N-HiTS › Visualizing NHITS

use_mls_style()

# Demonstrate neural interpolation concept
class SimpleNHITSBlock(nn.Module):
    """Simplified NHITS block for demonstration"""

    def __init__(self, input_size, hidden_size, output_size):
        super().__init__()
        self.encoder = nn.Linear(input_size, hidden_size)
        self.processor = nn.Linear(hidden_size, hidden_size)
        self.decoder = nn.Linear(hidden_size, output_size)
        self.activation = nn.ReLU()

    def forward(self, x):
        h = self.activation(self.encoder(x))
        h = self.activation(self.processor(h))
        return self.decoder(h)


# Create a simple example
input_window = 10
forecast_horizon = 5
hidden_size = 32


# Generate training data
def create_sequences(data, window_size, forecast_horizon):
    X, y = [], []
    for i in range(len(data) - window_size - forecast_horizon + 1):
        X.append(data[i : i + window_size])
        y.append(data[i + window_size : i + window_size + forecast_horizon])
    return np.array(X), np.array(y)


# Create sequences for different hierarchical levels
X_orig, y_orig = create_sequences(x, input_window, forecast_horizon)
X_level2, y_level2 = create_sequences(
    hierarchical_data["Level 2"], input_window // 2, forecast_horizon // 2
)

# Convert to tensors
X_orig_tensor = torch.FloatTensor(X_orig)
y_orig_tensor = torch.FloatTensor(y_orig)
X_level2_tensor = torch.FloatTensor(X_level2)
y_level2_tensor = torch.FloatTensor(y_level2)

print(
    f"Original level - Input shape: {X_orig_tensor.shape}, Target shape: {y_orig_tensor.shape}"
)
print(
    f"Level 2 - Input shape: {X_level2_tensor.shape}, Target shape: {y_level2_tensor.shape}"
)