:param var_collections: name of collections to store the created variables. :return: dictionary to index the created variables.

:param var_collections: name of collections to store the created variables. :return: dictionary to index the created variables.

:param var_collections: name of collections to store the created variables. :return: dictionary to index the created variables.

Return a standard feed dictionary for this dataset. :param name: if not None, register this supplier in dict NAMED_SUPPLIERS (this c

:param pred: output of the neural networks :param label: the true label paired with the input :return: cross-entropy loss function

:return: The data dimensionality as an integer, if input are vectors, or a tuple in the general case

:return: The target dimensionality as an integer, if targets are vectors, or a tuple in the general case

Dot product between vectors `a` and `b` with optional name. If a and b are not vectors, formally this computes <vec(a), vec(b)>.

:return: A generator for the dynamics (state_variable_{k+1})

Generate the feed_dicts for boml_optimizer.run() with data_batch and the instance of BOMLExperiment :param data_batch: each batch of data for

Generate the feed_dicts for boml_optimizer.run() with data_batch and the instance of BOMLExperiment :param data_batch: each batch of data for

Generate the feed_dicts for boml_optimizer.run() with lists of :param dat_lst: :param exs: :return:

Return the L1 regularization item :param var_list: list of parameters :param rate: regularization rate :return: L1_regularization_par

Return the L1 regularization item :param var_list: list of parameters :param rate: regularization rate :return: L2_regularization_par

:return:

:param boml_net: instance of BOMLNet :param i: int32, i-th layer to be defined :param initializer: function to initialize the bias :r

:param boml_net: instance of BOMLNet :param i: int32, the i-th layer to be defined :param initializer: function to initialize the weights

:param dim: dimension of the identity matrix :param name: name for the variable :param conv: BOOLEAN, define variables as matrix or vecto

Creates an hyperparameter variable, which is a GLOBAL_VARIABLE and HYPERPARAMETER. Mirrors the behavior of `tf.get_variable`. :param nam

Utility methods for getting a `RandomState` object. :param rand: rand can be None (new State will be generated), np.rand

Utility methods for getting a `RandomState` object. :param rand: rand can be None (new State will be generated), np.rand

Utility methods for getting a `RandomState` object. :param rand: rand can be None (new State will be generated), np.rand

:param boml_net: instance of BOMLNet :param layer: int32, i-th layer to be defined :param initializer: function to initializer the weight

:param boml_net: instance of BOMLNet :param layer: int32, i-th layer to be defined :param initializer: function to initializer the weight

:return: globalstep used in whole graph.

List of tensors and/or variables in the collection OUTERGRADIENTS. These variables are created by `far.HyperGradient`. :param scope: (s

:return: The initialization dynamics if it has been set, or `None` otherwise.

:return: a list of operations that return the values of the state variables for this learning dynamics after the executio

Initialize the model. If `deterministic_initialization` is set to true, saves the initial weight in numpy which will be used for subs

:return: the innergradient object underlying this wrapper.

Performs a descent step (as return by `tf.train.Optimizer.apply_gradients`) and computes the values of the variables after it.

List of variables in the collection LAGRANGIAN_MULTIPLIERS. These variables are created by `far.ReverseHG`. :param scope: (str) an opti

Implements leaky relu with negative coefficient `alpha`

:return: the outergradient object underlying this wrapper.

:param file_path: path to save the pickle file :return:

return a if b is None else a + b

Calls obj with args and kwargs and return its result if obj is callable, otherwise returns obj.

Calls obj with args and kwargs and return its result if obj is callable, otherwise returns obj.

Run or eval `a` and returns the result if possible. :param a: object, or `tf.Variable` or `tf.Tensor` :param ss: `tf.Session` or get def

:param obj: object :param i: the index :return: the i-th item if the `obj` instantiates the __getitem__ function

Computes mean, standard deviation and 95% half-confidence interval for a list of measures. :param measures: list :param mul: optional mu

Computes mean, standard deviation and 95% half-confidence interval for a list of measures. :param measures: list :param mul: optional mu

Merges dictionaries recursively. Accepts also `None` and returns always a (possibly empty) dictionary

:return: the outergradient object underlying this wrapper.

Load a meta-datasets from Mini-ImageNet. Returns a Datasets of MetaDataset s, :param folder: base folder :param sub_folders: optional su

:return: the created BMLNet object

Loading function for Omniglot dataset in learning-to-learn version. Use image data as obtained from https://github.com/cbfinn/maml/blob/maste

:return: the method for whole algorithm.

:return: temporary weights dictionary used in maml and fomaml for back propagation

:param pred: output of the neural networks :param label: the true label paired with the input :return: msw loss function

Unfortunately this method doesn't return a very specific name....It gets a little messy :param var_dict: :param vars_: :return:

:return: Number of examples in this dataset

Extract from omniglot dataset with rotated images, 100 classes, 3 examples per class in training set 3 examples per class in validation s

:return: the current output of the BOMLNet

:return: temporary weights dictionary used in maml and fomaml for back propagation

List of variables in the collection HYPERPARAMETERS. Hyperparameters constructed with `get_outerparameter` are in this collection by default

:return: the outergradient object underlying this wrapper.

:return: dict that holds hyper_params used in the inner optimization process.

reuses defined convolutional networks with new input and update the output results :param new_input: new input with same shape as the

reuses defined convolutional networks with new input and update the output results :param new_input: new input with same shape as the

:param file_path: path to save the pickle file :param obj: :return:

Stack some datasets calling stack for each dataset. :param datasets_s: :return: a new dataset