datasets_phenom


source

datasets_phenom

 datasets_phenom (models_class=<andi_datasets.models_phenom.models_phenom
                  object at 0x7543dd0f3910>)

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


source

create_dataset

 create_dataset (dics:list|dict|bool=None, T:None|int=None,
                 N_model:None|int=None, path:str='', save:bool=False,
                 load:bool=False)

Given a list of dictionaries, generates trajectories of the demanded properties. The only compulsory input for every dictionary is model, i.e. the model from which trajectories must be generated. The rest of inputs are optional. You can see the input parameters of the different models in andi_datasets.models_phenom, This function checks and handles the input dataset and the manages both the creation, loading and saving of trajectories.

Type Default Details
dics list | dict | bool None - if list or dictionary: the function generates trajectories with the properties stated in each dictionary.
- if bool: the function generates trajectories with default parameters set for the ANDI 2 challenge (phenom) for every available diffusion model.
T None | int None - if int: overrides the values of trajectory length in the dictionaries.
- if None: uses the trajectory length values in the dictionaries.
Caution: the minim T of all dictionaries will be considered!
N_model None | int None - if int: overrides the values of number of trajectories in the dictionaries.
- if None: uses the number of trajectories in the dictionaries
path str Path from where to save or load the dataset.
save bool False If True, saves the generated dataset (see self._save_trajectories).
load bool False If True, loads a dataset from path (see self._load_trajectories).
Returns tuple - trajs (array TxNx2): particles’ position. N considers here the sum of all trajectories generated from the input dictionaries. Note: if the dimensions of all trajectories are not equal, then trajs is a list.
- labels (array TxNx2): particles’ labels (see ._multi_state for details on labels)

In the example below we create two dictionaries and generate a dataset with it. See the corresponding tutorial for more details.

L = 50
dict_model3 = {'model': 'dimerization', 
               'L': L,
               'Pu': 0.1, 'Pb': 1}
dict_model5 = {'model': 'confinement',
               'L': L, 
               'trans': 0.2}

dict_all = [dict_model3, dict_model5]

trajs, labels = datasets_phenom().create_dataset(N_model = 10, # number of trajectories per model
                                                 T = 200,
                                                 dics = dict_all
                                                )
plot_trajs(trajs, L , N = 10, 
           num_to_plot = 3,
           labels = labels,
           plot_labels = True
          )
False
False

Creating, saving and loading trajectories

These auxiliary functions used in create_trajectories that allow for manipulate trajectories in various forms.


source

_create_trajectories

 _create_trajectories ()

Given a list of dictionaries, generates trajectories of the demanded properties. First checks in the .csv of each demanded model if a dataset of similar properties exists. If it does, it loads it from the corresponding file.

L = 20
dict_1 = {'model': 'single_state', 
          'L': L}
dict_2 = {'model': 'immobile_traps', 
               'L': L}
dict_all = [dict_1, dict_2]

DP = datasets_phenom()
trajs, labels = DP.create_dataset(N_model = 13, # number of trajectories per model
                                 T = 20,
                                 dics = dict_all                                            
                                )
plot_trajs(trajs, L , N = 10, 
           num_to_plot = 3,
           labels = labels,
           plot_labels = True
          )


source

_save_trajectories

 _save_trajectories (trajs, labels, dic, df, dataset_idx, path)

Given a set of trajectories and labels, saves two things:
- In the .csv corresponding to the demanded model, all the input parameters of the generated dataset. This allows to keed that of what was created before. - In a .npy file, the trajectories and labels generated.

trajs, labels = DP.create_dataset(N_model = 10, # number of trajectories per model
                                     T = 20,
                                     dics = dict_all,
                                     save = True, path = 'datasets_folder/'
                                    )
plot_trajs(trajs, L , N = 3)


source

_load_trajectories

 _load_trajectories (model_name, dataset_idx, path)

Given the path for a dataset, loads the trajectories and labels

# You must run to cells above for this one to work. Check that this are the 
# exact same trajectories.
trajs, labels = DP.create_dataset(N_model = 10, # number of trajectories per model
                                                 T = 20,
                                                 dics = dict_all[0],
                                                 load = True, path = 'datasets_folder/'
                                                )
plot_trajs(trajs, L , N = 3 )

Managing parameters and dictionaries

dictm = {'model': 'immobile_traps', 
               'L': 10}


DP = datasets_phenom()
DP.N_model = 10
DP.T = 20
DP.load = True
DP.save = False
DP.path = 'datasets_folder/'
try:
    DP._inspect_dic(copy.deepcopy(dictm))
except Exception as e: 
    print(e)
The dataset you want to load does not exist.
/home/gorka/miniconda3/envs/andi/lib/python3.10/site-packages/fastcore/docscrape.py:225: UserWarning: potentially wrong underline length... 
Returns 
----------- in 
Checks the information of the input dictionaries so that they fulfil the constraints of the program , completes missing information
with default values and then decides about loading/saving depending on parameters....
  else: warn(msg)

source

_inspect_dic

 _inspect_dic (dic)

Checks the information of the input dictionaries so that they fulfil the constraints of the program , completes missing information with default values and then decides about loading/saving depending on parameters.

Type Details
dic dict Dictionary with the information of the trajectories we want to generate
Returns tuple df: dataframe collecting the information of the dataset to load.
dataset_idx: location in the previous dataframe of the particular dataset we want to generate.

source

_get_args

 _get_args (model, return_defaults=False)

Given the name of a diffusion model, return its inputs arguments.

Type Default Details
model str Name of the diffusion model (see self.available_models_name)
return_defaults bool False If True, the function will also return the default values of each input argument.
Returns tuple args (list): list of input arguments.
defaults (optional, list): list of default value for the input arguments.