beetroots.sampler.saver package

Submodules

beetroots.sampler.saver.abstract_saver module

class beetroots.sampler.saver.abstract_saver.Saver(N: int, D: int, D_sampling: int, L: int, scaler: Scaler, results_path: str = '', batch_size: int | None = None, freq_save: int = 1, save_forward_map_evals: bool = False, list_idx_sampling: List[int] | None = None)[source]

Bases: object

enable to regularly save the progression of the Markov chain to a .hdf5 file

Parameters:

  • N (int) – total number of pixels to reconstruct

  • D (int) – total number of physical parameters

  • D_sampling (int) – number of physical parameters that are optimized / sampled

  • L (int) – number of observed lines

  • scaler (Scaler) – contains the transformation of the Theta values from their scaled space (in which the sampling happens) to their natural space

  • results_path (str) – path towards the .hdf5 output file, by default “”

  • batch_size (int, optional) – number of iterations between two saves on file, by default None

  • freq_save (int, optional) – save one sample in every (freq_save). Used to save disk space., by default 1

  • save_forward_map_evals (bool, optional) – wether to save the forward model evaluations and gradients, by default False

  • list_idx_sampling (Optional[List[int]], optional) – contains the indices of the physical parameters to be sampled

D

total number of physical parameters

Type:

int

D_sampling

number of physical parameters that are optimized / sampled

Type:

int

L

number of observed lines

Type:

int

N

total number of pixels to reconstruct

Type:

int

batch_size

frequency of saves, i.e., “every batch_size new iterates to be saved, the memory is saved to an .hdf5 file and re-initialized

Type:

int

check_need_to_save(t: int) → bool[source]

checks wether or not the memory should be saved to a .hdf5 file

Parameters:

t (int) – current iteration index

Returns:

wether or not to save to disk now

Return type:

bool

check_need_to_update_memory(t: int) → bool[source]

checks wether or not the memory should be updated

Parameters:

t (int) – current iteration index

Returns:

wether or not to save to update the memory

Return type:

bool

final_next_batch_size

int:

freq_save

frequency of saved iterates during the run (1 means that every iteration is saved)

Type:

int

abstract initialize_memory(T_MC: int, t: int, Theta: ndarray, forward_map_evals: dict = {}, nll_utils: dict = {}, dict_objective: dict = {}, additional_sampling_log: dict = {}) → None[source]

initializes the memory with the correct shapes

list_idx_sampling

contains the indices of the physical parameters to be sampled

Type:

1D np.ndarray

memory

stores the values before saving them to file

Type:

dict[str, Union[float, np.ndarray]]

next_batch_size

number of iterates to be stored in the next batch, i.e., until next save to file

Type:

int

results_path

path towards the .hdf5 output file

Type:

str

save_additional(list_arrays: List[ndarray], list_names: List[str]) → None[source]

saves additional content to a .hdf5 file

Parameters:

  • list_arrays (List[np.ndarray]) – list of the arrays to be saved

  • list_names (List[str]) – list of names for the arrays to be saved in the .hdf5 file

save_forward_map_evals

wether to save the forward model evaluations and gradients

Type:

bool

save_to_file()[source]

Saves the current memory content to a .hdf5 file

scaler

contains the transformation of the Theta values from their natural space to their scaled space (in which the sampling happens)

Type:

Scaler

set_results_path(results_path: str) → None[source]

sets the path of the .hdf5 file to a new value

Parameters:

results_path (str) – path towards the .hdf5 output file

t_last_init

time index of the last memory initialization

Type:

int

t_last_save

time index of the last save of the memory to .hdf5 file

Type:

int

abstract update_memory(t: int, Theta: ndarray, forward_map_evals: dict = {}, nll_utils: dict = {}, dict_objective: dict = {}, additional_sampling_log: dict = {}, rng_state_array: ndarray | None = None, rng_inc_array: ndarray | None = None) → None[source]

updates the memory with new information. All of the potential entries are optional except for the current iterate.

beetroots.sampler.saver.my_saver module

class beetroots.sampler.saver.my_saver.MySaver(N: int, D: int, D_sampling: int, L: int, scaler: Scaler, results_path: str = '', batch_size: int | None = None, freq_save: int = 1, save_forward_map_evals: bool = False, list_idx_sampling: List[int] | None = None)[source]

Bases: Saver

Parameters:

  • N (int) – total number of pixels to reconstruct

  • D (int) – total number of physical parameters

  • D_sampling (int) – number of physical parameters that are optimized / sampled

  • L (int) – number of observed lines

  • scaler (Scaler) – contains the transformation of the Theta values from their scaled space (in which the sampling happens) to their natural space

  • results_path (str) – path towards the .hdf5 output file, by default “”

  • batch_size (int, optional) – number of iterations between two saves on file, by default None

  • freq_save (int, optional) – save one sample in every (freq_save). Used to save disk space., by default 1

  • save_forward_map_evals (bool, optional) – wether to save the forward model evaluations and gradients, by default False

  • list_idx_sampling (Optional[List[int]], optional) – contains the indices of the physical parameters to be sampled

D

total number of physical parameters

Type:

int

D_sampling

number of physical parameters that are optimized / sampled

Type:

int

L

number of observed lines

Type:

int

N

total number of pixels to reconstruct

Type:

int

batch_size

frequency of saves, i.e., “every batch_size new iterates to be saved, the memory is saved to an .hdf5 file and re-initialized

Type:

int

final_next_batch_size

int:

freq_save

frequency of saved iterates during the run (1 means that every iteration is saved)

Type:

int

initialize_memory(T_MC: int, t: int, Theta: ndarray, forward_map_evals: dict = {}, nll_utils: dict = {}, dict_objective: dict = {}, additional_sampling_log: dict = {}) → None[source]

initializes the memory with the correct shapes

Parameters:

  • T_MC (int) – size of the Markov chain to be sampled / optimization procedure

  • t (int) – current iteration index

  • Theta (np.ndarray) – current iterate in the Markov chain / optimization run

  • forward_map_evals (dict, optional) – evaluations of the forward map and potentially derivatives, by default dict()

  • nll_utils (dict, optional) – evaluation of utilitary values of the Likelihood class, by default dict()

  • dict_objective (dict, optional) – contains the negative log posterior value and detailed components, by default dict()

  • additional_sampling_log (dict, optional) – additional data on the sampling / optimization run, by default dict()

list_idx_sampling

contains the indices of the physical parameters to be sampled

Type:

1D np.ndarray

memory

stores the values before saving them to file

Type:

dict[str, Union[float, np.ndarray]]

next_batch_size

number of iterates to be stored in the next batch, i.e., until next save to file

Type:

int

results_path

path towards the .hdf5 output file

Type:

str

save_forward_map_evals

wether to save the forward model evaluations and gradients

Type:

bool

scaler

contains the transformation of the Theta values from their natural space to their scaled space (in which the sampling happens)

Type:

Scaler

t_last_init

time index of the last memory initialization

Type:

int

t_last_save

time index of the last save of the memory to .hdf5 file

Type:

int

update_memory(t: int, Theta: ndarray, forward_map_evals: dict = {}, nll_utils: dict = {}, dict_objective: dict = {}, additional_sampling_log: dict = {}, rng_state_array: ndarray | None = None, rng_inc_array: ndarray | None = None) → None[source]

updates the memory with new information. All of the potential entries are optional except for the current iterate.

Parameters:

  • t (int) – current iteration index

  • Theta (np.ndarray) – current iterate in the Markov chain / optimization run

  • forward_map_evals (dict, optional) – evaluations of the forward map and potentially derivatives, by default dict()

  • nll_utils (dict, optional) – evaluation of utilitary values of the Likelihood class, by default dict()

  • dict_objective (dict, optional) – contains the negative log posterior value and detailed components, by default dict()

  • additional_sampling_log (dict, optional) – additional data on the sampling / optimization run, by default dict()

  • rng_state_array (Optional[np.ndarray], optional) – current state of the random generator (saved for sampling reproducibility), by default None

  • rng_inc_array (Optional[np.ndarray], optional) – current inc of the random generator (saved for sampling reproducibility), by default None

Module contents