pypfilt.examples

The pypfilt.examples module includes a chaotic dynamical system, a predator-prey system, a Gaussian random walk, and several epidemic SIR models.

Dynamical systems models

class pypfilt.examples.lorenz.Lorenz63

The Lorenz-63 system:

\[\begin{split}\frac{dx}{dt} &= \sigma (y - x) \\ \frac{dy}{dt} &= x (\rho - z) - y \\ \frac{dz}{dt} &= xy - \beta z \\ x_t &= [\sigma, \rho, \beta, x, y, z]^T\end{split}\]

This system exhibits chaotic behaviour in the neighbourhood of \(\sigma = 10\), \(\rho = 28\), \(\beta = \frac{8}{3}\).

field_types(ctx)

Define the state vector \([\sigma, \rho, \beta, x, y, z]^T\).

d_dt(time, xt, ctx, is_forecast)

The right-hand side of the ODE system.

Parameters:
  • time – The current time.

  • xt – The particle state vectors.

  • ctx – The simulation context.

  • is_forecast – True if this is a forecasting simulation.

can_smooth()

Indicate which state vector fields can be smoothed.

class pypfilt.examples.lorenz.ObsLorenz63(obs_unit, settings)

An observation model for the Lorenz-63 system:

\[y_t \sim N(\mu = x_t, \sigma = 1.5)\]

The observation unit must be the name of a field in the state vector. For example, the observation unit must be "y" to observe \(y(t)\):

[observations.y]
model = "pypfilt.examples.lorenz.ObsLorenz63"
distribution(ctx, snapshot)

Return a frozen scipy.stats distribution that defines the observation model for each particle.

Parameters:
  • ctx – The simulation context.

  • snapshot (Snapshot) – The current particle states.

pypfilt.examples.lorenz.lorenz63_simulate_toml()

A scenario for the Lorenz63 model, which can be used for simulating observations.

Returns:

The scenario definition, represented as a TOML string.

Return type:

str

pypfilt.examples.lorenz.lorenz63_forecast_toml()

A scenario for the Lorenz63 model, which can be used for forecasting.

Returns:

The scenario definition, represented as a TOML string.

Return type:

str

pypfilt.examples.lorenz.lorenz63_forecast_regularised_toml()

A scenario for the Lorenz63 model, which can be used for forecasting, and enabled post-regularisation.

Returns:

The scenario definition, represented as a TOML string.

Return type:

str

pypfilt.examples.lorenz.lorenz63_all_scenarios_toml()

All example scenarios for the Lorenz63 model.

Returns:

The scenario definitions, represented as a TOML string.

Return type:

str

Predator-prey system

Models

class pypfilt.examples.predation.LotkaVolterra

An implementation of the (continuous) Lotka-Volterra equations.

field_types(ctx)

Return a list of (field_name, field_dtype, field_shape) tuples that define the state vector.

The third element, field_shape, is optional and contains the shape of this field if it forms an array of type field_dtype.

These tuples must be in the same order as the state vector itself.

Parameters:

ctx – The simulation context.

d_dt(time, xt, ctx, is_forecast)

Calculate the derivatives of x(t) and y(t).

can_smooth()

Return the set of field names in the state vector that can be smoothed by the post-regularised particle filter (see post_regularise()).

Note

Models should only implement this method if the state vector contains parameters that can be smoothed.

class pypfilt.examples.predation.ObsModel(obs_unit, settings)
distribution(ctx, snapshot)

Return a frozen scipy.stats distribution that defines the observation model for each particle.

Parameters:
  • ctx – The simulation context.

  • snapshot (Snapshot) – The current particle states.

Example files

pypfilt.examples.predation.write_example_files()

Save the following example files to the working directory:

  • The forecast scenario file “predation.toml”;

  • The observations file “predation-counts-x.ssv”;

  • The observations file “predation-counts-y.ssv”;

  • The forecast scenario file “predation-datetime.toml”;

  • The observations file “predation-counts-x-datetime.ssv”; and

  • The observations file “predation-counts-y-datetime.ssv”;

pypfilt.examples.predation.example_toml_data()

Return the contents of the example file “predation.toml”.

pypfilt.examples.predation.example_obs_x_data()

Return the contents of the example file “predation-counts-x.ssv”.

pypfilt.examples.predation.example_obs_y_data()

Return the contents of the example file “predation-counts-y.ssv”.

pypfilt.examples.predation.example_toml_datetime_data()

Return the contents of the example file “predation-datetime.toml”.

pypfilt.examples.predation.example_obs_x_datetime_data()

Return the contents of the example file “predation-counts-x-datetime.ssv”.

pypfilt.examples.predation.example_obs_y_datetime_data()

Return the contents of the example file “predation-counts-y-datetime.ssv”.

Generating forecasts

pypfilt.examples.predation.forecast(data_file)

Run a suite of forecasts against generated observations, using a scalar time scale.

Parameters:

date_file – The name of the output HDF5 file.

pypfilt.examples.predation.plot(data_file, png=True, pdf=True)

Save the plots produced by plot_params() and plot_forecasts().

This will save the plots to files whose names begin with “predation_params” and “predation_forecasts”.

Parameters:
  • png – Whether to save plots as PNG files.

  • pdf – Whether to save plots as PDF files.

pypfilt.examples.predation.plot_params(param_cints, pdf_file=None, png_file=None)

Plot the parameter posteriors over the estimation run.

pypfilt.examples.predation.plot_forecasts(state_cints, x_obs, y_obs, pdf_file=None, png_file=None)

Plot the population predictions at each forecasting date.

Other functions

pypfilt.examples.predation.default_priors()

Define default model prior distributions.

pypfilt.examples.predation.predation_instance(toml_file)

Return an instance of the predation scenario from the specified TOML file.

pypfilt.examples.predation.predation_scalar_instance()

Return an instance of the predation scenario using a scalar time scale.

pypfilt.examples.predation.predation_datetime_instance()

Return an instance of the predation scenario using a datetime time scale.

pypfilt.examples.predation.apply_ground_truth_prior(instance)

Define the predation model prior distribution for fixed ground truth.

pypfilt.examples.predation.save_scalar_observations(obs_tables, x_obs_file, y_obs_file)

Save simulated observations to disk.

Gaussian random walk

Models

class pypfilt.examples.simple.GaussianWalk

A Gaussian random walk.

\[\begin{split}x_t &= x_{t-1} + X_t \\ X_t &\sim N(\mu = 0, \sigma = 1)\end{split}\]

The initial values \(x_0\) are defined by the prior distribution for "x":

[prior]
x = { name = "uniform", args.loc = 10.0, args.scale = 10.0 }
field_types(ctx)

Return a list of (field_name, field_dtype, field_shape) tuples that define the state vector.

The third element, field_shape, is optional and contains the shape of this field if it forms an array of type field_dtype.

These tuples must be in the same order as the state vector itself.

Parameters:

ctx – The simulation context.

update(ctx, time_step, is_fs, prev, curr)

Perform a single time-step.

class pypfilt.examples.simple.GaussianObs(obs_unit, settings)

A Gaussian observation model for the GaussianWalk model.

\[\mathcal{L}(y_t \mid x_t) \sim N(\mu = x_t, \sigma = s)\]

The observation model has one parameter: the standard deviation \(s\), whose value is defined by the "parameters.sdev" setting:

[observations.x]
model = "pypfilt.examples.simple.GaussianObs"
parameters.sdev = 0.2
distribution(ctx, snapshot)

Return a frozen scipy.stats distribution that defines the observation model for each particle.

Parameters:
  • ctx – The simulation context.

  • snapshot (Snapshot) – The current particle states.

Support functions

pypfilt.examples.simple.gaussian_walk_toml_data()

Return the contents of the example file “gaussian_walk.toml”.

pypfilt.examples.simple.gaussian_walk_instance()

Return an instance of the simple example scenario.

Epidemic SIR models

Models

class pypfilt.examples.sir.SirCtmc

A continuous-time Markov chain implementation of the SIR model.

The model settings must include the following keys:

  • population_size: The number of individuals in the population.

field_types(ctx)

Define the state vector structure.

can_smooth()

The fields that can be smoothed by the post-regularisation filter.

init(ctx, vec)

Initialise the state vectors.

update(ctx, time_step, is_forecast, prev, curr)

Update the state vectors to account for all events that occur up to, and including, time.

active_particles(vec, stop_time)

Return a Boolean array that identifies the particles whose most recent event occurred no later than stop_time.

select_next_event(ctx, vec, stop_time)

Calculate the next event time and event type for each active particle.

class pypfilt.examples.sir.SirDtmc

A discrete-time Markov chain implementation of the SIR model.

The model settings must include the following keys:

  • population_size: The number of individuals in the population.

field_types(ctx)

Define the state vector structure.

can_smooth()

The fields that can be smoothed by the post-regularisation filter.

init(ctx, vec)

Initialise the state vectors.

update(ctx, time_step, is_forecast, prev, curr)

Update the state vectors.

class pypfilt.examples.sir.SirOdeEuler

An ordinary differential equation implementation of the SIR model, which uses the forward Euler method.

The model settings must include the following keys:

  • population_size: The number of individuals in the population.

field_types(ctx)

Define the state vector structure.

can_smooth()

The fields that can be smoothed by the post-regularisation filter.

init(ctx, vec)

Initialise the state vectors.

update(ctx, time_step, is_forecast, prev, curr)

Update the state vectors.

class pypfilt.examples.sir.SirOdeRk

An ordinary differential equation implementation of the SIR model, which uses the explicit Runge-Kutta method of order 5(4).

The model settings must include the following keys:

  • population_size: The number of individuals in the population.

field_types(ctx)

Define the state vector structure.

can_smooth()

The fields that can be smoothed by the post-regularisation filter.

init(ctx, vec)

Initialise the state vectors.

d_dt(time, xt, ctx, is_forecast)

The right-hand side of the system.

class pypfilt.examples.sir.SirSde

A stochastic differential equation implementation of the SIR model.

The model settings must include the following keys:

  • population_size: The number of individuals in the population.

field_types(ctx)

Define the state vector structure.

can_smooth()

The fields that can be smoothed by the post-regularisation filter.

init(ctx, vec)

Initialise the state vectors.

update(ctx, time_step, is_forecast, prev, curr)

Update the state vectors.

class pypfilt.examples.sir.SirObs(obs_unit, settings)

A binomial observation model for the example SIR models.

\[ \begin{align}\begin{aligned}\mathcal{L}(y_t \mid x_t) &\sim B(n, p)\\n &= S(t-\Delta) - S(t)\end{aligned}\end{align} \]
Parameters:
  • obs_unit – A descriptive name for the data.

  • settings – The observation model settings dictionary.

The settings dictionary should contain the following keys:

  • observation_period: The observation period \(\Delta\).

For example, for daily observations that capture 80% of new infections:

[observations.cases]
model = "pypfilt.examples.sir.SirObs"
observation_period = 1
parameters.p = 0.8
new_infections(ctx, snapshot)

Return the number of new infections \(S(t-\Delta) - S(t)\) that occurred during the observation period \(\Delta\) for each particle.

distribution(ctx, snapshot)

Return the observation distribution for each particle.

Example files

pypfilt.examples.sir.sir_toml_data()

Return the contents of the example file “sir.toml”.