biophysics_fitting ❭ exploration_from_seedpoint ❭ RW ❭ RW

RW

class biophysics_fitting.exploration_from_seedpoint.RW.RW(df_seeds=None, param_ranges=None, params_to_explore=None, evaluation_function=None, MAIN_DIRECTORY=None, min_step_size=0, max_step_size=0.02, checkpoint_every=100, checkpoint_by_time=None, concat_every_n_save=60, n_iterations=60000, mode=None, aim_params=None, stop_n_inside_with_aim_params=-1)

Class to perform RW exploration from a seedpoint.

Parameters:

  • df_seeds (pd.DataFrame) – individual seed points as rows and the parameters as columns

  • param_ranges (pd.DataFrame) – parameters as rows and has a min_ and max_ column denoting range of values this parameter may take

  • params_to_explore (list) – list of parameters that should be explored. If None, all parameters are explored.

  • evaluation_function (Callable) – takes one argument (a new parameter vector) and returns:

    • inside: boolean that indicates if the parameter vector is within experimental constraits (i.e. results in acceptable physiology) or not.

    • evaluation: dictionary that will be saved alongside the parameters. For example, this should contain ephys features.

  • checkpoint_every (int) – save the results every n iterations

  • check_point_by_time (float) – time interval in minutes for checkpointing for using time-based checkpointing. If both checkpoint_every and checkpoint_by_time are set, checkpointing will be done by time.

  • concat_every_n_save (int) – number of checkpoints after which the intermediate .pickle` files are concatenated to a single ``.parquet dataframe.

  • mode (str) – Random walk mode. Options: (None, ‘expand’). default: None ‘expand’: only propose new points that move further away from seedpoint

  • aim_params (dict) – this param will make the exploration algorithm propose only new points such that a set of parameters aims certain values during exploration. Default: {}

  • stop_n_inside_with_aim_params (int) – number of successful models / set of parameters inside space with aim_params to find before stopping exploration

  • MAIN_DIRECTORY (str) – output directory in which results are stored.

Attributes:

df_seeds

individual seed points as rows and the parameters as columns

Type:

pd.DataFrame

param_ranges

parameters as rows and has a min_ and max_ column denoting range of values this parameter may take

Type:

pd.DataFrame

params_to_explore

list of parameters that should be explored. If None, all parameters are explored.

Type:

list

evaluation_function

Function use to evaluate a vector of biphysical parameters. Must take one argument (a parameter vector) and return a tuple of (inside, evaluation):

  • inside: boolean that indicates if the parameter vector is within experimental constraits

    (i.e. results in acceptable physiology) or not.

  • evaluation: dictionary that will be saved alongside the parameters. For example, this should contain

    ephys features.

This function is usually evaluation_function_incremental_helper().

Type:

Callable

MAIN_DIRECTORY

output directory in which results are stored.

Type:

str

min_step_size

minimum step size for the random walk

Type:

float

max_step_size

maximum step size for the random walk

Type:

float

checkpoint_every

save the results every n iterations

Type:

int

checkpoint_by_time

time interval in minutes for checkpointing for using time-based checkpointing. If both checkpoint_every and checkpoint_by_time are set, checkpointing will be done by time.

Type:

float

concat_every_n_save

number of checkpoints after which the pickle files are concatenated and cleaned

Type:

int

n_iterations

number of iterations to run the random walk

Type:

int

mode

Random walk mode. Options: (None, ‘expand’). default: None ‘expand’: only propose new points that move further away from seedpoint

Type:

str

aim_params

this param will make the exploration algorithm propose only new points such that a set of parameters aims certain values during exploration. Default: {}

Type:

dict

stop_n_inside_with_aim_params

number of successful models / set of parameters inside space with aim_params to find before stopping exploration

Type:

int

normalized_aim_params

normalized aim parameters

Type:

pd.Series

Methods:

_normalize_aim_params(aim_params)

Normalize aim parameters to be between 0 and 1.

_normalize_params(p)

Normalize parameters to be between 0 and 1.

_unnormalize_params(p)

Unnormalize parameters to be between min and max.

_concatenate_and_clean(seed_folder, particle_id, iteration)

Concatenate the intermediate .pickle results and save as one parquet file.

_clean_the_pickles(outdir, files, iteration)

Remove the pickle files that correspond to the intermediate results of a iteration.

_load_pickle_or_parquet(outdir, iteration, mode)

Load the results of a iteration from a pickle or parquet file.

assess_aim_params_reached(normalized_params, tolerance)

Check whether the aim parameters have been reached.

run_RW(selected_seedpoint, particle_id, seed)

Run random walk exploration from a seed point.