How It Works
The chart above shows raw, noisy data (grey) and the Exponential Moving Average (green). EMA gives more weight to recent data points, making it responsive yet smooth.
EMA_t = α * x_t + (1 - α) * EMA_{t-1}
Adjust the Smoothing Factor (α) to see how EMA behaves:
- High α (e.g., 0.8): EMA follows the raw data closely. It reacts quickly to changes but is also very "jittery" (less smoothing).
- Low α (e.g., 0.1): EMA is very smooth and stable. It filters out noise effectively but reacts slowly to real trend changes (lag).
- In AI (like Adam optimizer), we often use a "decay rate" β, where α = 1 - β. A common β is 0.9, which means α = 0.1.