fixed issues with Pytorch in tests. Now NARNN must have tests too

sysidentpy/neural_network/tests/test_narxnn.py

@@ -0,0 +1,100 @@
+import numpy as np
+import torch
+from numpy.testing import assert_almost_equal, assert_equal, assert_raises
+
+from sysidentpy.basis_function import Fourier, Polynomial
+from sysidentpy.neural_network import NARXNN
+from sysidentpy.utils.generate_data import get_siso_data
+
+
+def create_test_data(n=1000):
+    # np.random.seed(42)
+    # x = np.random.uniform(-1, 1, n).T
+    # y = np.zeros((n, 1))
+    theta = np.array([[0.6], [-0.5], [0.7], [-0.7], [0.2]])
+    # lag = 2
+    # for k in range(lag, len(x)):
+    #     y[k] = theta[4]*y[k-1]**2 + theta[2]*y[k-1]*x[k-1] + theta[0]*x[k-2] \
+    #         + theta[3]*y[k-2]*x[k-2] + theta[1]*y[k-2]
+
+    # y = np.reshape(y, (len(y), 1))
+    # x = np.reshape(x, (len(x), 1))
+    # data = np.concatenate([x, y], axis=1)
+    data = np.loadtxt("examples/datasets/data_for_testing.txt")
+    x = data[:, 0].reshape(-1, 1)
+    y = data[:, 1].reshape(-1, 1)
+    return x, y, theta
+
+
+x, y, _ = create_test_data()
+train_percentage = 90
+split_data = int(len(x) * (train_percentage / 100))
+
+X_train = x[0:split_data, 0]
+X_test = x[split_data::, 0]
+
+y1 = y[0:split_data, 0]
+y_test = y[split_data::, 0]
+y_train = y1.copy()
+
+y_train = np.reshape(y_train, (len(y_train), 1))
+X_train = np.reshape(X_train, (len(X_train), 1))
+
+y_test = np.reshape(y_test, (len(y_test), 1))
+X_test = np.reshape(X_test, (len(X_test), 1))
+
+
+def test_default_values():
+    default = {
+        "ylag": 1,
+        "xlag": 1,
+        "model_type": "NARMAX",
+        "batch_size": 100,
+        "learning_rate": 0.01,
+        "epochs": 200,
+        # "loss_func": "mse_loss",
+        "optimizer": "Adam",
+        "net": None,
+        "train_percentage": 80,
+        "verbose": False,
+        "optim_params": None,
+        # "device": "cpu",
+    }
+    model = NARXNN(basis_function=Polynomial())
+    model_values = [
+        model.ylag,
+        model.xlag,
+        model.model_type,
+        model.batch_size,
+        model.learning_rate,
+        model.epochs,
+        # model.loss_func,
+        model.optimizer,
+        model.net,
+        model.train_percentage,
+        model.verbose,
+        model.optim_params,
+        # model.device,
+    ]
+    assert list(default.values()) == model_values
+
+
+def test_validate():
+    assert_raises(ValueError, NARXNN, ylag=-1, basis_function=Polynomial(degree=1))
+    assert_raises(ValueError, NARXNN, ylag=1.3, basis_function=Polynomial(degree=1))
+    assert_raises(ValueError, NARXNN, xlag=1.3, basis_function=Polynomial(degree=1))
+    assert_raises(ValueError, NARXNN, xlag=-1, basis_function=Polynomial(degree=1))
+
+
+def test_fit_raise():
+    assert_raises(
+        ValueError,
+        NARXNN,
+        basis_function=Polynomial(degree=1),
+        model_type="NARARMAX",
+    )
+
+
+def test_fit_raise_y():
+    model = NARXNN(basis_function=Polynomial(degree=2))
+    assert_raises(ValueError, model.fit, X=X_train, y=None)