TildAlice – Tech Blog for Developers
Python, AI, Deep Learning, Data Analysis, Computer Vision, Reinforcement Learning โ tips and tutorials
Latest Posts
-
RAFT vs FlowFormer Optical Flow: Speed, Accuracy, and Real Benchmarks
FlowFormer beats RAFT on Sintel by 15% but runs 3x slower. Real optical flow benchmarks with inference time, EPE, and optimization tricks.
-
RLHF Review: InstructGPT and Human Feedback Training
How InstructGPT's 1.3B model with RLHF beats 175B GPT-3. Full breakdown of SFT, reward modeling, and PPO training โ with the catches they skip.
-
Stock Time Series: Why ARIMA Failed and LSTM Barely Worked
ARIMA just repeats yesterday's price. LSTM hits 52% directional accuracy. Why stock forecasting tutorials hide the real success rates.
-
XAI in Practice: SHAP and LIME for Model Interpretation
SHAP vs LIME for XGBoost: when game theory beats local approximations. Interactive tutorial reveals what drives your model's predictions.
-
Part 3: Handling Imbalanced Data: SMOTE, ADASYN, and Beyond
SMOTE vs ADASYN vs cost-sensitive learning on Kaggle fraud data. Benchmark shows which method wins at precision, recall, and edge cases.
-
Part 2: Statistical Rigor: Why Your Correlation Might Be Spurious
Catch 4 statistical traps that make correlations spurious: Simpson's Paradox, p-hacking, confounders. Includes bootstrap tests you can run today.
-
Interactive Dashboards with Plotly and Streamlit
Build a live data dashboard in 50 lines: Plotly + Streamlit with callbacks, filters, and hover tooltips. No JavaScript required.
-
Part 6: Beyond Simulation: Addressing the Sim-to-Real Gap
Bridge the sim-to-real gap: domain randomization cut our robot failure rate by 73%. Here's the exact transfer learning pipeline we used.