Vera AI Nlp Generating

NLP Text Classification --- Full Analysis & Manuscript Generation

Table of Contents

• Scope Boundary
• Workflow
• Additional Inputs
• Output Structure
• Key References (read before generation)
• Reporting Standards
• Method Status
• Configuration Defaults
• Why These Defaults
• GPU Availability
• Minimal Smoke Test
• Cross-Skill Interface

Open-source skill. Read reference/specs/output-variation-protocol.md before every generation --- apply all variation layers for natural, diverse output.

Scope Boundary

Use this skill when:

• vera-ai-nlp-reviewing has already established a credible baseline and the user wants a fuller model battery.
• The task is supervised text classification, optionally with a small set of numeric side features.

Do not use this skill when:

• The task is generation, retrieval, summarization, sequence labeling, or prompt-only classification.
• The desired method is not in the shipped model list below.
• The available hardware cannot support anything beyond CPU smoke testing and the user expects transformer-scale benchmarking.

Workflow

Continues from where vera-ai-nlp-reviewing stopped (PART 0-2 done).

Additional Inputs

Collect if not already provided:

• Target discipline (for reporting conventions)
• Target journal or style (ACL, EMNLP, NeurIPS, etc.)
• Research question / hypothesis
• Subgroup variable or text property for stratification

Output Structure

output/
├── methods.md
├── results.md
├── tables/             ← Markdown + CSV per table
├── figures/            ← PNGs, 300 DPI
├── references.bib
└── code.py             ← Style-varied

Key References (read before generation)

Reporting Standards

Same as vera-ai-nlp-reviewing, plus:

• All models: report F1 (weighted) and AUC (macro) with bootstrapped 95% CIs
• Deep learning: report training epochs, best epoch, learning rate, batch size
• ALBERT: report pre-trained model name, whether base was frozen
• Feature importance: unified 0-100 scale across ML and DL models
• Model comparison: frame as convergent findings, not horse race
• Tree-based with small N: frame as "exploratory"; never claim generalizability

Method Status

Configuration Defaults

Pipeline constants live in config/default.json. Read it before generation. Adds to the testing-skill config:

• ml_models.* — SVM/RF/LightGBM hyperparameter grids
• deep_models.* — GRU/TextCNN/ALBERT hyperparameter grids, epochs, batch size, learning rate
• gpu.{required_for, fallback_models} — e.g., ALBERT requires GPU; falls back to GRU+TextCNN if unavailable (see GPU Availability section below)
• feature_importance.scale (0, 100) — unified scale across model families
• subgroup.{min_size, interaction_alpha} — subgroup-analysis gating

To override: create config/local.json.

Why These Defaults

• ALBERT is the shipped transformer because it is materially lighter than full BERT-style baselines while still exercising a modern contextual encoder path.
• GRU + TextCNN are the CPU fallback because they preserve two complementary inductive biases, sequential recurrence and local n-gram filters, without pretending a transformer run happened.
• Tree-based models on sparse text remain marked "exploratory" at small N because they are useful robustness checks, not generally the most defensible primary NLP model.

GPU Availability

Check at workflow step 06 (advanced models) start. Implementation pattern:

import torch
GPU_AVAILABLE = torch.cuda.is_available()
MODELS_TO_FIT = ["GRU", "TextCNN"] + (["ALBERT"] if GPU_AVAILABLE else [])
if not GPU_AVAILABLE:
    print("NOTE: CUDA not available; falling back to GRU + TextCNN. ALBERT skipped.")

Do NOT silently proceed without a model — the fallback must always include at least GRU + TextCNN on CPU.

Minimal Smoke Test

• Smoke-test prompt: "Continue from vera-ai-nlp-reviewing on a 4-class 20 Newsgroups subset. Fit the additional ML models and the CPU-safe deep models, then produce the manuscript-ready artifacts."
• CPU pass condition: SVM, Random Forest, LightGBM, GRU, and TextCNN run; ALBERT may be skipped if CUDA is unavailable, but the fallback decision must be logged.
• Expected artifacts: methods.md, results.md, tables/, figures/, references.bib, and one consolidated code.py.

Cross-Skill Interface

Method Unit Contract:
├── code_python      → .py script (style-varied)
├── methods_md       → methods.md (varied structure)
├── results_md       → results.md (varied phrasing)
├── tables/          → Markdown + CSV
├── figures/         → PNGs 300 DPI (varied layout)
├── references_bib   → .bib with cited references
└── comparison       → cross-method narrative (in results.md)