pyhgf.distribution.logp#
- pyhgf.distribution.logp(mean_1: float, mean_2: float, mean_3: float, precision_1: float, precision_2: float, precision_3: float, tonic_volatility_1: float, tonic_volatility_2: float, tonic_volatility_3: float, tonic_drift_1: float, tonic_drift_2: float, tonic_drift_3: float, volatility_coupling_1: float, volatility_coupling_2: float, input_precision: float, response_function_parameters: Array | ndarray | bool_ | number | bool | int | float | complex, input_data: Array | ndarray | bool_ | number | bool | int | float | complex, time_steps: Array | ndarray | bool_ | number | bool | int | float | complex, response_function_inputs: ndarray | Array | bool_ | number | bool | int | float | complex | None, response_function: Callable | None, hgf: HGF) float[source]#
Compute the log-probability of a decision model under belief trajectories.
This function returns the evidence of a single Hierarchical Gaussian Filter given network parameters, input data and behaviours under a decision model.
- Parameters:
- mean_1
The mean at the first level of the HGF. For the continuous HGF, this is the mean of the first value parent (x_1). For the binary HGF this is the mean of the binary state node (x_0).
- mean_2
The mean at the second level of the HGF. For the continuous HGF, this is the mean of the first volatility parent (x_2). For the binary HGF this is the mean of the first continuous state node (x_1).
- mean_3
The mean at the third level of the HGF. The value of this parameter will be ignored when using a two-level HGF (n_levels=2). For the continuous HGF, this is the mean of the second volatility parent (x_3). For the binary HGF this is the mean of the first volatility parent (x_2).
- precision_1
The precision at the first level of the HGF. For the continuous HGF, this is the precision of the first value parent (x_1). For the binary HGF this is the precision of the binary state node (x_0).
- precision_2
The precision at the second level of the HGF. For the continuous HGF, this is the precision of the first volatility parent (x_2). For the binary HGF this is the precision of the first continuous state node (x_1).
- precision_3
The precision at the third level of the HGF. The value of this parameter will be ignored when using a two-level HGF (n_levels=2). For the continuous HGF, this is the precision of the second volatility parent (x_3). For the binary HGF this is the precision of the first volatility parent (x_2).
- tonic_volatility_1
The tonic volatility at the first level (x_1 for the continuous HGF, x_2 for the binary HGF). This parameter represents the tonic part of the variance (the part that is not inherited from parent nodes).
- tonic_volatility_2
The tonic volatility at the second level (x_2 for the continuous HGF, x_3 for the binary HGF). This parameter represents the tonic part of the variance (the part that is not inherited from parent nodes).
- tonic_volatility_3
The tonic volatility at the third level of the HGF. This parameter represents the tonic part of the variance (the part that is not inherited from parent nodes). This parameter is only used for a three-level continuous HGF.
- tonic_drift_1
The tonic drift at the first level of the HGF (x_1 for the continuous HGF, x_2 for the binary HGF). This parameter represents the drift of the random walk.
- tonic_drift_2
The tonic drift at the second level of the HGF (x_2 for the continuous HGF, x_3 for the binary HGF). This parameter represents the drift of the random walk.
- tonic_drift_3
The tonic drift at the third level of the HGF. This parameter represents the drift of the random walk. This parameter is only used for a three-level continuous HGF.
- volatility_coupling_1
The volatility coupling between the first and second levels of the HGF (between x_1 and x_2 for a continuous HGF, and between x_2 and x_3 for a binary HGF). This represents the phasic part of the variance (the part affected by the parent nodes). Defaults to 1.0 (full connectivity).
- volatility_coupling_2
The volatility coupling between the second and third levels of the HGF (x_2 and x_2 for a continuous HGF, not applicable to a binary HGF). This represents the phasic part of the variance (the part affected by the parent nodes). Defaults to 1.0 (full connectivity). The value of this parameter will be ignored when using a two-level HGF (n_levels=2).
- input_precision
The expected precision associated with the continuous input.
- response_function_parameters
An array of additional parameters that will be passed to the response function to compute the surprise. This can include values over which inference is performed in a PyMC model (e.g. the inverse temperature of a binary softmax).
- input_data
An array of input time series where the first dimension is the number of models to fit in parallel.
- time_steps
An array of input time steps where the first dimension is the number of models to fit in parallel.
- response_function_inputs
An array of behavioural inputs passed to the response function where the first dimension is the number of models to fit in parallel.
- response_function
The response function that is used by the decision model.
- hgf
An instance of a two or three-level Hierarchical Gaussian Filter.
- Returns:
- logp
The log-probability (negative surprise).