training-data
Prepare, validate, and train SketchCode paired PNG/GUI datasets, including vocabulary, preprocessing, model architecture, and legacy training troubleshooting.
- License 许可
- MIT
- 最近更新
- 2026-10-08
- Stars
- ★ 11
- Forks
- 3
01 前言 Front Matter
- name
- training-data
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Prepare, validate, and train SketchCode paired PNG/GUI datasets, including vocabulary, preprocessing, model architecture, and legacy training troubleshooting.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/api-reference.md5.2 KB
- 参考文档references/data-formats.md3.8 KB
- 参考文档references/model-overview.md3.7 KB
- 参考文档references/training-workflow.md5.9 KB
- 参考文档references/troubleshooting.md6.6 KB
03 SKILL.md 正文
SketchCode training-data
Use this sub-skill when the task involves preparing paired SketchCode training data, validating .png/.gui layouts, understanding vocabulary.vocab, running or adapting legacy training, fine-tuning from model JSON/weights, inspecting image preprocessing, or explaining the CNN-plus-GRU model structure.
Route here for
- Training SketchCode from scratch or from existing
model_json.jsonand weights. - Checking a flat dataset of paired
sample_id.pngandsample_id.guifiles before training. - Debugging
training_set,validation_set,.npzfeature outputs, augmentation, or validation split surprises. - Explaining vocabulary tokens, GUI DSL tokenization, sequence generation, and model save files.
Route elsewhere
- Converting wireframe PNGs to
.guior HTML belongs toconversion-inference. - BLEU scoring or comparing predicted/original
.guifiles belongs toevaluation. - Downloading public data or pretrained model assets belongs to the root SketchCode asset workflow, not this training sub-skill.
Operating sequence
-
Read references/data-formats.md to confirm the dataset shape, vocabulary, tokenization, and preprocessing outputs.
-
Run the bundled validator before expensive or destructive training:
python sub-skills/training-data/scripts/validate_training_dataset.py DATASET_DIR -
Use the guarded bundled training wrapper when a real run is requested; it dry-runs by default and requires explicit destructive-split acknowledgement:
python sub-skills/training-data/scripts/run_training.py --help -
Read references/training-workflow.md before launching training; it explains command templates, flags/defaults, fine-tuning inputs, and the destructive split directory behavior.
-
Read references/model-overview.md when the task asks about architecture, optimizer, callbacks, saved files, sequence lengths, or memory expectations.
-
Use references/api-reference.md for a concise source-level API and CLI map.
-
If anything fails, start with references/troubleshooting.md, especially for missing pairs, duplicate GUI layouts, legacy TensorFlow/Keras/OpenCV imports, and the
VOCAB_FILErelative-path issue.
Safety defaults
- Validate first; full training is long-running and mutates filesystem state.
- Stage data in a scratch parent directory. The legacy split code deletes and recreates sibling
training_setandvalidation_setdirectories under the parent of--data_input_path. - Do not fine-tune in-place inside a directory that contains important
training_setorvalidation_setfolders. - Use the bundled wrapper's dry-run output to inspect the split directories and model output location before passing
--run --allow-destructive-split.
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.