`datasets_theory`

Main class for generating datasets of theoretical trajectories. For details on this class, please check this tutorial.

datasets_theory

 datasets_theory ()

This class generates, saves and loads datasets of theoretical trajectories simulated from various diffusion models (available at andi_datasets.models_theory).

source

create_dataset

 create_dataset (T, N_models, exponents, models, dimension=1,
                 save_trajectories=False, load_trajectories=False,
                 path='datasets/', N_save=1000, T_save=1000)

Creates a dataset of trajectories via the theoretical models defined in .models_theory. Check our tutorials for use cases of this function.

	Type	Default	Details
T	int		Length of the trajectories.
N_models	int, numpy.array		- if int, number of trajectories per class (i.e. exponent and model) in the dataset. - if numpy.array, number of trajectories per classes: size (number of models)x(number of classes)
exponents	float, array		Anomalous exponents to include in the dataset. Allows for two digit precision.
models	bool, int, list		Labels of the models to include in the dataset. Correspodance between models and labels is given by self.label_correspodance, defined at init. If int/list, choose the given models. If False, choose all of them.
dimension	int	1
save_trajectories	bool	False	If True, the module saves a .h5 file for each model considered, with N_save trajectories and T = T_save.
load_trajectories	bool	False	If True, the module loads the trajectories of an .h5 file.
path	str	datasets/	Path to the folder where to save/load the trajectories dataset.
N_save	int	1000	Number of trajectories to save for each exponents/model. Advise: save at the beggining a big dataset (t_save ~ 1e3 and N_save ~ 1e4) which then allows you to load any other combiantion of T and N_models.
T_save	int	1000
Returns	numpy.array		- Dataset of trajectories of lenght Nx(T+2), with the following structure: o First column: model label o Second column: value of the anomalous exponent o 2:T columns: trajectories

source

create_segmented_dataset

 create_segmented_dataset (dataset1, dataset2, dimension=1,
                           final_length=200, random_shuffle=False)

Creates a dataset with trajectories which change diffusive feature (either model or anomalous exponent) after a time ‘t_change’.

	Type	Default	Details
dataset1	numpy.array		Array of size Nx(t+2), where the first columns values correspond to the labels of the model and anomalous exponent. The rest correspond to the trajectories of length t.
dataset2	numpy.array		Same as dataset1
dimension	int	1	Dimensions of the generated trajectories. Three possible values: 1, 2 and 3.
final_length	int	200	Length of the output trajectories.
random_shuffle	bool	False	If True, shuffles the first axis of dataset1 and dataset2.
Returns	numpy.array		Array of size Nx(t+5) whose columns represent: o Column 0: changing time o Column 1,2: labels first part of the trajectory (model, exponent) o Column 3,4: labels second part of the trajectory (model, exponent) o Column 5:(t+5): trajectories of lenght t.

source

create_noisy_diffusion_dataset

 create_noisy_diffusion_dataset (dataset=False, T=False, N=False,
                                 exponents=False, models=False,
                                 dimension=1,
                                 diffusion_coefficients=False,
                                 save_trajectories=False,
                                 load_trajectories=False,
                                 path='datasets/', N_save=1000,
                                 t_save=1000)

Create a dataset of noisy trajectories. This function creates trajectories with _create_trajectories and then adds given noise to them.
All arguments are the same as _create_trajectories but dataset and diffusion_coefficients.

	Type	Default	Details
dataset	bool	False	- If False, creates a dataset with the given parameters. - If numpy array, dataset to which the function applies the noise.
T	bool	False
N	bool	False
exponents	bool	False
models	bool	False
dimension	int	1
diffusion_coefficients	bool	False
save_trajectories	bool	False
load_trajectories	bool	False
path	str	datasets/
N_save	int	1000
t_save	int	1000
Returns	numpy.array		Dataset of trajectories of lenght Nx(T+2), with the following structure: o First column: model label o Second column: value of the anomalous exponent o 2:T columns: trajectories

source

create_noisy_localization_dataset

 create_noisy_localization_dataset (dataset=False, T=False, N=False,
                                    exponents=False, models=False,
                                    dimension=1, noise_func=False,
                                    sigma=1, mu=0,
                                    save_trajectories=False,
                                    load_trajectories=False,
                                    path='datasets/', N_save=1000,
                                    t_save=1000)

Create a dataset of noisy trajectories. This function creates trajectories with _create_trajectories and then adds given noise to them.
All parameters are the same as _create_trajectories but noise_func.

	Type	Default	Details
dataset	bool	False	If False, creates a dataset with the given parameters. If numpy array, dataset to which the function applies the noise.
T	bool	False
N	bool	False
exponents	bool	False
models	bool	False
dimension	int	1
noise_func	bool	False	If False, the noise added to the trajectories will be Gaussian distributed, with variance sigma and mean value mu. If function, uses the given function to generate noise to be added to the trajectory. The function must have as input two ints, N and M and the output must be a matrix of size NxM.
sigma	int	1
mu	int	0
save_trajectories	bool	False
load_trajectories	bool	False
path	str	datasets/
N_save	int	1000
t_save	int	1000
Returns	numpy.array		Dataset of trajectories of lenght Nx(T+2), with the following structure: o First column: model label o Second column: value of the anomalous exponent o 2:T columns: trajectories