Module connectome.models.framework

framework to open up saved models as well as train and save new models

Functions

def model_framework(X_train, y_train, model: str, pretrained: bool = True, model_path: str = None, save: bool = False, t_direct: str = None, **kwargs)

Function that lets the user decide what type of model he wants to use for his data and which parameters to use. See the documentation of the modelling functions for more information on further parameters that can be specified.

Examples:

>>> # CNN:
>>> # Different optimizer and loss function
>>> import tensorflow as tf
>>> from tf.keras.optimizers import Nadam
>>> from tf.keras.losses import Hinge
>>> model = model_framework(X_train = X_train,
                            y_train = y_train,
                            model = "cnn",
                            pretrained = False,
                            model_path = None,
                            epochs = 500,
                            patience = 10,
                            optimizer= Nadam(),
                            loss= Hinge())
>>> 
>>> # Elastic Net:
>>> # classification task:
>>> model = model_framework(X_train = X_train,
                            y_train = y_train,
                            model = "elnet",
                            pretrained = False,
                            n_alphas_logreg = 5, 
                            cv_logreg = 3, 
                            l1_ratios_logreg = [0.0, 0.1])
>>> # you can transform the values before modelling, e.g. take the absolute values (remember to also transform X_test for evaluation):
>>> X_train = prepare_data_elastic_net(data = X_train,
                                                     option = "abs")
>>> X_test = prepare_data_elastic_net(data = X_test,
                                                     option = "abs")
>>> # regression task:
>>> model = model_framework(X_train = X_train,
                            y_train = y_train,
                            model = "elnet",
                            pretrained = False,
                            n_alphas_linreg = 20,
                            cv_linreg = 5, 
                            l1_ratios_linreg = [0.01, 0.1])
>>> # you don't have to specifiy n_alphas_logreg/linreg, cv_logreg/linreg or l1_ratios_logreg/linreg and take the default values instead (see documentation of elastic net function)
>>> 
>>> # Random Forest:
>>> model = model_framework(X_train = X_train,
                            y_train = y_train,
                            model = "rf",
                            pretrained = False,
                            n_estimators = 100)  

Args

X_train

The training dataset

y_train

The true labels

pretrained

If a new model should be trained or a pretrained model should be used (True/False, default True uses pretrained model

model

A string to enter which model to use ("elnet" = elastic net, "gboost" = gradient boosting, "rf" = random forest, "cnn" = convolutional neural network, "ebm" = explainable boosting machine)

model_path

Full path to the folder where the selected pretrained model is stored

save

If the newly trained model should be saved

t_direct

Target directory to save the model in

**kwargs

Additional parameters and options depending on the selected model

Returns

A fitted model

Raises

FileNotFoundError IOError