10M SIMS cap
One binary. One file in > the analysis out.
Cross‑industry, offline, zero disk writes.
Quant, insurance, energy, telecom, genomics, climate, manufacturing — the same engine adapts to your domain. Run the modules you need.
Mapped read‑only (MAP_PRIVATE | MAP_NORESERVE). Never copied, never decompressed fully. No temp files, no caches — each run leaves no trace.
Large files (VCF, FASTA, HDF5) stream directly from the mmap. Swap‑friendly, RAM‑efficient, and fast — even on 8GB machines.
Performance, proven
simulations per second on a Ryzen 5 1600X
50M draws in 234 ms — real DJIA params
faster than pandas
177×177 Pearson matrix in 10.99 ms vs 136.50 ms
correlations per second
2534 observations, full 177×177 matrix
Tested on a 10‑year‑old AMD Ryzen 5 1600X with 8GB DDR4. Pure AVX2. No custom kernel.
The more cores and RAM you have, the faster it runs. AVX‑512 builds coming soon.
Modules
Selectable modules. Run what you need with --modules.
The Monte Carlo core runs automatically.
| Name | Domain | What it computes |
|---|---|---|
| montecarlo | core | high-throughput simulation of the series |
| correlation | cross | Pearson / Spearman / Kendall, partial correlation, significance |
| bayesian | cross | Normal-Normal belief update + change-point |
| heston | quant | Heston stochastic-volatility calibration + simulation |
| evt | quant/risk | Extreme Value Theory (POT / GPD) + return levels |
| copula | quant | Gaussian and C-vine copula dependence |
| timeseries | quant/ML | SARIMA + GJR-GARCH volatility |
| portfolio | quant | Markowitz portfolio + VaR / CVaR |
| cointegration | quant | Engle-Granger cointegration / pairs |
| hmm | ML | Gaussian hidden Markov regime detection |
| fixedincome | quant | Nelson-Siegel yield curve |
| randomforest | ML | Random Forest regression / classification |
| qmc | numeric | Quantum Monte Carlo (harmonic oscillator) |
| variants | genomics | VCF variant calling / cohort summary |
| popgen | genomics | population genetics |
--modules heston,evt,copula · --modules all · --modules 0x1400
Formats
djehuti_prep.py
djehuti_prep.py
Data preparation
The engine reads clean binary formats. Use the included djehuti_prep.py
helper to turn messy spreadsheets into a clean .npy.
# Install dependencies $ pip install numpy pandas pyarrow openpyxl # Convert CSV, TSV, Excel, Parquet, Feather to clean .npy $ python djehuti_prep.py messy.csv clean.npy $ python djehuti_prep.py data.parquet clean.npy $ python djehuti_prep.py book.xlsx clean.npy --sheet 0 # Output: clean.npy + clean.cols.json (column names)
Recipes
The engine's strength is the mix. Pick the modules that match your question. The Monte Carlo core runs automatically.
| Industry |
--modules recipe |
Reads as |
|---|---|---|
| Finance / quant | heston,evt,copula,cointegration,portfolio |
pricing + tail risk + dependence + pairs + allocation |
| Insurance | evt,copula,bayesian |
extreme losses + dependence + capital scenarios |
| Energy / utilities | timeseries,evt,hmm |
demand forecast + peak extremes + load regimes |
| Telecom | timeseries,randomforest,bayesian,correlation,hmm |
traffic forecast + churn drivers + anomaly + KPI links + usage regimes |
| Agriculture | timeseries,randomforest,evt,popgen,correlation |
yield forecast + driver attribution + drought/flood + breeding genetics |
| Water / hydrology | evt,timeseries,hmm,correlation |
flood/drought return levels + flow forecast + wet/dry regimes |
| Climate / environment | timeseries,evt,hmm,correlation |
trend + extremes + regimes + scenarios |
| Healthcare / genomics | popgen,variants,bayesian,randomforest |
population structure + variant summary + inference + classification |
| Manufacturing / quality | evt,randomforest,correlation |
process variation + defect extremes + driver analysis |
| Retail / e-commerce | timeseries,randomforest,bayesian |
demand seasonality + drivers + change-point + scenarios |
| Public sector / economics | cointegration,timeseries,correlation,evt |
long-run equilibria + forecasts + shocks |
| Mining / commodities | timeseries,evt,copula,cointegration |
price dynamics + spikes + co-movement + spreads |
| Research / academia* | all |
the full cross-sector mix |
* We are actively working on Djehuti 2 Academic version, with more capabilities and enhanced formulas for each module — especially the genomics module. Stay up to date!
System requirements
Any modern Linux with glibc 2.27+ (Ubuntu 18.04+, Debian 10+, RHEL 8+)
AVX2 is mandatory. The engine uses AVX2/FMA instructions. Most VPS instances (4-6 core EPYC) are Zen2+ or later and support AVX2 natively.
Linux: AppImage or raw binary (no dependencies)
Windows: Single .exe (no DLLs, no install)
Mac: Not currently supported.
# Linux — run the AppImage $ ./Djehuti-10M-x86_64.AppImage data.npy # Windows — run the .exe C:\> djehuti2-10M.exe data.npy # Pick modules $ ./Djehuti-10M data.npy --modules heston,evt,copula # See all options $ ./Djehuti-10M --help
Limited licensing offer to the first 50 buyers only.
Free updates for 1-year & Free upgrade to the 50M SIMS build when out. Then price will change.
Launch offer — $200 off
To get your refund:
We will first try to resolve your issue within 48-72 hours. If we can't, we will process the refund.
Not eligible for refund: