Quickstart

HeteroSymNN follows a Scikit-Learn style API. Here is how to create a “Cocktail Layer” that mixes periodic and linear features.

Basic Training Example

from HeteroSymNN.API.wrapper import Wraper
from HeteroSymNN.Core.Nets.neural_nets import FlexibleNN

# 1. Define the architecture and activation functions
# Note: You can mix strings like "sin(x)" with parameter tuples
model = FlexibleNN(
    nodes_structure=[10, 25, 25, 1],
    activation_config=[
        "sin(x)",
        "num",
        ("tanh(z)*a", {"a": 2})
    ],
    training_mode="mini-batch",
    batch_size=32,
    num_treaning_iter=200
)

# 2. Wrap the model for training (Regression mode)
agent = Wraper(model, work_type="reg")

# 3. Load data and train
agent.fit(X_train, y_train)