Projects
9 longer programs that combine several Deepbox modules. Read the source or run them yourself.
9 of 9 projects
- 01Financial Portfolio Risk AnalysisBuilds a synthetic eight-asset portfolio and reports its risk: VaR and CVaR, Sharpe and Sortino ratios, drawdown, correlations, optimized weights, a Monte Carlo forecast and a few stress scenarios.TensorsLinear AlgebraStatisticsDataFrame
- 02Neural Network Image ClassifierTrains a small multi-layer perceptron on the 8x8 digits dataset with mini-batch Adam, then reports test accuracy, precision, recall, F1 and a confusion matrix.Neural NetworksOptimizationTensorsMetrics
- 03Customer Churn PredictionPredicts churn on 1,000 synthetic customers. Six classifiers are trained and compared, then the best one by F1 is cross-validated and inspected.MLPreprocessingMetricsDataFrame
- 04Time Series Stock Price ForecastingForecasts next-day returns on 500 days of synthetic prices with lagged returns and technical indicators as features. Linear regression and ridge regression are compared with a mean baseline.TensorsStatisticsMLPreprocessing
- 05Movie Recommendation EngineCollaborative filtering on a synthetic rating matrix of 200 users and 50 movies, with user clustering and a 2D projection.TensorsMLMetricsDataFrame
- 06Sentiment AnalysisClassifies 500 synthetic reviews as positive or negative. Reviews are turned into TF-IDF vectors, and two classifiers are compared.MLPreprocessingMetricsDataFrame
- 07Fraud Detection PlatformScores 1,500 synthetic card transactions in four parts: reporting with a DataFrame, a supervised classifier with calibrated probabilities, three anomaly detectors trained on legitimate rows only, and model inspection.MLPreprocessingMetricsDataFrame
- 08Support Ticket TriageRoutes 640 synthetic support tickets to one of four queues (billing, account access, outage, bug report). Three text vectorizers are compared with logistic regression and Naive Bayes, and logistic regression is tuned with
GridSearchCV.PreprocessingMLMetricsDataFrame - 09Experimentation PlatformAnalyzes a synthetic checkout experiment with three variants (control, streamlined checkout, smart bundle) and 3,600 sessions. It builds scorecards, tests each variant against control, estimates uplift with a bootstrap and plans the sample size for a follow-up test.DataFrameStatisticsVisualizationRandom