Skip to content

Latest commit

 

History

History
65 lines (43 loc) · 2.96 KB

README.md

File metadata and controls

65 lines (43 loc) · 2.96 KB
Raw

PsycheSail Backend Documentation 🧠⛵

Welcome to the PsycheSail Backend repository! This repository hosts the Python server codebase for the accompanying Flutter application, PsycheSail.

Overview 🌟

  • main.py: This file comprises the core FastAPI implementation, orchestrating API endpoints and invoking corresponding functionalities.
  • data.json: Here resides the dataset utilized for training the stress level detection model.
  • trainingTF/trainingSVM notebooks: Jupyter Notebooks dedicated to training and storing the NLP model for stress level prediction in chats.
  • inferencing.py: A script designed to load and predict stress levels using the saved model, providing insightful analytics.
  • VectorSearch.py: A program engineered to facilitate text-based search using Vertex AI Vector Search, pinpointing relevant chatrooms based on shared experiences.

Stress Detection 💆‍♂️

The Stress Detection module predicts user progress over time, furnishing users with analytics showcasing their improvement trajectory, fostering a sense of accomplishment and well-being.

TensorFlow NLP Model 🧠🔍

Trained on a corpus of 10,000 chats, our NLP model leverages TensorFlow. Saved in the .h5 format, it efficiently discerns stress levels in user interactions.

Here's a succinct rundown for your README file:

Model Architecture:

  • Embedding Layer: Converts text data into dense vector representations.
  • SimpleRNN Layer: Captures temporal dependencies in sequential input data.
  • Dense Layers: Learns intricate patterns within the data.
  • Output Layer: Generates probability distributions across stress level classes via softmax activation.

Training Process:

  • Data Preprocessing: Tokenizes and pads messages for uniform input size.
  • Model Compilation: Adam optimizer and sparse categorical cross-entropy loss function are applied.
  • Training: The model undergoes 20 epochs of training on preprocessed data.

Usage:

  • Training: Prepare a JSON dataset with message-text and corresponding stress-level fields. Customize hyperparameters and execute the training script.
  • Evaluation: Assess model performance using metrics like accuracy.
  • Inference: Utilize the trained model to predict stress levels for new textual inputs.

Vector Search 🔍📊

Our Vector Search module employs text embeddings to index and cluster messages. By identifying main topics and matching user input with relevant chatroom topics, it efficiently returns matching chatroom IDs.

FastAPI Server 🚀

The FastAPI server facilitates seamless integration between our Flutter application and the Python backend, enabling ML model processing and predictive functionalities.

Usage 🛠️

  1. Clone the repository:
git clone https://github.com/PSYCHE-SAIL/Backend
  1. Install the requirements (Recommended Python 3.10.x):
pip install -r requirements.txt
  1. Run the FastAPI Server:
uvicorn main:app --reload