How ghaniketrajputp005/adobe-gensolve-hackathon-2024 works
Project Type and Purpose
• It includes dataset generation, model training, evaluation, and model artifact storage.
File Structure
• Top-level directories and files:
• “problems/”: CSV files with doodle fragments (raw input data).
• “src/”: Python scripts for dataset generation (generate_dataset.py), model creation (model_creation_d.py, model_creation_dc.py), and testing (test_model_d.py, test_model_dc.py).
• “CURVES_MODELS/”: pre-trained .h5 model artifacts.
• “INSTRUCTIONS/”: documentation (PDF).
• requirements.txt / setup.py: dependency definitions.
• Virtual-environment implied by README (venv).
• Cloud integration mention—presumed storage bucket or object store for model artifacts.
README
• Dependencies managed via requirements.txt.
• Cloud storage used to persist trained models.
Main Components and Their Interactions
a. Data Ingestion Component
• Source: CSV files in “problems/”
• Responsible for reading raw doodle fragments.
b. Dataset Generation Component
• Script: src/generate_dataset.py
• Processes CSVs → cleans, augments, and outputs training/validation datasets.
c. Model Training Component
• Scripts: model_creation_d.py (simple model), model_creation_dc.py (complex model)
• Inputs: processed datasets → output: .h5 model files.
d. Model Evaluation Component
• Scripts: test_model_d.py, test_model_dc.py
• Runs on validation/test splits → produces accuracy reports.
e. Artifact Storage Component
• Local: “CURVES_MODELS/”
• Remote: Cloud storage (e.g., AWS S3, GCP Storage) for model persistence.
f. Development Environment
• Virtual environment (venv), dependencies via requirements.txt.
g. Documentation
• “INSTRUCTIONS/Curvetopia1-compressed…pdf”
Architectural Patterns and Design Principles
• Pipeline pattern: chaining together data ingestion → preprocessing → training → evaluation → storage.
• Separation of concerns: one script per pipeline stage.
• Reproducibility: dependency isolation (venv) + requirements.txt.
• Artifact versioning: distinct .h5 model files.
Technologies and Frameworks
• Python 3.x
• ML libraries (TensorFlow / Keras implied by .h5 models)
• CSV processing (pandas or built-in)
• Cloud SDK (e.g., boto3 or google-cloud-storage) for model upload (per README “Cloud Integration”).
Essential Elements Checklist
– [ ] Script/process names and locations
– [ ] Data sources and formats
– [ ] Output artifacts and storage targets
– [ ] Dependency/environment boundary