- Update skill count badges and descriptions from 146 to 147 skills
- Add TileDB-VCF to genomic tools list in bioinformatics section
- Add variant database management use case for TileDB-VCF
- Add comprehensive TileDB-VCF entry to docs/scientific-skills.md
- Delete references/population_genomics.md
- Remove all references to deleted documentation files
- Clean up References section since no reference files remain
- Simplify skill to standalone main file only
- Delete detailed export and ingestion reference documentation
- Update main skill to remove references to deleted files
- Simplify skill to focus on core querying and population genomics
- Keep querying.md and population_genomics.md reference files
- All documentation is at https://cloud.tiledb.com/academy/
- Remove incorrect service URLs (docs.tiledb.com, support portal, etc.)
- Consolidate to academy and main platform URLs only
- Update contact information to sales@tiledb.com
- Replace incorrect subcommands (create-dataset, ingest, list-datasets)
- Use correct subcommands: create, store, export, list, stat, utils, version
- Update examples to match actual CLI usage patterns
- Add comprehensive list of all available subcommands with descriptions
- VCFs must be single-sample (multi-sample not supported)
- Index files (.csi or .tbi) are required for all VCF/BCF files
- Add indexing examples with bcftools and tabix
- Document requirements prominently in both main skill and ingestion guide
- Correct method: tiledb.cloud.vcf.read() not query_variants()
- Fix parameter: attrs not attributes
- Add namespace parameter for billing account
- Add .to_pandas() conversion step
- Use realistic example with TileDB-Inc dataset URI
- Remove Java references (focus on Python and CLI)
- Move all TileDB-Cloud content to bottom of document
- Update export example to show VCF format with .export() method
- Simplify 'When to Use' section focusing on open source capabilities
- Better document organization with cloud scaling at the end
- Add preferred conda environment setup with Python <3.10
- Include M1 Mac specific configuration (CONDA_SUBDIR=osx-64)
- Install tiledbvcf-py via mamba from tiledb channel
- Restore normal Python examples (not Docker-only)
- Keep Docker as alternative installation method
- Correct installation method: Docker images, not pip packages
- Update examples to show Docker container usage
- Based on actual TileDB-VCF repository documentation
- Add comprehensive TileDB-VCF skill by Jeremy Leipzig
- Covers open source TileDB-VCF for learning and moderate-scale work
- Emphasizes TileDB-Cloud for large-scale production genomics (1000+ samples)
- Includes detailed reference documentation:
* ingestion.md - Dataset creation and VCF ingestion
* querying.md - Efficient variant queries
* export.md - Data export and format conversion
* population_genomics.md - GWAS and population analysis workflows
- Features accurate TileDB-Cloud API patterns from official repository
- Highlights scale transition: open source → TileDB-Cloud for enterprise
Anomaly detection fixes:
- Fix critical quantile index bug: index 0 is mean not q10; correct indices are q10=1, q20=2, q80=8, q90=9
- Redesign test: use all 36 months as context, inject 3 synthetic anomalies into future
- Result: 3 CRITICAL detected (was 11/12 — caused by test-set leakage + wrong indices)
- Update severity labels: CRITICAL = outside 80% PI, WARNING = outside 60% PI
Covariates fixes:
- Fix variable-shadowing bug: inner dict comprehension overwrote outer loop store_id
causing all stores to get identical covariate arrays (store_A's price for everyone)
- Give each store a distinct price baseline (premium $12, standard $10, discount $7.50)
- Trim CONTEXT_LEN from 48 → 24 weeks; CSV now 108 rows (was 180)
- Add NOTE ON REAL DATA comment: temp file pattern for large external datasets
Both scripts regenerated with clean outputs.
- Created generate_html.py to embed JSON data directly in HTML
- No external fetch() needed - works when opened directly in browser
- File size: 149.5 KB (self-contained)
- Shows forecast horizon (12-36 months) in stats
- Each forecast now extends to 2025-12 regardless of historical data length
- Step 1 (12 points): forecasts 36 months ahead to 2025-12
- Step 25 (36 points): forecasts 12 months ahead to 2025-12
- GIF shows full forecast horizon at every animation step
- X-axis fixed to 2022-01 to 2025-12 (full data range)
- Y-axis fixed to 0.72°C to 1.52°C (full value range)
- Background shows all observed data (faded gray) + final forecast reference (faded red dashed)
- Foreground shows current step data (bright blue) + current forecast (bright red)
- GIF size reduced from 918KB to 659KB
Create an all-out demonstration showing how TimesFM forecasts evolve
as more historical data is added:
- generate_animation_data.py: Runs 25 incremental forecasts (12→36 points)
- interactive_forecast.html: Single-file HTML with Chart.js slider
- Play/Pause animation control
- Shows historical data, forecast, 80%/90% CIs, and actual future data
- Live stats: forecast mean, max, min, CI width
- generate_gif.py: Creates animated GIF for embedding in markdown
- forecast_animation.gif: 25-frame animation (896 KB)
Interactive features:
- Slider to manually step through forecast evolution
- Auto-play with 500ms per frame
- Shows how each additional data point changes the forecast
- Confidence intervals narrow as more data is added
- Change title from 'Above 1951-1980 Baseline' to clearer example description
- New title: 'TimesFM Zero-Shot Forecast Example / 36-month Temperature Anomaly → 12-month Forecast'
- Makes it clear this is a demonstration with limited input data
This branch was originally created from feat/markdown-mermaid-writing-skill
for development purposes, but the timesfm-forecasting skill should be
independent of PR #50.
- Remove scientific-skills/markdown-mermaid-writing/ directory
- Remove reference to markdown-mermaid-writing from SKILL.md integration section
- This PR now stands alone and does not require PR #50 to be merged first
Model weights (~800 MB) download on-demand from HuggingFace when skill
is first used. Preflight checker ensures sufficient resources before
any download begins.
Add comprehensive TimesFM forecasting skill with mandatory system
preflight checks (RAM/GPU/disk), end-to-end CSV forecasting script,
full API reference, data preparation guide, and hardware requirements
documentation. Supports TimesFM 2.5 (200M), 2.0 (500M), and legacy
v1.0 with automatic batch size recommendations based on hardware.
Per borealBytes review comment, removing the docs/project directory
from this PR since only the skill content should be included.
The docs/project content remains in my local fork for reference.
Refs: PR #50
Added '## ⚠️ Common pitfalls' section covering:
- Radar chart syntax (radar-beta vs radar, axis vs x-axis, curve syntax)
- XY Chart vs Radar syntax comparison table
- Accessibility notes for diagrams that don't support accTitle/accDescr
Prevents the x-axis → radar-beta confusion that occurred in the example
research report.
Changed from invalid 'radar' with 'x-axis' syntax to proper 'radar-beta'
syntax with axis/curve keywords as per references/diagrams/radar.md.
Also removed accTitle/accDescr (radar-beta doesn't support them) and
added italic description above the code block per accessibility requirements.
