BehaviourSim is an open-source Python engine and cloud API designed to model complex human and system trajectories. Generate deterministic or stochastic traces governed by Markov dynamics, Dirichlet priors, and state-conditioned feature emissions.
BehaviourSim structures behavioral generation into a clean, decoupled generative pipeline. Every observed event is conditioned on underlying transition dynamics.
s_0 \sim \pi_0Agents sample an initial state from a baseline prior distribution or profile vector.
P(s_{t+1} \mid s_t, \theta)Stochastic transition matrix evaluates temporal inertia, absorption, and Markov steps.
s_t \in \mathcal{S}Discrete behavioral status (e.g. browsing, trading, researching, churned).
x_t \sim \mathcal{F}(s_t)Continuous and discrete emissions sampled conditionally on the current active state.
y_t = g(s_t, x_t)Deterministic or stochastic threshold checks flag alerts, anomalies, or conversions.
\mathcal{D} = \{(x_t, y_t)_{t=1}^T\}Structured pandas DataFrame exported with full lineage, timestamps, and sequence IDs.
Generate rich tabular datasets in three lines of code using domain-specific presets or custom stochastic state definitions.
123456789101112131415# 1. Initialize simulator from a registered domain preset from behaviorsim import Simulator # Presets: 'finance', 'healthcare', 'education', 'mobile_app' sim = Simulator.from_preset("finance") # 2. Generate reproducible multi-agent sequential behavioral traces df = sim.generate( num_interactions=100, num_sequences=10, seed=42 ) # 3. Inspect generated dataset (profiles, latent states, observable metrics) print(df[["sequence_id", "interaction_id", "state", "portfolio_value", "risk_alert"]].head())
Domain presets configure realistic states, transition probability matrices, and continuous feature distributions calibrated for specific behavioral domains. Read the domain presets reference or follow the quickstart guide.
| seq_id | step | state | metric_val | alert |
|---|---|---|---|---|
| 1 | 0 | evaluating | 104.28 | 0 |
| 1 | 1 | transacting | 108.91 | 0 |
| 1 | 2 | hedging | 92.45 | 1 |
| 2 | 0 | evaluating | 101.12 | 0 |
Pass a fixed integer seed to guarantee identical sequence generation across operating systems and hardware configurations.
Constructed for high-fidelity experimentation where standard uniform or Gaussian random generators fail to capture sequential dependencies.
First-order and higher-order discrete state graphs with customizable transition probabilities, inertia weights, and absorbing states. Learn more in transitions & dynamics.
Emit continuous or categorical features sampled from state-conditioned Normal, Beta, Poisson, or custom parametric distributions. Read our distribution methodology.
Built-in domain factories for finance, healthcare, educational engagement, and mobile application retention modeling. Explore practical use cases.
High-throughput REST API with quota tracking, session management, and parameter evaluation hosted at api.behavioursim.vedaangsharma.in. Test the API Explorer.
BehaviourSim is an open-source Python-based platform for generating, calibrating, validating, and analyzing reproducible synthetic sequential behavioral data.
behaviorsim, providing Simulator, State, and distribution samplers.Direct, factual answers to common questions about synthetic sequential data generation, Markov modeling, and BehaviourSim capabilities.
BehaviourSim is an open-source Python framework (distributed as 'behaviorsim' on PyPI) and cloud API designed for generating reproducible synthetic sequential behavioral data. It models human and system trajectories as discrete Markov state machines that emit parametric features at each interaction step.
Synthetic sequential behavioral data consists of artificially generated time-series or event-log trajectories where each record depends chronologically on previous states and actions. Unlike independent, identically distributed (IID) tabular rows, sequential behavioral data preserves temporal dependencies, transition dynamics, and user state progressions.
Install the package via 'pip install behaviorsim', initialize a simulator using Simulator.from_preset('finance') (or custom states), and call sim.generate(num_interactions=100, num_sequences=10, seed=42). The method returns a structured pandas DataFrame.
In BehaviourSim, a behavioral state represents a discrete internal or operational condition of an agent (such as 'browsing', 'evaluating', 'transacting', 'churned', 'adherent', or 'stressed'). Active states govern which feature distributions are sampled and constrain valid transitions to subsequent states.
Generation follows a two-stage decoupled stochastic process: First, an agent transitions across discrete states according to a Markov transition probability matrix or conditional rules. Second, observable continuous and categorical features are sampled from probability distributions parameterized specifically for that active state.
Yes. BehaviourSim guarantees deterministic reproducibility across platforms by accepting an explicit integer seed. Running simulation workloads with the same seed, state definitions, and emission parameters produces bit-for-bit identical sequence datasets.
BehaviourSim supports 8 verified parametric emission distributions: Normal (Gaussian), Lognormal, Exponential, Poisson, Uniform, Uniform Discrete, Bernoulli (binary flags), and Categorical (multinomial choices).
Yes. Agents can be initialized with distinct profile vectors and prior distributions (Dirichlet priors), allowing simulations to model heterogeneous cohorts such as novice versus power users, high-frequency traders, or adherent versus non-adherent patients.
BehaviourSim includes 4 built-in domain presets: Finance (trading states, portfolio values, volatility alerts), Healthcare (vital signs, adherence, dosage levels), Education (engagement states, accuracy, response times), and Mobile App (onboarding, active session, checkout funnel, churn). Note: presets are simulation models for testing and research, not clinical or financial advice.
Yes. Researchers and ML engineers use BehaviourSim to benchmark recurrent neural networks (RNNs/LSTMs/Transformers), evaluate feature engineering pipelines, simulate temporal concept drift, and test anomaly detection without exposing sensitive real-world user data.
Yes. BehaviourSim provides a hosted cloud REST API at api.behavioursim.vedaangsharma.in with endpoints for preset discovery, asynchronous simulation job dispatch, status polling, and API key management.
Yes. The BehaviourSim Web platform at behavioursim.vedaangsharma.in provides an interactive Simulation Playground, statistical telemetry visualizers, transition matrix heatmaps, and comprehensive documentation.
Generic random generators sample rows independently without temporal continuity or state awareness. BehaviourSim enforces temporal inertia, absorbing Markov states, state-conditioned multivariate distributions, and deterministic lineage required for realistic sequence simulation.
BehaviourSim is licensed under the permissive MIT License. Full test suites, transition matrix verification, and simulation recipes are open for community inspection.