How xtiwby/metaphor_detector2 works
This repository is a staged multimodal metaphor-detection research pipeline: it prepares Flickr-based inputs to train a Word2Vec semantic disambiguator, applies that model to MET-Meme examples, then trains and evaluates a metaphor classifier. The README also documents a Contrast classifier branch, although its implementation is not shown in the sampled source. Training, evaluation, and experiment artifacts form the main internal layer. Some intermediate file-writing details and the Contrast implementation were not sampled, so those links rely on README statements.
Main entry points: run_classifier_train.py Classifier training · run_disambiguator_train.py Disambiguator training · trainer.py Training and evaluation · calculate_disambiguator_input.py Flickr encoding
Components 16
Input preparation
- Flickr encoding
calculate_disambiguator_input.py - Flickr datasets
data_processing/flickr_dataset.py
Disambiguation training
- Disambiguator training
run_disambiguator_train.py - W2V disambiguator
model/clip_w2v_gpt.py
Metaphor classification
- MET-Meme encoding
calculate_classifier_input.py - Embedded dataset
data_processing/metaphor_dataset.py - Classifier training
run_classifier_train.py - GPT2 classifier
model/clip_w2v_gpt.py - Contrast branch
model/clip_w2v_gpt.py
Training support
- Training and evaluation
train_eval/trainer.py - Model and metrics
train_eval/trainer.py - Seed settings
utils/setting.py - Progress logging
utils/logger_handler.py
Other components
- Researcher
- Flickr30k data
- MET-Meme data
Connections 21
- Researcher → Flickr encoding: prepares inputs
README.md - Flickr30k data → Flickr encoding: encodes captions
calculate_disambiguator_input.py - Flickr encoding → Disambiguator training: supplies features
README.md - Flickr30k data → Flickr datasets: loads examples
data_processing/flickr_dataset.py - Flickr datasets → Disambiguator training: provides batches
run_disambiguator_train.py - Researcher → Disambiguator training: starts training
README.md - Disambiguator training → W2V disambiguator: instantiates
run_disambiguator_train.py - Disambiguator training → Training and evaluation: delegates training
run_disambiguator_train.py - Disambiguator training → Seed settings: sets seed
run_disambiguator_train.py - Training and evaluation → Model and metrics: saves model
train_eval/trainer.py - MET-Meme data → MET-Meme encoding: encodes examples
calculate_classifier_input.py - W2V disambiguator → MET-Meme encoding: disambiguates features
calculate_classifier_input.py - MET-Meme encoding → Embedded dataset: writes embedded data
README.md - Embedded dataset → Classifier training: provides batches
run_classifier_train.py - Researcher → Classifier training: starts training
README.md - Classifier training → GPT2 classifier: instantiates
run_classifier_train.py - Classifier training → Training and evaluation: delegates evaluation
run_classifier_train.py - Classifier training → Seed settings: sets seed
run_classifier_train.py - Classifier training → Progress logging: configures logging
run_classifier_train.py - Training and evaluation → Model and metrics: reports metrics
train_eval/trainer.py - Contrast branch → Model and metrics: yields results
README.md