Feature Engineering for Channel Allocation

In reinforcement learning for dynamic channel allocation problems (such as wireless communication networks), the raw state is typically a binary occupancy matrix showing which channels are in use at which cells. To use function approximation methods, we need to extract meaningful features from this raw state. Implement a function that takes a channel allocation state and cell adjacency information, and produces a feature vector suitable for RL value function approximation. The state is a binary matrix of shape (n cells, n channels) where 1 indicates a channel is in use. The adjacency matrix of shape (n cells, n cells) indicates which cells are neighbors (the diagonal is always 0). Your function should compute the following feature groups in this order : 1. Cell utilization (n cells features): For each cell, the fraction of its channels currently in use. 2. Neighbor average utilization (n cells features): For each cell, the mean utilization of its neighboring cells. If a cell has no neighbors, this value is 0. 3. Channel conflict ratio (n cells features): For each cell, the fraction of its in use channels that are also in use by at least one neighbor. If a cell has no neighbors, this is 0. 4. Jointly available channel ratio (n cells features): For each cell, the fraction of channels that are free in this cell AND free in ALL of its neighbors simultaneously. If a cell has no neighbors, count channels free in the cell only. 5. Global statistics (3 features): The overall utilization across all cells, the maximum cell utilization, and the minimum cell utilization. The total feature vector length is 4 n cells + 3. Args: state: A list of lists of shape (n cells, n channels), binary values adjacency: A list of lists of shape (n cells, n cells), binary adjacency matrix Return the feature vector as a list of floats.