feature-column-inputs
Build and validate DeepCTR-Torch feature columns and model_input dictionaries from tabular CTR, recommender, dense, sparse, and sequence data.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- feature-column-inputs
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Build and validate DeepCTR-Torch feature columns and model_input dictionaries from tabular CTR, recommender, dense, sparse, and sequence data.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/data-preprocessing.md6.0 KB
- 参考文档references/feature-columns-and-inputs.md8.8 KB
- 参考文档references/troubleshooting.md7.2 KB
- 脚本scripts/validate_feature_input.py23.2 KB
03 SKILL.md 正文
feature-column-inputs
Use this sub-skill when an agent must turn raw tabular CTR/recommender data into DeepCTR-Torch feature-column lists and the model_input dictionary consumed by model constructors, fit, predict, and evaluate.
What this sub-skill owns
- Choosing and declaring
SparseFeat,DenseFeat, andVarLenSparseFeatcolumns. - Building
linear_feature_columns,dnn_feature_columns, andfeature_names = get_feature_names(...). - Preparing sparse categorical ids, dense numeric arrays, sequence/multi-value arrays, and optional sequence length columns.
- Validating names, batch sizes, sparse id ranges, dense vector widths, sequence max lengths, padding,
length_name, sharedembedding_name, andgroup_nameusage before model construction. - Explaining common input/feature-column errors and safe fixes.
Route away
- Model-family choice, compile/fit/predict loops, metrics, callbacks, save/load, and single-output training belong in
../single-task-modeling/SKILL.md. - DIN/DIEN behavior histories,
behavior_feature_list, attention, GRU options, and negative sampling belong in../sequence-and-interest-models/SKILL.md; use this sub-skill only for the underlyingVarLenSparseFeatinput contract. - Shared inputs for multi-output labels can be built here, but MTL constructors, task names/types, target matrix shape, losses, and prediction interpretation belong in
../multitask-modeling/SKILL.md.
Operating sequence
-
Inventory columns by semantic type: categorical sparse ids, dense numeric/scaled vectors, sequence or multi-value categorical histories, and labels.
-
Preprocess categorical data into integer ids; reserve id
0for padding when a sequence uses mask-based pooling withoutlength_name. -
Declare feature columns using the exact constructor signatures and shape rules in feature-columns-and-inputs.
-
Build
feature_names = get_feature_names(linear_feature_columns + dnn_feature_columns)and thenmodel_input = {name: data[name] for name in feature_names}. -
Run the bundled validator before constructing the model:
python scripts/validate_feature_input.py python scripts/validate_feature_input.py --spec feature_input_spec.json -
If validation fails, resolve the matching symptom in troubleshooting before routing to a modeling sub-skill.
References and helper
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).