How karitselmuthu/enterprise_ticket_transfomer works
This is an educational ticket-classification project spanning V1–V7 training and evaluation, plus a V8 FastAPI inference service. The main workflow reads labeled tickets, fixes a reproducible split, trains a selected model generation, evaluates it, and loads an explicitly selected artifact for prediction. The graph preserves the distinct baseline, embedding, recurrent, Transformer, and pretrained-model stages and the production promotion check. Relationships for unsampled modules are included only where the README or SOURCE INDEX supports them; the source excerpts do not establish every internal training detail.
Main entry points: split.py Locked split · runtime.py Artifact runtime · train_neural.py V3–V5 trainer · main.py Prediction API
Components 24
Data preparation
- Ticket CSV
data/samples/tickets.csv - Locked split
src/training/split.py - Tokenizer
src/preprocessing/tokenizer.py - Neural vocabulary
src/preprocessing/vocabulary.py
Model training
- V1 trainer
src/training/train.py - Naive Bayes
src/baselines/traditional.py - V2 trainer
src/training/train_v2.py - Word2Vec classifier
src/baselines/word2vec.py - V3–V5 trainer
src/training/train_neural.py - Checkpoint preparation
src/training/prepare_pretrained.py - V6–V7 trainer
src/training/train_pretrained.py
Evaluation and promotion
- Model evaluation
src/evaluation/evaluate.py - Data audit
src/evaluation/data_audit.py - Generation comparison
src/evaluation/compare_all.py - Promotion validation
src/inference/promotion.py
Inference service
- Pretrained inference
src/inference/pretrained.py - Artifact runtime
src/inference/runtime.py - Prediction API
api/main.py
Model internals
- Skip-gram vectors
src/embeddings/word2vec.py - LSTM models
src/baselines/lstm.py - Transformer classifier
src/transformer/model.py - Self-attention
src/transformer/attention.py - Positional encoding
src/transformer/positional_encoding.py
Other components
- Engineer
Connections 38
- Engineer → Ticket CSV: provides data
README.md - Engineer → Locked split: creates split
README.md - Ticket CSV → Locked split: partitions rows
src/training/split.py - Locked split → V1 trainer: supplies manifest
src/training/train.py - Ticket CSV → V1 trainer: loads rows
src/training/train.py - V1 trainer → Naive Bayes: fits classifier
src/training/train.py - Naive Bayes → Tokenizer: tokenizes text
src/baselines/traditional.py - Engineer → V2 trainer: selects generation
README.md - Locked split → V2 trainer: supplies manifest
src/training/train_v2.py - V2 trainer → Word2Vec classifier: trains classifier
src/training/train_v2.py - Word2Vec classifier → Skip-gram vectors: trains vectors
src/baselines/word2vec.py - Skip-gram vectors → Tokenizer: tokenizes text
src/embeddings/word2vec.py - Engineer → V3–V5 trainer: selects generation
README.md - Locked split → V3–V5 trainer: loads split
src/training/train_neural.py - V3–V5 trainer → Neural vocabulary: builds vocabulary
src/training/train_neural.py - Neural vocabulary → Tokenizer: tokenizes text
src/preprocessing/vocabulary.py - V3–V5 trainer → LSTM models: constructs V3–V4
src/training/train_neural.py - V3–V5 trainer → Transformer classifier: constructs V5
src/training/train_neural.py - Transformer classifier → Self-attention: uses attention
src/transformer/model.py - Transformer classifier → Positional encoding: adds positions
src/transformer/model.py - Engineer → Checkpoint preparation: prepares checkpoint
README.md - Checkpoint preparation → V6–V7 trainer: provides local weights
README.md - Engineer → V6–V7 trainer: selects V6–V7
README.md - Locked split → V6–V7 trainer: supplies manifest
src/training/train_pretrained.py - Engineer → Model evaluation: requests evaluation
README.md - Model evaluation → Locked split: checks fingerprint
src/evaluation/evaluate.py - Engineer → Data audit: audits data
README.md - Data audit → Locked split: uses split logic
src/evaluation/data_audit.py - Engineer → Generation comparison: compares models
README.md - Prediction API → Artifact runtime: loads predictor
api/main.py - Prediction API → Promotion validation: validates production
api/main.py - Artifact runtime → Naive Bayes: dispatches V1
src/inference/runtime.py - Artifact runtime → Word2Vec classifier: dispatches V2
src/inference/runtime.py - Artifact runtime → V3–V5 trainer: constructs neural model
src/inference/runtime.py - Artifact runtime → Neural vocabulary: encodes input
src/inference/runtime.py - Artifact runtime → Pretrained inference: dispatches V6–V7
src/inference/runtime.py - Engineer → Prediction API: submits ticket
README.md - Prediction API → Artifact runtime: predicts text
api/main.py