KandiSeema

Project : Classification–Reconstruction Learning for Open Set Recognition using Python, PyTorch, NumPy, scikit-learn, Matplotlib This prototype demonstrates how the Open Set Recognition (OSR) model was trained, deployed, and tested using Microsoft Azure services to ensure scalability, reproducibility, and cloud-based execution. Data Upload & Cloud Storage (Azure Blob Storage) The dataset consisting of known-class samples (training) and unknown-class samples (testing) was uploaded to Azure Blob Storage. Python code running on Azure accessed the dataset directly from the blob container using the Azure Storage SDK. This enabled cloud-based data loading and eliminated the need for local dependencies. Model Training Using Azure Machine Learning (Azure ML Studio) A deep learning model combining classification (softmax) and reconstruction (autoencoder) architectures was trained using PyTorch. Training jobs were run on Azure ML Compute, using GPU-backed virtual machines. The cloud environment automatically handled: Package installation (PyTorch, NumPy, scikit-learn) Versioning of the training scripts Logging of accuracy, loss curves, and reconstruction error Model Evaluation & Open-Set Testing Using Azure ML pipelines, the model was evaluated on: Known-class data → classification accuracy Unknown-class data → reconstruction error + thresholding The results were visualized using Matplotlib, with plots stored back into Azure Blob Storage. Examples generated: Reconstruction error graphs ROC curve for known vs unknown detection Feature embeddings for visual separation Deployment as a Cloud Endpoint The trained OSR model was deployed as an Azure ML Endpoint (REST API). Input images/data are sent to the endpoint, which returns: Predicted class (if known) Reconstruction score Decision: “Known” or “Unknown” Live Testing of the Prototype A Python testing script hosted on Azure Notebook calls the endpoint. When a known class sample is submitted, the model returns a class label with low reconstruction error. When an unknown sample is sent, the endpoint detects high reconstruction error and classifies it as “Unknown.” This demonstrates real-time cloud-based Open Set Recognition using Microsoft Azure.

(Handwritten Character Recognition, Titanic EDA, Twitter Sentiment Analysis, Memory Game using Microsoft Azure) Step-by-Step Demonstration 1.Data Ingestion (Azure Blob Storage) The handwritten character dataset is stored in Azure Blob Storage. Python application fetches the images directly using Azure Storage SDK. 2.Model Training (Locally or Azure ML Studio) A classification model using NumPy, pandas, scikit-learn is trained. The model learns to identify characters based on pixel features. 3.Prediction Interface A simple UI (Python/Notebook) accepts a handwriting image. The trained model predicts the character. Results are logged to Azure if using cloud notebooks. 4.Cloud Integration The model can be deployed as a REST API using Azure ML Endpoint, allowing images to be sent from any app and receive a predicted output.

Gameplay Demonstration 1.UI Creation (Tkinter) Python Tkinter interface displays random numbers/colors for a few seconds. Game Logic (Random + Time modules) Random module generates sequences. Time module controls display intervals. User Input & Scoring User repeats the sequence. Correct answers increase score; mistakes end the round. 2. Azure Integration Game scores/results can be logged to Azure Table Storage or CosmosDB. The game executable can be hosted via Azure App Service as a desktop-style web app.