tutorial.md

1. Node & Environment Variables

Argument	Type	Default	Description
--node_name	str	"./"	Node identifier for certificate management
--local_port	int	8081	Local port for client-server communication
--sandbox_path	str	"/sandbox"	Path to the client sandbox
--certs_path	str	"/certs"	Path to SSL certificates
--data_path	str	"/data"	Path to the dataset
--production_mode	str	"True"	Enable production mode (minimal logs)
--experiment_name	str	"experiment_1"	Folder to store experiment outputs

2. Dataset Variables

Argument	Type	Default	Description
--dataset	str	"dt4h_format"	Dataset loader type
--data_id	str	"data_id.parquet"	Dataset filename
--normalization_method	str	"IQR"	Normalization method: IQR, MIN_MAX, STD
--train_labels	list[str]	None	List of feature columns for training
--target_labels	list[str]	None	List of target columns (labels)
--train_size	float	0.7	Fraction for training dataset
--validation_size	float	0.2	Fraction for validation dataset
--test_size	float	0.1	Fraction for testing dataset

3. Training Variables

Argument	Type	Default	Description
--num_rounds	int	50	Number of federated iterations
--lr	float	1e-3	Learning rate (when applicable)
--checkpoint_selection_metric	str	"precision"	Metric used to select checkpoint models
--seed	int	42	Random seed
--num_clients	int	1	Number of clients in federation (informative only)

4. General Model Variables

Argument	Type	Default	Description
--model	str	"random_forest"	Model type (random_forest, xgb, nn, linear, cox)
--n_feats	int	0	Number of input features (can be inferred from dataset)
--n_out	int	0	Number of output features (depends on target type)
--task	str	"None"	Task type (classification, regression, survival)
--device	str	"cpu"	Device for training (cpu, cuda)
--local_epochs	int	10	Number of epochs per federated round
--batch_size	int	8	Batch size for training
--penalty	str	"none"	Regularization penalty: none, l1, l2, elasticnet, smooth l1

5. Model-Specific Variables Linear Models / SVR

Argument	Type	Default	Description
--solver	str	"saga"	Optimization solver
--l1_ratio	float	0.5	L1-L2 ratio for ElasticNet
--max_iter	int	100000	Maximum optimizer iterations
--tol	float	0.001	Tolerance for optimization
--kernel	str	"linear"	Kernel for SVR (linear, poly, rbf, sigmoid)
--degree	int	3	Polynomial degree (for poly kernel)
--gamma	str	"scale"	Gamma parameter (for rbf / poly kernels)
Random Forest
	Argument	Type	Default
------------------------	----	---------------	------------------------------
--balanced	str	"True"	Balanced class weights
--n_estimators	int	100	Number of trees
--max_depth	int	2	Maximum tree depth
--class_weight	str	"balanced"	Class weight strategy
--levelOfDetail	str	"DecisionTree"	Logging detail level
--regression_criterion	str	"squared_error"	Criterion for regression trees

XGBoost

Argument	Type	Default	Description
--booster	str	"gbtree"	Booster type (gbtree, gblinear, dart)
--tree_method	str	"hist"	Tree construction method (exact, hist)
--train_method	str	"bagging"	Training method (bagging, cyclic)
--eta	float	0.1	Learning rate

Neural Networks

Argument	Type	Default	Description
--dropout_p	float	0.0	Monte Carlo dropout probability
--T	int	20	Number of MC dropout samples

 Survival

Argument	Type	Default	Description
--time_col	str	"time"	Column containing survival times
--event_col	str	"event"	Column indicating event occurrence
--negative_duration_strategy	str	"clip"	Strategy for negative durations
--l1_penalty	float	0.0	L1 regularization penalty

1. General Settings

Argument	Type	Default	Description
--model	str	None	Model type
--task	str	None	Task type
--num_rounds	int	50	Number of federated iterations
--num_clients	int	1	Total number of clients
--min_fit_clients	int	0	Minimum clients required to perform fit
--min_evaluate_clients	int	0	Minimum clients required to perform evaluation
--min_available_clients	int	0	Minimum clients required to start a round
--seed	int	42	Random seed
--sandbox_path	str	"/sandbox"	Path to sandbox directory
--local_port	int	8081	Server listening port
--production_mode	str	"True"	Production mode (minimal logs)

2. Strategy Settings

Argument	Type	Default	Description
--strategy	str	"FedAvg"	Federated aggregation strategy
--smooth_method	str	"EqualVoting"	Weight smoothing method
--smoothing_strenght	float	0.5	Weight smoothing strength
--dropout_method	str	None	Dropout strategy for clients
--dropout_percentage	float	0.0	Ratio of dropout nodes
--checkpoint_selection_metric	str	"precision"	Metric used for checkpoint selection
--metrics_aggregation	str	"weighted_average"	Aggregation method for metrics
--experiment_name	str	"experiment_1"	Directory for experiment outputs