A-TIER technique
VectorBT Grid Search — mmap data, joblib batching and vectorized backtests
100k ft — What is this?
VectorBT Grid Search — mmap data, joblib batching and vectorized backtests
50k ft — Why care?
A speed recipe for strategy grid searches with VectorBT: load OHLCV via numpy memmap (mmap_mode='r', copy=False) to avoid RAM blowups, fan thousands of backtests across cores with joblib Parallel, batch tasks in groups of 1,000, and run Portfolio.from_signals with fees and slippage as Numba-compiled matrix ops. Illustrated with a BTC moving-average crossover return surface.
10k ft — Key takeaways
• Always pass fees and slippage into Portfolio.from_signals so grid winners are not fee artifacts.
• Batch joblib tasks to keep all cores saturated and cut process overhead.
• More accessible entry point for retail → quant transition
• Use np.load(..., mmap_mode='r') and DataFrame(copy=False) to backtest many tickers without RAM spikes.
• VectorBT enables fast vectorized strategy backtesting
5k ft — Tools & techniques
NumPy, Numba, Python, VectorBT, joblib, pandas
Vectorized Backtesting, algorithmic trading, backtesting, memory mapping, parallel batching, parameter grid search, vectorized backtesting
1k ft — Deep-dive / try it
VectorBT explained for Retail Bros
#finance #quantfinance #trading #quantrading
rishabh_singh05
26w
How to join ur classes?
More from @quant.dhawan
- B-TIER Heston Model — stochastic volatility for realistic options pricing
- B-TIER Cox Process — doubly stochastic Poisson for bursty market events
- A-TIER Hidden Markov Models — detecting market regime shifts automatically
- B-TIER Avellaneda-Stoikov — optimal spread calculation for market makers
- A-TIER Ornstein-Uhlenbeck Process — mean reversion for pairs trading