`ibmfl.model`¶

Base Class¶

class ibmfl.model.fl_model.FLModel(model_type, model_spec, **kwargs)[source]¶

Base class for all FLModels. This includes supervised and unsupervised ones.

get_model_absolute_path()[source]¶

Construct absolute path using MODEL_DIR env variable

Parameters: filename (str) – Name of file
Returns: absolute_path; constructed absolute path using model_dir
Return type: str

get_gpu_config_auto()[source]¶

Extract num of gpus and gpu ids automatically using gpu util

Parameters: gpu_config (dict) – Provided gpu config
Returns: num_gpus, gpu_ids
Return type: int, str

get_gpu_config_manual()[source]¶

Extract num of gpus and gpu ids automatically using gpu util

Parameters: gpu_config (dict) – Provided gpu config
Returns: num_gpus, gpu_ids
Return type: int, str

Decision Tree¶

class ibmfl.model.dt_fl_model.DTFLModel(model_name, model_spec, dt_model=None, latest_counts=None)[source]

Class implementation for basic decision tree

__init__(model_name, model_spec, dt_model=None, latest_counts=None, **kwargs)[source]¶

Create a 'DTFLModel' instance for decision tree model. For the given model_spec, it will extract the data information required to grow the tree including: list_of_features, feature_values, list_of_labels. If an initial tree structure is given, model_spec is optional.

Parameters

model_name (str) – String specifying the name of the model
model_spec (dict) – Specification of the tree model
dt_model (dict) – A tree structure
latest_counts (list) – A list of last_counts saved from last query call, served as party model_update information
kwargs (dict) – A dictionary contains other parameter settings on to initialize a decision tree model.

Returns

None

fit_model(train_data, fit_params, **kwargs)[source]¶

Computes the count according to split information and provided training data

Parameters

train_data (pandas.core.frame.DataFrame) – Training dataset
fit_params (dict) – A dictionary containing the aggregator query information, e.g., ‘splits’, which is a list of node split candidates in the current tree; ‘list_of_labels’, which is a list of labels needs count information
kwargs (dict) – Training dataset

Returns

None

update_model(model_update=None, new_feature_values=None, new_list_of_features=None)[source]¶

Update the feature list and their corresponding range of feature values.

Parameters

model_update (ModelUpdate) – Optional, an ModelUpdate object that contains information to update the DTFLModel.
new_feature_values (list) – Optional, new range of feature values
new_list_of_features (list) – Optional, new list of feature under consideration

Returns

None

get_model_update(self)[source]¶

Generate a ModelUpdate object that will be sent to the entities

Returns: counts - a list of counts for the latest query
Return type: ModelUpdate

predict(self, x_test, node=None)[source]¶

Perform prediction for a data instance.

Parameters

x_test (category) – A data instance in the format as expected by the model
node (dict) – A tree root to start searching for leaf node

Returns

A list of predictions

Return type

list

evaluate(self, test_dataset, **kwargs)[source]¶

Evaluates the model given testing data

Parameters

test_dataset (np.ndarray) – Testing data, a tuple given in the form (x_test, test) or a datagenerator of of type keras.utils.Sequence, keras.preprocessing.image.ImageDataGenerator
kwargs (dict) – Dictionary of metrics available for the model

Returns

dict_metrics

Return type

dict

evaluate_model(self, x, y=None, class_report=False, **kwargs)[source]¶

Evaluate decision tree given x and y

Parameters

x (pandas.core.frame.DataFrame) – Feature vectors
y (list) – Optional, corresponding labels of x, could be given in x as its last column
class_report (bool) – A boolean value indicating if classification report will be included in the evaluation results.
kwargs (dict) – Dictionary of metrics available for the model

Returns

dict_metrics

Return type

dict

save_model(filename=None, path=None)[source]¶

Save the tree to a file in the format specific to the framework requirements.

Parameters

filename (str) – Name of the file to store the tree
path (str) – Path of the folder to store the tree. If no path is specified, the tree will be stored in the default path

Returns

filename

Return type

str

print_tree(node=None, depth=0)[source]¶

Print the tree tht stores in the DTFLModel object.

Parameters

node (dict) – A given tree node to start printing the tree
depth (int) – Tree depth

Returns

print_string

Return type

str

static load_data_info_from_spec(spec)[source]¶

Loads data information from provided model_spec, where model_spec is a dict that contains three items:

1) data information, which is a dictionary that contains list_of_features, feature_values, list_of_labels, all in ‘list’ type;

2) dt_model optional, a dictionary contains the tree structure

3) last_counts optional, a list served as a place to store

Parameters: spec (dict) – Model specification contains required information, like list_of_features, feature_values, list_of_labels, to initialize a DTFLModel.
Returns: (list_of_features, feature_values, list_of_labels) where list_of_features is a list of features in training data, feature_values is a list of value ranges for each feature in the feature list, and list_of_labels is a list of possible labels for training data
Return type: tuple

Pytorch Wrapper¶

class ibmfl.model.pytorch_fl_model.PytorchFLModel(model_name, model_spec, keras_model=None)[source]¶

Wrapper class for importing a Pytorch model.

__init__(self, model_name, model_spec=None, pytorch_module=None, module_init_params=None, **kwargs)[source]¶

Create a `PytorchFLModel` instance from a Pytorch model. If pytorch_model is provided, it will use it; otherwise it will take the model_spec to create the model.

Parameters

model_name (str) – String specifying the type of model e.g., Pytorch_NN
model_spec (dict) – A dictionary specifying path to saved nn.sequence container
pytorch_module (torch.nn.Module class reference) – uninstantiated pytorch model class
module_init_params (dict) – A dictionary with the values for the model's init arguments. The key for the init parameters must be prefixed with module__ (ex. if the parameter name is hidden_size, then key must be module__hidden_size)
kwargs (dict) – A dictionary contains other parameter settings on to initialize a PyTorch model.

Returns

None

initialize_model(self, pytorch_module, optimizer, criterion, module_init_params)[source]¶

Initializes a pytorch model via skorch library.

Parameters

pytorch_module (torch.nn.Module) – uninstantiated pytorch model class
optimizer (pytorch optimizer class) – the optimizer to use
criterion (pytorch loss function class) – the loss function to use
module_init_params (dict) – A dictionary with the values for the model's init arguments. The key for the init parameters must be prefixed with module__ (ex. if the parameter name is hidden_size, then key must be module__hidden_size)

Returns

an initialized skorch model

Return type

skorch.NeuralNet

fit_model(self, train_data, fit_params=None, **kwargs)[source]¶

Fits current model with provided training data.

Parameters

model_update (ModelUpdate) – Training data, a tuple given in the form (x_train, y_train). Otherwise, input compatible with skorch.dataset.Dataset

Returns

None

update_model(self, model_update)[source]¶

Update model with provided model_update, where model_update should be generated according to `PytorchFLModel.get_model_update()`.

Parameters

model_update (ModelUpdate) – `ModelUpdate` object that contains the weights that will be used to update the model.

Returns

None

get_model_update(self)[source]¶

Generates a `ModelUpdate` object that will be sent to other entities.

Returns: ModelUpdate
Return type: ModelUpdate

predict(self, x)[source]¶

Perform prediction for a batch of inputs. Note that for classification problems, it returns the resulting probabilities.

Parameters

x (np.ndarray) – Samples with shape as expected by the model.

Returns

Array predictions

Return type

np.ndarray

predict_generator(self, dataloader)[source]¶

Performs predictions using a pytorch dataloader.

Parameters

x (torch.utils.data.DataLoader) – A pytorch dataloader with a dataset for predicting

Returns

Array predictions

Return type

np.ndarray

evaluate(self, test_dataset, eval_metrics=None, **kwargs)[source]¶

Evaluates the model given testing data.

Parameters

test_dataset (np.ndarray) – Testing data, a tuple given in the form (x_test, y_test) or a pytorch DataLoader
eval_metrics (sklearn.metrics or function) – A list of sklearn.metric class, or a function for evaluating a pytorch dataloader batch
kwargs (dict) – Dictionary of metrics available for the model

Returns

metric_dict - Dictionary of metrics

Return type

dict

evaluate_model(self, x, y, eval_metrics=None, **kwargs)[source]¶

Evaluates the model given x and y.

Parameters

x (np.ndarray) – Samples with shape as expected by the model.
y (np.ndarray) – Corresponding labels to x.
eval_metrics (sklearn.metrics or function) – A list of sklearn.metric class or a functions with the signature (y_true, y_pred, **kwargs) => float.

Returns

metric_dict - Dictionary of metrics

Return type

dict

evaluate_generator_model(self, dataloader, eval_metrics=None, **kwargs)[source]¶

Evaluates the model based on the provided dataloader.

Parameters

dataloader (torch.utils.data.DataLoader) – A pytorch dataloader with test dataset and labels
eval_metrics (function) – a function for how to evaluate the batched examples and labels must have the signature (x_batch, y_batch) => float

Returns

metric_dict - Dictionary of metrics

Return type

dict

save_model(self, filename=None, optimizer_filename=None, history_filename=None)[source]¶

Save a model to file in the format specific to the backend framework.

Parameters

filename (str) – Name of the file that contains the model to be loaded.
optimizer_filename (str) – Name of the file that contains the optimizer to be loaded.
history_filename (str) – Name of the file that contains the model history to be loaded.

Returns

None

load_model(self, pytorch_module, model_filename, optimizer_filename=None, history_filename=None, optimizer=None, module_init_params=None)[source]¶

Loads a model from disk given the specified file_name.

Parameters

pytorch_module (torch.nn.Module class reference) – Uninstantiated pytorch model class
model_filename (str) – Name of the file that contains the model to be loaded.
optimizer_filename (str) – Name of the file that contains the optimizer to be loaded.
history_filename (str) – Name of the file that contains the model history to be loaded.
optimizer (pytorch optimizer class) – The optimizer to be loaded.
module_init_params (dict) – A dictionary with the values for the model's init arguments. The key for the init parameters must be prefixed with module__ (ex. if the parameter name is hidden_size, then key must be module__hidden_size).

Returns

None

get_weights(self, to_numpy=False)[source]¶

Returns the weights of the model.

Parameters

to_numpy (boolean) – Determines whether the weights should be returned as numpy array, or tensor.

Returns

List of model weights

Return type

list

parameters_to_numpy(self, params)[source]¶

Transforms parameter tensors to numpy arrays.

Parameters

params – The parameter tensor to be transformed.

Returns

Numpy array of parameters

Return type

nd.array

load_model_from_spec(self, model_spec)[source]¶

Loads model from provided model_spec, where model_spec is a `dict` that contains one item: model_spec['model_definition'], which has a pointer to the file where an nn.sequence container is saved.

Parameters

model_spec (dict) – A dictionary contains model specification

Returns

model

Return type

nn.sequence

get_gradient(self, train_data)[source]¶

Returns the gradients for each layer in the model.

Parameters

train_data (tuple) - Training data, a tuple given in the form (x_train, y_train). otherwise, input compatible with skorch.dataset.Dataset

Returns

gradients

Return type

list

static valid_acc(net, x, y)[source]¶

Returns the gradients for each layer in the model.

Parameters

net (nn.sequence) - The model that will be used
x (nd.array) - the validation data set
y (nd.array) - the training targets

Returns

Accuracy of the validation training pass

Return type

float

static loss(y, y_pred, net)[source]¶

Returns resulting loss computed based on the provided dataset.

Parameters

y (nd.array)
y_pred (nd.array)
net (nn.sequence)

Returns

loss

static __expand_linear_layer__(net, layer_name, new_dimensions, layer_idx)[source]¶

Expands a linear layer of a pytorch module.

Parameters

net (torch.nn.Module) - an instantiated pytorch module
layer_name (str) - layer variable name which needs to be expanded
new_dimensions (list) - new dimensions
layer_idx (int) - layer idx which needs to be expanded. this corresponds with the index of `new_dimensions`

Returns

None

expand_model_by_layer_name(self, new_dimension, layer_name="dense")[source]¶

Expands the current PyTorch models layers with the provided dimensions.

Parameters

new_dimension (list) - new dimensions of the particular `layer_name`
layer_name (str) - layer variable name which needs to be expanded

Returns

None

is_fitted(self)[source]¶

Return a boolean value indicating if the model is fitted or not. In particular, it calls `skorch.utils.check_is_fitted` to checks whether the net is initialized. If it has, return True; otherwise return false.

Parameters: None
Returns: res
Return type: bool

get_loss(self, dataset)[source]¶

Return the resulting loss computed based on the provided dataset.

Parameters

train_data (tuple) - Training data, a tuple given in the form (x_train, y_train). otherwise, input compatible with skorch.dataset.Dataset

Returns

orig_loss_value - the resulting loss.

Return type

float

TensorFlow Wrapper¶

class ibmfl.model.tensorflow_fl_model.TensorFlowFLModel(model_name, model_spec, keras_model=None)[source]¶

Wrapper class for importing tensorflow models.

__init__(self, model_name, model_spec, tf_model=None, **kwargs)[source]¶

Create a `PytorchFLModel` instance from a Pytorch model. If pytorch_model is provided, it will use it; otherwise it will take the model_spec to create the model.

Parameters

model_name (str) – String specifying the type of model e.g., tf_CNN
model_spec (dict) – Specification of the `tf_model`
tf_model (tf.keras.Model) – Compiled TensorFlow model
kwargs (dict) – A dictionary contains other parameter settings on to initialize a TensorFlow model.

Returns

None

fit_model(self,train_data, fit_params=None, **kwargs)[source]¶

Fits current model with provided training data.

Parameters

train_data (nd.ndarray) – Training data, a tuple given in the form (x_train, y_train).
fit_params (dict) – (optional) Dictionary with hyperparameters that will be used to call fit function. Hyperparameter parameters should match pytorch expected values e.g., `epochs`, which specifies the number of epochs to be run. If no `epochs` or `batch_size` are provided, a default value will be used (1 and 128, respectively).

Returns

None

update_model(self, model_update)[source]¶

Update TensorFlow model with provided model_update, where model_update should be generated according to `TensorFlowFLModel.get_model_update()`.

Parameters

model_update (ModelUpdate) – `ModelUpdate` object that contains the weights that will be used to update the model.

Returns

None

get_model_update(self)[source]¶

Generates a `ModelUpdate` object that will be sent to other entities.

Returns: ModelUpdate
Return type: ModelUpdate

predict(self, x, **kwargs)[source]¶

Perform prediction for a batch of inputs. Note that for classification problems, it returns the resulting probabilities.

Parameters

x (np.ndarray) – Samples with shape as expected by the model.
kwargs (dict) – Dictionary of tf-specific arguments.

Returns

Array predictions

Return type

np.ndarray

evaluate(self, test_dataset, **kwargs)[source]¶

Evaluates the model given testing data.

Parameters

test_dataset (np.ndarray) – Testing data, a tuple given in the form (x_test, y_test) or a datagenerator of type `keras.utils.Sequence`, `keras.preprocessing.image.ImageDataGenerator`
kwargs (dict) – Dictionary of metrics available for the model

Returns

metrics

Return type

dict

evaluate_model(x, y, batch_size=128, **kwargs)[source]¶

Evaluates the model given x and y.

Parameters

x (category) – Samples with shape as expected by the model.
y (dict) – Corresponding labels to x.
batch_size (int) – Size of batches.
kwargs (dict) – Dictionary of metrics available for the model

Returns

Dictionary of metrics

Return type

dict

evaluate_generator_model(dataloader, eval_metrics=None)[source]¶

Evaluates the model based on the provided dataloader.

Parameters

test_generator (`ImageDataGenerator` or `keras.utils.Sequence`) – Testing datagenerator of type `keras.utils.Sequence`, or `keras.preprocessing.image.ImageDataGenerator`.
kwargs (function) – Dictionary of metrics available for the model

Returns

metrics

Return type

dict

load_model(file_name, custom_objects={})[source]¶

Loads a model from disk given the specified file_name.

Parameters

file_name (str) – Name of the file that contains the model to be loaded.
custom_objects (dict)

Returns

TensorFlow model loaded to memory

Return type

tf.keras.models.Model

save_model(filename=None)[source]¶

Save a model to file in the format specific to the backend framework.

Parameters

filename (str) – Name of the file where to store the model.

Returns

filename

Return type

str

model_from_json_via_tf_keras(json_file_name)[source]¶

Loads a model architecture from disk via tf.keras given the specified json file name.

Parameters

json_file_name (str) Name of the file that contains the model architecture to be loaded.

Returns

tf.keras model with only model architecture loaded to memory

Return type

tf.keras.models.Model

load_model_from_spec(model_spec)[source]¶

Loads model from provided model_spec, where model_spec is a `dict` that contains one item: model_spec['model_definition'], which has a pointer to the file where an nn.sequence container is saved.

Parameters

model_spec

Returns

model

Return type

keras.models.Model

get_gradient(train_data)[source]¶

Returns the gradients for each layer in the model.

Parameters

train_data (nd.array)

Returns

gradients

Return type

list of tf.Tensor

expand_model_by_layer_name(new_dimension, layer_name="dense")[source]¶

Expand the current Keras model with provided dimension of the hidden layers or model weights. This method by default expands the dense layer of the current neural network. It can be extends to expand other layers specified by `layer_name`, for example, it can be use to increase the number of CNN filters or increase the hidden layer size inside LSTM.

Parameters

new_dimension (list) - an instantiated pytorch module
layer_name (str) - layer variable name which needs to be expanded

Returns

None

is_fitted()[source]¶

Return a boolean value indicating if the model is fitted or not. In particular, check if the tensorflow model has weights. If it has, return True; otherwise return false.

Parameters: None
Returns: res
Return type: bool

get_loss(dataset)[source]¶

Return the resulting loss computed based on the provided dataset.

Parameters

train_data ((`nd.array`, nd.array`)) - Training data, a tuple given in the form (x_train, y_train). otherwise, input compatible with skorch.dataset.Dataset

Returns

The resulting loss.

Return type

float

Keras Wrapper¶

class ibmfl.model.keras_fl_model.KerasFLModel(model_name, model_spec, keras_model=None)[source]¶

Wrapper class for importing keras models.

fit_model(filename)[source]¶

Fits current model with provided training data.

Parameters

train_data (category) – Training data, a tuple given in the form (x_train, y_train) or a datagenerator of type keras.utils.Sequence, keras.preprocessing.image.ImageDataGenerator
train_data (nd.ndarray) – Training data, a tuple given in the form (x_train, y_train) or a datagenerator of type keras.utils.Sequence, keras.preprocessing.image.ImageDataGenerator
fit_params (dict) – (optional) Dictionary with hyperparameters that will be used to call Keras fit function. Hyperparameter parameters should match keras expected values e.g., `epochs`, which specifies the number of epochs to be run. If no `epochs` or `batch_size` are provided, a default value will be used (1 and 128, respectively).

Returns

None

fit(x_test, node=None)[source]¶

Fits current model using model.fit with provided training data.

Parameters

x_test (category) – A data instance in the format as expected by the model
node (dict) – A tree root to start searching for leaf node

Returns

A list of predictions

Return type

list

predict(x_test, node=None)[source]¶

Perform prediction for a data instance.

Parameters

x_test (category) – A data instance in the format as expected by the model
node (dict) – A tree root to start searching for leaf node

Returns

A list of predictions

Return type

list

update_model(x_test, node=None)[source]¶

Perform prediction for a data instance.

Parameters

x_test (category) – A data instance in the format as expected by the model
node (dict) – A tree root to start searching for leaf node

Returns

A list of predictions

Return type

list

get_model_update(x_test, node=None)[source]¶

Perform prediction for a data instance.

Parameters

x_test (category) – A data instance in the format as expected by the model
node (dict) – A tree root to start searching for leaf node

Returns

A list of predictions

Return type

list

evaluate(x_test, node=None)[source]¶

Perform prediction for a data instance.

Parameters

x_test (category) – A data instance in the format as expected by the model
node (dict) – A tree root to start searching for leaf node

Returns

A list of predictions

Return type

list

evaluate_model(x_test, node=None)[source]¶

Perform prediction for a data instance.

Parameters

x_test (category) – A data instance in the format as expected by the model
node (dict) – A tree root to start searching for leaf node

Returns

A list of predictions

Return type

list

load_model(x_test, node=None)[source]¶

Perform prediction for a data instance.

Parameters

x_test (category) – A data instance in the format as expected by the model
node (dict) – A tree root to start searching for leaf node

Returns

A list of predictions

Return type

list

load_model_via_keras(x_test, node=None)[source]¶

Perform prediction for a data instance.

Parameters

x_test (category) – A data instance in the format as expected by the model
node (dict) – A tree root to start searching for leaf node

Returns

A list of predictions

Return type

list

load_model_via_tf_keras(x_test, node=None)[source]¶

Perform prediction for a data instance.

Parameters

x_test (category) – A data instance in the format as expected by the model
node (dict) – A tree root to start searching for leaf node

Returns

A list of predictions

Return type

list

Differentially Private Gaussian Naive Bayes Wrapper¶

class ibmfl.model.naive_bayes_fl_model.NaiveBayesFLModel(config)[source]¶

RL Import Wrapper¶

class ibmfl.model.rllib_fl_model.RLlibFLModel(model_name, model_spec, keras_model=None)[source]¶

Wrapper class for importing keras models.

fit_model(train_data, fit_params)[source]¶

Fits current model with provided training data.

Parameters

train_data (category) – Samples with shape as expected by the model.
fit_params (dict) – (Optional) Dictionary with hyperparameters that will be used to train the model.

Returns

model

Return type

nn.sequence

update_model(model_update, **kwargs)[source]¶

Update RLlib model with provided model_update, where model_update should be generated according to `RLlibFLModel.get_model_update()`.

Parameters

model_update (category) – `ModelUpdate` object that contains the weights that will be used to update the model

Returns

None

get_model_update(filename)[source]¶

Generates a `ModelUpdate` object that will be sent to other entities.

Returns: ModelUpdate
Return type: ModelUpdate

evaluate(test_dataset, **kwargs)[source]¶

Evaluates the model given testing data.

Parameters

test_dataset (category) – esting data, a tuple given in the form (x_test, test) or a datagenerator of of type `keras.utils.Sequence`, `keras.preprocessing.image.ImageDataGenerator`
kwargs (dict) – Dictionary of metrics available for the model.

Returns

model

Return type

nn.sequence

evaluate_model(x=None, y=None, batch_size=128, **kwargs)[source]¶

Evaluates the RLlib model with test environment and latest checkpoint using RLlib rollout.

Parameters

x (category) – Samples with shape as expected by the model.
y (dict) – Corresponding labels to x.
batch_size (int) – Size of batches.
kwargs (dict) – Dictionary of metrics available for the model

Returns

Mean Episode Reward

Return type

dict

save_model(filename=None, path=None)[source]¶

Save a model to file in the ray results folder.

Parameters

filename (str) – Name of the file where to store the model.
path (str) – Path of the folder where to store the model. If no path is specified, the model will be stored in the default data location of the library `DATA_PATH`.

Returns

None

load_model(file_name)[source]¶

Loads a model from disk given the specified file_name.

Parameters

file_name (str) – Name of the file that contains the model to be loaded.

Returns

None

create_rl_trainer(env_class_ref)[source]¶

Creates RL trainer using model_spec where model_spec['model_definition'] specifies RL algorithm and env_class_ref contains environment definition.

Parameters

env_class_ref (category) – Samples with shape as expected by the model.
train_data (nd.array) – Corresponding labels to x.
test_data (nd.array) – A list of sklearn.metric class or a functions with the signature (y_true, y_pred, **kwargs) => float.

Returns

None

Sklearn Wrapper¶

class ibmfl.model.sklearn_fl_model.SklearnFLModel(connection, synch=False, max_timeout=None)[source]¶

Wrapper class for sklearn models.

__init__(connection, synch=False, max_timeout=None)[source]¶

Create a SklearnFLModel instance from a sklearn model. If sklearn_model is provided, it will use it; otherwise it will take the model_spec to create the model.

Parameters

model_name (str) – A name specifying the type of model, e.g., linear_SVM
model_spec (dict) – A dictionary contains model specification
sklearn_model (sklearn) – A compiled sklearn model
kwargs (dict) – tA dictionary contains other parameter settings on to initialize a sklearn model.

fit_model(party)[source]¶

Fits current model with provided training data.

Parameters: party (PartyConnection) – party details that are needed by the protocol handler
Return id: guide assigned to the party

update_model(model_update, **kwargs)[source]¶

Update sklearn model with provided model_update.

Parameters

model_update (category) – `ModelUpdate` object that contains the weights that will be used to update the model

Returns

None

get_model_update(filename)[source]¶

Generates a `ModelUpdate` object that will be sent to other entities.

Returns: ModelUpdate
Return type: ModelUpdate

predict(x, **kwargs)[source]¶

Evaluates the model given testing data.

Parameters

test_dataset (category) – esting data, a tuple given in the form (x_test, test) or a datagenerator of of type `keras.utils.Sequence`, `keras.preprocessing.image.ImageDataGenerator`
kwargs (dict) – Dictionary of metrics available for the model.

Returns

model

Return type

nn.sequence

evaluate_model(x=None, y=None, batch_size=128, **kwargs)[source]¶

Evaluates the model given test data x and the corresponding labels y.

Parameters

x (category) – Samples with shape as expected by the model.
y (dict) – Corresponding labels to x.
batch_size (int) – Size of batches.
kwargs (dict) – Dictionary of metrics available for the model

Returns

Mean Episode Reward

Return type

dict

save_model(filename=None, path=None)[source]¶

Save a model to file in the ray results folder.

Parameters

filename (str) – Name of the file where to store the model.
path (str) – Path of the folder where to store the model. If no path is specified, the model will be stored in the default data location of the library `DATA_PATH`.

Returns

None

load_model(file_name)[source]¶

Loads a model from disk given the specified file_name.

Parameters

file_name (str) – Name of the file that contains the model to be loaded.

Returns

None

Sklearn K Means Wrapper¶

class ibmfl.model.sklearn_kmeans_fl_model.SklearnKMeansFLModel(connection, synch=False, max_timeout=None)[source]¶

Wrapper class for sklearn.cluster.KMeans

__init__(connection, synch=False, max_timeout=None)[source]¶

Create a SklearnKMeansFLModel instance from a sklearn model. If sklearn_model is provided, it will use it; otherwise it will take the model_spec to create the model.

Parameters

model_name (str) – A name specifying the type of model, e.g., linear_SVM
model_spec (dict) – A dictionary contains model specification
sklearn_model (sklearn) – A compiled sklearn model
kwargs (dict) – tA dictionary contains other parameter settings on to initialize a sklearn model.

fit_model(party)[source]¶

Fits current model with provided training data.

Parameters: party (PartyConnection) – party details that are needed by the protocol handler
Return id: guid assigned to the party

update_model(model_update, **kwargs)[source]¶

Update sklearn model with provided model_update.

Parameters

model_update (category) – `ModelUpdate` object that contains the weights that will be used to update the model

Returns

None

get_model_update(filename)[source]¶

Generates a `ModelUpdate` object that will be sent to other entities.

Returns: ModelUpdate
Return type: ModelUpdate

predict(x, **kwargs)[source]¶

Evaluates the model given testing data.

Parameters

test_dataset (category) – testing data, a tuple given in the form (x_test, test) or a datagenerator of of type `keras.utils.Sequence`, `keras.preprocessing.image.ImageDataGenerator`
kwargs (dict) – Dictionary of metrics available for the model.

Returns

model

Return type

nn.sequence

evaluate(x=None, y=None, batch_size=128, **kwargs)[source]¶

Evaluates the model given testing data.

Parameters

test_dataset (np.ndarray) – Testing data, a tuple given in the form (x_test, test) or a datagenerator of of type `keras.utils.Sequence`, `keras.preprocessing.image.ImageDataGenerator`
kwargs (dict) – Dictionary of metrics available for the model

Returns

acc

Return type

dict

evaluate_model(x=None, y=None, batch_size=128, **kwargs)[source]¶

Evaluates the model given test data x and the corresponding labels y.

Parameters

x (category) – Samples with shape as expected by the model.
y (dict) – Corresponding labels to x.
batch_size (int) – Size of batches.
kwargs (dict) – Dictionary of metrics available for the model

Returns

acc

- Dictionary with all evaluation metrics provided by specific implementation

Return type

dict

save_model(filename=None, path=None)[source]¶

Save a model to file in the ray results folder.

Parameters

filename (str) – Name of the file where to store the model.
path (str) – Path of the folder where to store the model. If no path is specified, the model will be stored in the default data location of the library `DATA_PATH`.

Returns

None

load_model_from_spec(file_name)[source]¶

Loads model from provided model_spec, where model_spec is a `dict` that contains the following items: model_spec['model_definition'] contains the model definition as type sklearn.cluster.KMeans

Parameters

model_spec (dict) – Model specification contains a compiled sklearn model.

Returns

model

Return type

sklearn.cluster

SGD Classifier Wrapper¶

class ibmfl.model.sklearn_SGD_linear_fl_model.SklearnSGDFLModel(config)[source]¶

Prejudice Remover¶

class ibmfl.model.prej_remover_fl_modelPrejRemoverFLModel(config)[source]¶

ibmfl.model¶

Base Class¶

Decision Tree¶

Pytorch Wrapper¶

TensorFlow Wrapper¶

Keras Wrapper¶

Differentially Private Gaussian Naive Bayes Wrapper¶

RL Import Wrapper¶

Sklearn Wrapper¶

Sklearn K Means Wrapper¶

SGD Classifier Wrapper¶

Prejudice Remover¶

`ibmfl.model`¶