How gauravbh1010tt/deeplearn works
This DeepLearn repository is a multi-project research hub for deep learning, NLP, CV, and ML implementations. Each project folder corresponds to a research paper’s experimental code – from attention-based CNNs (ABCNN) for sentence matching to Siamese LSTM models for sentence similarity, and even specialized models like CorrNet, neural tensor networks, and fake news stance identification. The overall design is modular: each project is self-contained with its own main script, model definitions, and dependency listing, while shared code and utilities (under _deeplearn_utils) support data processing, feature extraction, and deep learning layer definitions.
Main components
• Projects/Experiments: Represent each research implementation as a separate module. For example:
• Attention_Based_CNN (ABCNN)
• Attention_recurrent_models
• CorrNet
• MaLSTM (Siamese)
• TrecQA_CNN+Sim
• WikiQA_CNN+Feat
• convolution neural tensor network
• corrMCNN
• neural tensor network
• fake news challenge (FNC-1)
• Shared Utility Modules: Capture the _deeplearn_utils folder that includes:
• Extracted Features (for different datasets such as Trec and WikiQA)
• Data directories holding training and test CSV/TSV files, numpy arrays, and pickle files
• Deep learning layers and text utilities (dl_layers, dl_text, various utility scripts)
• External Dependencies and Integrations:
• Requirements.txt files that list dependencies for each project
• The use of external modules such as DL-text (as noted in the README) that prepare datasets and possibly handle other NLP utilities.
Out the relationships and interactions
• Each project (represented as its own component) has a “main.py” that drives the execution. This, in turn, calls the corresponding model definition (for example, model_abcnn.py, model_Siam_LSTM.py, etc.).
• Data flow starts from the raw data files present under _deeplearn_utils/data. They are pre-processed (possibly via scripts in dl_text or utility scripts) and then fed into model training and evaluation pipelines in the project modules.
• Each project’s configuration (requirements.txt, README instructions) aligns with external deep learning libraries, so denote dependency flow (external libraries → projects).
Architectural Patterns/Design Principles
• Modularity: Each experiment is encapsulated in its own directory, adhering to separation-of-concerns.
• Reusability: The _deeplearn_utils module is designed to be reused across different implementations.
• Experiment Isolation: By utilizing separate main scripts and requirements for each research paper implementation, the design supports concurrent experimentation with differing architectures and models.
• Data Pipeline: There is a distinct flow from raw data to extracted features and then model ingestion, supporting a clear separation between data handling and model computation.
• Experiment modules (e.g., rounded rectangles for projects)