fivedreg Package

class fivedreg.DataLoader(data_path: str)

Bases: object

__init__(data_path: str)

Initialize DataLoader with path to pickle file.

Args:

data_path: Path to the pickle file containing the dataset dictionary

load_data() → Dict[str, Any]

Load data from pickle file.

Expected format: {

‘X’: numpy array of shape (n_samples, n_features), ‘y’: numpy array of shape (n_samples,), ‘metadata’: dict with dataset information

}

Returns:

Dictionary containing ‘X’, ‘y’, and ‘metadata’ keys

get_data_summary(data: Dict[str, Any]) → Dict[str, Any]

Get a structured summary of the data dictionary.

Args:

data: Dictionary with ‘X’, ‘y’, and ‘metadata’ keys

Returns:

Dictionary containing structured summary data

class fivedreg.LightweightNN(hidden_layers: List[int] = [64, 32, 16], learning_rate: float = 0.001, max_iter: int = 1000, activation: str = 'relu', output_activation: str | None = None, random_state: int | None = None, verbose: int = 0)

Bases: object

A lightweight neural network using TensorFlow with configurable architecture.

Designed for fast training on CPU (< 1 minute for datasets up to 10,000 samples). Works with 5D pandas DataFrame input data.

Parameters:

• hidden_layers (list of int, default=[64, 32, 16]) – Number of neurons in each hidden layer. The number of layers is determined by the length of this list.

• learning_rate (float, default=0.001) – Learning rate for the optimizer.

• max_iter (int, default=1000) – Maximum number of training iterations (epochs).

• activation (str, default='relu') – Activation function for hidden layers (‘relu’, ‘tanh’, ‘sigmoid’).

• output_activation (str, default=None) – Activation function for output layer. None means linear activation.

• random_state (int, default=None) – Random seed for reproducibility.

• verbose (int, default=0) – Verbosity level. 0 = silent, 1 = progress updates.

model

The compiled TensorFlow Keras model.

Type:

tf.keras.Model

is_fitted_

Whether the model has been fitted to data.

Type:

bool

input_dim_

Number of input features (determined during fit).

Type:

int

output_dim_

Number of output features (determined during fit).

Type:

int

Examples

>>> import pandas as pd
>>> import numpy as np
>>> from fiveD_NN.neural_network import LightweightNN
>>>
>>> # Create sample 5D data
>>> X = pd.DataFrame(np.random.randn(1000, 5))
>>> y = pd.DataFrame(np.random.randn(1000, 1))
>>>
>>> # Initialize and train model
>>> model = LightweightNN(hidden_layers=[64, 32, 16], learning_rate=0.001, max_iter=500)
>>> model.fit(X, y)
>>>
>>> # Make predictions
>>> predictions = model.predict(X)

__init__(hidden_layers: List[int] = [64, 32, 16], learning_rate: float = 0.001, max_iter: int = 1000, activation: str = 'relu', output_activation: str | None = None, random_state: int | None = None, verbose: int = 0)

reset_keras()

Reset the Keras session to clear memory and avoid memory leaks.

fit(X: DataFrame | ndarray, y: DataFrame | ndarray, early_stopping: bool = True, vs: float | None = None) → LightweightNN

Train the neural network on the provided data.

Parameters:

• X (pd.DataFrame or np.ndarray) – Training input features (n_samples, n_features). For 5D_NN package, this should be 5 features.

• y (pd.DataFrame or np.ndarray) – Training target values (n_samples, n_outputs).

• early_stopping (bool, default=True) – Whether to use early stopping during training. If True, training will stop early if the loss does not improve for a certain number of epochs.

• vs (float, optional, default=None) – Validation split ratio. If None, no validation split is performed. If a float, indicates the proportion of data to use for validation (must be < 1.0).

Returns:

self – Returns self for method chaining.

Return type:

LightweightNN

Examples

>>> import pandas as pd
>>> import numpy as np
>>> from fiveD_NN.neural_network import LightweightNN
>>>
>>> X = pd.DataFrame(np.random.randn(1000, 5))
>>> y = pd.DataFrame(np.random.randn(1000, 1))
>>> model = LightweightNN(max_iter=500)
>>> model.fit(X, y)
>>> # Disable early stopping
>>> model.fit(X, y, early_stopping=False)
>>> # Use validation split
>>> model.fit(X, y, vs=0.2)

get_history() → History

Get the training history.

Returns:

history – Training history.

Return type:

keras.callbacks.History

predict(X: DataFrame | ndarray) → ndarray

Make predictions using the trained model.

Parameters:

X (pd.DataFrame or np.ndarray) – Input features for prediction (n_samples, n_features).

Returns:

predictions – Predicted values (n_samples, n_outputs).

Return type:

np.ndarray

Raises:

ValueError – If the model has not been fitted yet.

Examples

>>> import pandas as pd
>>> import numpy as np
>>> from fiveD_NN.neural_network import LightweightNN
>>>
>>> X_train = pd.DataFrame(np.random.randn(1000, 5))
>>> y_train = pd.DataFrame(np.random.randn(1000, 1))
>>> X_test = pd.DataFrame(np.random.randn(100, 5))
>>>
>>> model = LightweightNN(max_iter=500)
>>> model.fit(X_train, y_train)
>>> predictions = model.predict(X_test)

set_params(**params)

Set model parameters.

Parameters:

**params (dict) – Parameter names and values to set.

Returns:

self – Returns self for method chaining.

Return type:

LightweightNN