generative-models
Plan and troubleshoot DGL-LifeSci molecular generative workflows with DGMG, JTVAE/JTNNVAE, vocabulary handling, reconstruction, pretraining, generation, and safe input validation.
- License 许可
- Apache-2.0
- 最近更新
- 2026-10-08
- Stars
- ★ 11
- Forks
- 3
01 前言 Front Matter
- name
- generative-models
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Plan and troubleshoot DGL-LifeSci molecular generative workflows with DGMG, JTVAE/JTNNVAE, vocabulary handling, reconstruction, pretraining, generation, and safe input validation.
- disable-model-invocation
- true
02 资源文件 3
- 参考文档references/data-formats.md6.4 KB
- 参考文档references/troubleshooting.md9.4 KB
- 参考文档references/workflows.md8.7 KB
03 SKILL.md 正文
Generative Models
Use this sub-skill when an agent needs to plan, adapt, or debug DGL-LifeSci molecular generation workflows built around dgllife.model.DGMG, dgllife.model.JTNNVAE, dgllife.utils.JTVAEVocab, JTVAE MolTree utilities, or JTVAE datasets/collators.
Use This For
- Choosing between DGMG autoregressive graph generation and JTVAE/JTNNVAE junction-tree variational autoencoding.
- Preparing SMILES text inputs, DGMG atom/bond vocabularies, JTVAE vocabulary files, reconstruction inputs, and checkpoint expectations.
- Planning DGMG train/eval/generate or JTVAE pretrain/vae-train/reconstruct workflows without launching long training jobs.
- Diagnosing RDKit parseability, vocabulary mismatch, invalid generated molecules, checkpoint shape errors, and CPU/GPU/runtime issues.
Route Elsewhere
- For generic SMILES cleaning, CSV schemas, graph featurizers, and molecule dataset basics, use
../molecule-data-prep/SKILL.md. - For generic model-zoo constructor inspection or pretrained model loading outside DGMG/JTVAE, use
../model-zoo-pretraining/SKILL.md. - For property-prediction metrics, split strategy, training loops, and supervised evaluation, use
../property-prediction/SKILL.md.
References
- Read references/workflows.md for DGMG/JTVAE planning, API entry points, CLI-style arguments, outputs, and safe/unsafe execution classes.
- Read references/data-formats.md for SMILES files, DGMG atom/bond types, JTVAE vocabulary/MolTree expectations, and checkpoint contracts.
- Read references/troubleshooting.md for import/install, optional dependencies, RDKit sensitivity, vocabulary mismatch, invalid molecules, checkpoint shape, and long-runtime failures.
- Run scripts/validate_generative_inputs.py to validate small SMILES fixtures and optional JTVAE vocabulary tokens without training or downloading data.
Fast Workflow
- Classify the task: DGMG for direct graph-generation decisions; JTVAE/JTNNVAE for junction-tree encoding, reconstruction, and latent sampling.
- Validate user SMILES or vocabulary files with
python scripts/validate_generative_inputs.py --smiles-file molecules.txt --max-rows 100before planning model code. - For JTVAE, add
--vocab-file vocab.txtor--derive-jtvae-vocabto detect malformed vocabulary tokens, duplicates, and missing MolTree vocabulary tokens. - Check
references/workflows.mdfor constructor signatures, expected checkpoint naming, and whether a requested action is safe as a smoke check or requires explicit user approval. - Use
references/troubleshooting.mdwhen RDKit, DGL, Torch, vocabulary, checkpoint, generation validity, or runtime symptoms appear.
Version Notes
The verified installed package was dgllife 0.3.1 with DGL 1.1.3, CPU Torch, RDKit, scikit-learn, NumPy/SciPy, and pandas importable. Verified generative signatures included DGMG(atom_types, bond_types, node_hidden_size=128, num_prop_rounds=2, dropout=0.2) and JTNNVAE(vocab, hidden_size, latent_size, depth, stereo=True).
04 同仓库更多 Skills
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
python-api
Programmatic Object-Detection-Metrics API guidance for BoundingBox, BoundingBoxes, Evaluator, enums, coordinate conversion, AP/mAP semantics, and source-style import diagnostics.
file-evaluation
Folder-based PASCAL VOC AP/mAP evaluation from ground-truth and detection text files.
object-detection-metrics
Guides legacy Object-Detection-Metrics PASCAL VOC AP/mAP evaluation from detection text folders and direct Python API classes.
nx-workspace-patterns
Configure and optimize Nx monorepo workspaces. Use when setting up Nx, configuring project boundaries, optimizing build caching, or implementing affected commands.