deprecate-upward-constructors)

Documentation

• Added ARCHITECTURE.md documenting the layered-dependency constitution that governs every module. Pull requests that violate the layer rule (lower layer importing from higher) should be rejected. (#architecture-doc)

• Added Article VII (Compartmentalization) to the architectural constitution: code is organized into focused modules and classes, with soft size triggers (~600-line modules, ~300-line classes, ~50-line methods) flagging refactor candidates. Several existing files exceed these limits and are scheduled for splitting in M3+. (#article-vii)

• Documentation overhaul: README and Sphinx docs rewritten around use cases.

  • New docs/how-to/ section with six task-focused guides — fitting per-group regressions, setting a discrete-property palette, pooling raw data across wells, exporting Petrel-syntax equations, building a SCAL/DG3 deliverable, and displaying single-well log plots.

  • docs/quickstart.md rewritten to a 7-line working deliverable using the post-M3 patterns (Manager substrate, add_regression_per, add_table_panel).

  • docs/index.md adds a “How-to guides” toctree section above the topical user guide.

  • README.md rewritten to lead with what the library lets you stop doing (no per-well loops, no monkey-patching, no matplotlib glue, no equation rewriting), with cross-references to the how-to pages and story_tests/ scripts.

  • User-guide pages loading-data, multi-well, visualization, regression updated to use canonical post-M3 patterns and corrected two real bugs: manager.add_dataframe (which doesn’t exist; it’s manager.load_properties) and manager.well_names (which is manager.wells).

  (#docs-overhaul)

[0.2.0] - 2026-03-05

Breaking Changes

• Package renamed to logSuite: PyPI package is now logsuite (was pylog). Import changes: from logsuite import WellDataManager, Well, Property. Install: pip install logsuite. All internal references, docs, CI/CD, and tests updated.

[0.1.159] - 2026-03-05

Breaking Changes

• Package renamed to pyLog: PyPI package name changed from well-log-toolkit to pylog.

[0.1.158] - 2026-03-05

Documentation

• Sphinx documentation: Full docs site with furo theme, MyST markdown, and autodoc API reference. Includes installation, quickstart, user guide (loading data, wells/properties, statistics, visualization, regression, multi-well), and API reference for all public classes and functions.

• ReadTheDocs configuration: .readthedocs.yaml for automated builds.

• Documentation dependencies: Added [docs] optional dependency group.

• Repo rename: Updated all project URLs from logsuite to logsuite to match PyPI package name.

[0.1.157] - 2026-03-05

Features

• geometric_mean() and harmonic_mean(): New statistical functions for permeability averaging (geometric) and parallel flow averaging (harmonic). Both support depth-weighted and arithmetic modes.

• Property.apply(func): Apply arbitrary functions to property values, returning a new Property on the same depth grid.

• Property.histogram(bins, weighted): Compute histogram of property values, optionally weighted by depth intervals.

• PolynomialExponentialRegression: Now exported at top-level (from logsuite import PolynomialExponentialRegression).

• Basic LAS 3.0 support: Read LAS 3.0 files with ~Log_Definition and ~Log_Data sections (single data section, tab-delimited).

[0.1.156] - 2026-03-05

Performance

• Vectorized compute_zone_intervals(): Replaced Python for-loop with numpy vectorized operations for zone interval calculation, matching the existing compute_intervals() pattern.

• Added performance benchmark test for compute_zone_intervals with 10K-point arrays.

[0.1.155] - 2026-03-05

Features

• Fuzzy name matching: Typos in property/well names now show “Did you mean: …” suggestions using difflib.get_close_matches().

• WellDataManager.validate(): New method to check data integrity across all wells (missing properties, depth monotonicity, length mismatches).

• Proxy stats warnings: manager.PHIE.mean() now warns when wells are silently skipped because they lack the requested property.

• Input validation: Property() now rejects mismatched depth/values lengths and non-monotonic depth arrays. LasFile() rejects non-.las file extensions.

Testing

• Added tests/conftest.py with 10 centralized pytest fixtures.

• Added 4 new test files: test_error_handling.py (20 tests), test_resample_edge_cases.py (6 tests), test_las_export_roundtrip.py (5 tests), test_sums_avg_report.py (8 tests).

• Fixed return True/False → assert/pytest.skip() across 20 test files, eliminating all 62 PytestReturnNotNoneWarning warnings.

• Moved 6 demo/example scripts from tests/ to examples/.

Documentation

• Added See Also cross-references to key methods (filter, sums_avg, resample, get_property, load_las, mean).

[0.1.154] - 2026-03-05

Breaking Changes

• Restructured package into domain-driven subpackages. Direct imports from flat modules (e.g., from logsuite.statistics import mean) must now use subpackage paths (e.g., from logsuite.analysis.statistics import mean). Top-level imports (from logsuite import Well, Property) are unchanged.

Internal Changes

• Created io/ subpackage with las_file.py for LAS file I/O.

• Created core/ subpackage with well.py, property.py, operations.py.

• Created analysis/ subpackage with statistics.py, regression.py, sums_avg.py.

• Created manager/ subpackage, splitting 3500-line manager.py into data_manager.py and proxy.py.

• Created visualization/ subpackage, splitting 5050-line visualization.py into template.py, wellview.py, crossplot.py.

• Extracted _version.py for dynamic version detection.

• Updated all test imports to match new package structure.

[0.1.153] - 2026-03-04

Internal Changes

• Moved test directory from pytest/ to tests/ to match standard Python conventions.

• CI now runs the full pytest suite across Python 3.10-3.13 instead of a smoke import.

• Dropped Python 3.9 support; minimum version is now 3.10.

• Added towncrier-based changelog system for fragment-based release notes.

• Updated project classifier from Alpha to Beta.

• Updated project description to reflect full petrophysical analysis capabilities.

• Fixed issues URL in README (was pointing to yourusername placeholder).

• Added PyPI and CI status badges to README.

[0.1.152] - 2024-12-29

Summary

This is a retrospective entry covering the cumulative state of the library at v0.1.152.

Features

• Lazy LAS 2.0 file reader with header-only parsing and on-demand data loading.

• Property class with depth-weighted statistics (mean, sum, std, percentile, mode).

• Chained hierarchical filtering: well.PHIE.filter('Zone').filter('Facies').sums_avg().

• Numpy-style operator overloading on Property objects (24 operators).

• WellDataManager for multi-well orchestration with property broadcasting.

• _ManagerPropertyProxy for cross-well operations: manager.PHIE.filter('Zone').mean().

• SumsAvgResult.report() with arithmetic, geometric, pooled, and sum aggregation.

• Template-driven well log visualization (Template, WellView).

• Interactive crossplots with multi-dimensional mapping (Crossplot).

• 6 regression types: linear, polynomial, exponential, logarithmic, power, polynomial-exponential.

• Parameter locking for regression coefficients.

• Discrete property support with label/color/style/thickness mappings.

• Source-aware property storage for round-trip LAS export.

• Computed property creation via well.HC = well.PHIE * (1 - well.SW).

• Project save/load to JSON with full metadata preservation.

• Depth interval computation using midpoint method with zone boundary truncation.

• Property resampling with configurable methods (linear, nearest, forward-fill).

• Strict depth alignment enforcement with detailed error guidance.