v0.4.0

• release: bump version to 0.4.0
• style: apply formatting-only updates across modules and tests
• remove automated bioconda PR, generate recipe as artifact only; fix recipe (Python >=3.9, add scipy)
• clean up duplicate imports, stale comments, and misplaced import re
• remove unused imports and stale comments in hyenadna files
• remove dead variables and unused import in clustering.py
• fix: --filter-only flag was silently ignored (missing from args Namespace)
• remove dead _leiden_clustering_on_graph() function
• remove dead PathManager class from utils.py
• docs: update README to match current pipeline and fix cli OPTION_GROUPS ghost reference
• remove dead greedy_min_score parameter (F1 scores are always >= 0)
• Revert "fix: tighten greedy clustering contamination cap to 5%"
• fix: tighten greedy clustering contamination cap to 5%
• fix: block rescue merges above 10% duplication
• feat: expose rescue duplication limits in CLI and enforce max total duplication
• Revert "refactor: update encoder architecture to be dynamic and balanced"
• refactor: update encoder architecture to be dynamic and balanced
• fix: rate limit unconditional fusion debug logs
• fix: rate limit debug logging for gating weights
• fix: switch debug prints to logger.debug
• feat: add debug prints for gating weights
• Revert to state after 'feat: add short-reads/sr mode' (77f1ea5)
• Update rescue threshold: 0.9 for single sample, 0.7 for coassemblies
• Add 0.05 to default Leiden resolutions for larger clusters
• Fix indentation bug in rescue loop causing duplicate merges
• Update rescue strategy: use global SCG count for contamination and enforce 10% ceiling
• feat: add short-reads/sr mode to enforce 1000bp min contig length
• feat: add contamination filter to greedy clustering
• tune: implement F1-score based bin quality metric inspired by SemiBin2
• tune: add 0.1 to greedy clustering resolutions
• tune: score bins by total core genes (unique families) - 5*dups
• tune: set miniprot query coverage 0.6 identity 0.4, revert global -N/-p
• fix: use query coverage instead of target coverage for miniprot filtering
• Revert "Increase contamination penalty from 5 to 7 in bin quality scoring"
• Revert "Remove Singleton Rescue step"
• Revert "Increase contamination penalty from 7 to 10"
• Revert "Add 0.1 to default greedy clustering resolutions"
• Revert "tune: tighten miniprot thresholds"
• Revert "chore: remove agents instructions"
• Revert "tune: lower contamination penalty"
• Revert "tune: score bins by total core genes"
• Revert "tune: relax miniprot filters"
• Revert "tune: restore miniprot outs"
• tune: restore miniprot outs
• tune: relax miniprot filters
• tune: score bins by total core genes
• tune: lower contamination penalty
• chore: remove agents instructions
• tune: tighten miniprot thresholds
• Add 0.1 to default greedy clustering resolutions
• Increase contamination penalty from 7 to 10
• Remove Singleton Rescue step
• Increase contamination penalty from 5 to 7 in bin quality scoring
• Add debug logging for singleton rescue
• Restore exact logging format
• Revert greedy clustering parallelization
• Replace joblib with concurrent.futures
• Parallelize greedy clustering resolution search
• Reduce rescue duplication tolerance to 3%
• Implement Singleton Rescue step
• Remove refinement step and related CLI options
• Restore missing _leiden_clustering_on_graph function
• Replace adaptive resolution with greedy Leiden clustering
• fix: handle duplicate alignment filenames by using parent directory for disambiguation
• Fix: Remove unexpected 'use_header_cache' argument from check_core_gene_duplications calls
• feat: Log command line arguments for reproducibility
• test: add reproduction test for issue verification
• feat: implement bin rescue strategy
• chore: update .gitignore
• test: Add tests for CLI default parameters
• fix(tests): Resolve existing test suite failures
• feat: Update coassembly defaults for learning rate and lambda
• Remove standalone directory
• Tune coassembly resolution
• chore: remove AGENTS documentation
  Full Changelog: v0.3.4...v0.4.0

Full Changelog: v0.3.4...v0.4.0

v0.3.4

• chore: place license-files under project section
• chore: fix license files config
• fix: make license field build-compatible
• Prepare v0.3.4 release
• Merge branch '2modelapproach-wip'
• Lower coassembly auto barlow lambda to 0.005
• Add filter-only CLI mode
• Skip euk filter in single-cell mode
• Set coassembly min contig length to 4096
• Raise min contig length for coassemblies
• Dial back single-cell k-NN default
• Increase single-cell k-NN default
• Fix single-cell resolution to 0.01
• Lower single-cell resolution sweep to <=0.5
• Cap single-cell resolution sweep at 1.0
• Add mode presets and single-cell clustering behavior
• Seed training for reproducibility
• Adjust LR and auto barlow defaults
• Raise miniprot alignment thresholds
• Loosen completeness early-stop threshold to 50%
• Drop resolution 3.0 from coassembly sweep
• Prefer cluster count when duplications tie
• Drop over-split resolutions and prioritize completeness in selection
• Add completeness-driven early stop in resolution sweep
• Refine resolution selection tie-breakers
• Raise coassembly resolution floor to 0.6
• Quiet cluster sizes to debug level
• Adjust resolution sweep for coassemblies
• Test higher Leiden resolution for coassemblies
• Limit high-resolution testing to coassemblies
• Filter singleton bins from clustering summary log
• Add auto default for Barlow lambda based on sample count
• Add CLI knob for Barlow Twins lambda
• Refine bin splitting with Leiden and update defaults
• Switch refinement to k-means
• Seed training setup deterministically
• Simplify duplication checks
• Seed HyenaDNA predictor randomness
• Refactor feature calculations
• Write bins after filtering
• Vectorize clustering utilities
• Update CLI defaults
• Fix adaptive resolution grid
• Prune unused helpers in utils
• Vectorize cluster contig mapping
• Clean HyenaDNA model comments
• Update plot features import
• Remove AGENTS.md
• feat: change default refinement threshold to conservative (min_dup=2)
• wip: save current 2-model approach experiments
• refactor: show training progress bar only in verbose mode
• fix: revert test_resolutions to v0.3.0 values (3 resolutions)
• revert: restore v0.3.0 fusion layer architecture
• revert: restore v0.3.0 adaptive refinement strategy
• fix: restore v0.3.0 coverage encoder layer sizes
• refactor: revert to duplication minimization for resolution selection
• feat: auto-adjust batch size when dataset is smaller
• feat: add LRU cache for k-mer mapping and save-bins-before-refinement option
• feat: improve reproducibility with seeded random number generation
• fix: add missing losses.py module
• perf: reuse cached feature tensor in SequenceDataset
• refactor: remove redundant eukaryotic classification logging from clustering
• refactor: remove obsolete reclustering check log message
• refactor: remove unused _leiden_clustering import
• fix: use hashlib for deterministic fragment generation
• refactor: remove unused extract_base_contig_name import
• refactor: remove unused tqdm import from output module
• refactor: replace transformers tokenizer with standalone implementation
• refactor: remove transformers dependency from core requirements
• feat: add version number to params.json output
• fix: correct BUSCO gene family parsing and lower quality thresholds
• perf: add deterministic fragments and k-NN graph disk caching
• fix: inline gene count extraction to avoid missing import
• fix: restore batch size default to 2048
• fix: enable refinement to work without -k flag
• perf: cache k-NN graph in adaptive resolution + log cleanup
• fix: remove logic that assigns unembedded contigs to largest cluster
• perf: add graph caching helper for refinement
• refactor: count only single-copy genes for completeness metric
• refactor: round resolution value in Leiden log message
• refactor: remove redundant resolution log message
• fix: add --save-filtered-contigs to Filtering & Processing section
• refactor: simplify CLI by removing parameter range restrictions
• chore: remove unused dependencies (psutil, biopython, joblib)
• refactor: remove 3 redundant refinement log messages
• refactor: merge graph info messages into one
• refactor: round resolution values in Leiden clustering message
• refactor: remove 'Adaptive resolution determination' message
• refactor: remove duplicate message and increase batch size
• perf: optimize miniprot execution with caching
• refactor: remove redundant filtered FASTA message
• refactor: clean up output logging
• refactor: move fragment augmentation message to debug
• refactor: simplify miniprot logging message
• refactor: clean up BAM processing progress display
• refactor: improve BAM processing logging and progress display
• feat: improve logging and add save-filtered-contigs option
• feat: update HyenaDNA classifier to use improved model
• refactor: improve CLI UX with positional args and optional output
• fix: add run_exports to bioconda recipe template
• fix: update bioconda recipe generation to fix test failures
• fix: resolve deprecation warnings in workflows and package config
• refactor: move calculate_raw_coverage.py to scripts directory
• refactor: centralize miniprot thresholds and adjust to 0.30/0.50
• chore: remove unused code from refinement and utils modules
• feat: auto-adjust batch size when dataset is smaller
• chore: add large data files to .gitignore
• docs: prepare release 0.3.0
• refactor: clean up unused adaptive strategy parameters
• refactor: reduce console output verbosity
• docs: add repository guidelines for development workflow
• feat: add standalone HyenaDNA predictor
• feat: add Barlow Twins training diagnostics
• feat: add adaptive resolution and performance optimizations
• Fix xgboost import and classification column names
• Replace XGBoost classifier with HyenaDNA LLM-based model
• Prepare 0.2.5 release
• chore: drop unused setup_logging import
• fix: retain best checkpoint state
• fix: guard zero-read coverage normalization
• chore: require python 3.9+
• Always write embeddings.csv regardless of --keep-intermediate flag
• Update fusion layer architecture
• Release v0.2.4
• Merge branch 'main' of https://github.com/danielzmbp/remag
• refinement: skip bins without duplication data; remove '(conservative approach)' from warnings
• Merge branch 'main' of https://github.com/danielzmbp/remag
• feat: implement conservative refinement strategy to preserve completeness
• Merge branch 'main' of https://github.com/danielzmbp/remag
• Fix critical edge cases and test infrastructure issues
• Merge branch 'main' of https://github.com/danielzmbp/remag
• Eliminate code duplication in DataFrame column initialization
• Fix security vulnerability and optimize performance
  Full Changelog: v0.3.3...v0.3.4

Full Changelog: v0.3.3...v0.3.4

v0.3.3

Full Changelog: ...v0.3.3

Full Changelog: https://github.com/danielzmbp/remag/commits/v0.3.3

v0.3.2

• chore: bump version to 0.3.2
• fix: suppress pandas FutureWarning for fillna downcasting
• refactor: remove obsolete reclustering check log message
• refactor: quality-aware resolution selection and improved refinement
• fix: use hashlib for deterministic fragment generation
• tune: adjust learning rate, coverage threshold, and refinement rounds
• refactor: remove unused tqdm import from output module
• refactor: replace transformers tokenizer with standalone implementation
• refactor: remove transformers dependency from core requirements
• feat: add version number to params.json output
• fix: correct BUSCO gene family parsing and lower quality thresholds
• perf: add deterministic fragments, k-NN caching, and simplified logging
• refactor: improve resolution testing and fix refinement without -k
• perf: cache k-NN graph in adaptive resolution + improve metrics
• fix: remove logic that assigns unembedded contigs to largest cluster
• perf: cache k-NN graph during refinement + extend resolution range + find most conservative solution
• fix: remove minimum resolution threshold in refinement
• refactor: simplify refinement with fixed resolution steps
• refactor: fix k-neighbors and threshold during refinement
• refactor: count only single-copy genes for completeness metric
• refactor: round resolution value in Leiden log message
• refactor: fix clustering parameters during resolution testing
• refactor: widen auto-resolution testing range
• refactor: increase refinement resolution steps for faster convergence
• chore: increase default minimum bin size to 300kb
• chore: change default base learning rate to 0.001
• refactor: remove redundant resolution log message
• refactor: improve auto-resolution metrics and refinement
• fix: add --save-filtered-contigs to Filtering & Processing section
• refactor: simplify CLI by removing parameter range restrictions
• refactor: remove experimental SCG loss functionality
• chore: remove unused dependencies (psutil, biopython, joblib)
• refactor: remove 3 redundant refinement log messages
• refactor: merge graph info messages into one
• refactor: round resolution values in Leiden clustering message
• refactor: remove 'Adaptive resolution determination' message
• refactor: remove duplicate message and increase batch size
• perf: optimize miniprot execution with caching
• refactor: move SCG feature matrix message to debug
• refactor: remove redundant filtered FASTA message
• refactor: clean up output logging
• refactor: move fragment augmentation message to debug
• refactor: move SCG gene mappings message to debug
• refactor: simplify miniprot logging message
• refactor: clean up BAM processing progress display
• refactor: improve BAM processing logging and progress display
• feat: improve logging and add save-filtered-contigs option
• feat: update HyenaDNA classifier to use improved model
• refactor: improve CLI UX with positional args and optional output
• refactor: move calculate_raw_coverage.py to scripts directory
• feat: add SCG-aware contrastive learning with consolidated miniprot execution
• fix: add run_exports to bioconda recipe template
  Full Changelog: v0.3.1...v0.3.2

Full Changelog: v0.3.1...v0.3.2

v0.3.1

• chore: bump version to 0.3.1
• fix: update bioconda recipe generation to fix test failures
• fix: resolve deprecation warnings in workflows and package config
  Full Changelog: v0.3.0...v0.3.1

Full Changelog: v0.3.0...v0.3.1

v0.3.0

Full Changelog: ...v0.3.0

Full Changelog: https://github.com/danielzmbp/remag/commits/v0.3.0

v0.2.5

• Prepare 0.2.5 release
• chore: drop unused setup_logging import
• fix: retain best checkpoint state
• fix: guard zero-read coverage normalization
• chore: require python 3.9+
• Always write embeddings.csv regardless of --keep-intermediate flag
• Update fusion layer architecture
  Full Changelog: v0.2.4...v0.2.5

Full Changelog: v0.2.4...v0.2.5

v0.2.4

• Release v0.2.4
• Merge branch 'main' of https://github.com/danielzmbp/remag
• refinement: skip bins without duplication data; remove '(conservative approach)' from warnings
• Merge branch 'main' of https://github.com/danielzmbp/remag
• feat: implement conservative refinement strategy to preserve completeness
• Merge branch 'main' of https://github.com/danielzmbp/remag
• Fix critical edge cases and test infrastructure issues
• Merge branch 'main' of https://github.com/danielzmbp/remag
• Eliminate code duplication in DataFrame column initialization
• Fix security vulnerability and optimize performance
• remove comment
• README: remove explicit conda install of miniprot; clarify it’s an automatic dependency
  Full Changelog: v0.2.3...v0.2.4

Full Changelog: v0.2.3...v0.2.4

v0.2.3

• Prepare v0.2.3 release - Bug fixes and dependency updates
• Fix undefined variable error in clustering and clean up imports
• Update recipe dependencies for bioconda
• Update README.md for v0.2.2 - Remove obsolete k-means references and update version examples
  Full Changelog: v0.2.2...v0.2.3

Full Changelog: v0.2.2...v0.2.3

v0.2.2

• Update CHANGELOG.md for v0.2.2 release
• Prepare v0.2.2 release - Final cleanup and version bump
• Fix undefined variable error in clustering
• Refactor codebase for improved maintainability and reduced complexity
• clean up redundant code and comments
• update dependencies for bioconda
• remove small euk db
• add args parameter to _construct_knn_graph function
• add --outs=0.95 parameter to miniprot command
• remove unnecessary noise handling in leiden clustering
• limit bins.csv to contig and cluster columns
• fix: remove duplicate v prefix in Zenodo title
  Full Changelog: v0.2.1...v0.2.2

Full Changelog: v0.2.1...v0.2.2