Source code for or_suite.agents.ambulance.median_sklearn

import numpy as np
import sys
import sklearn_extra.cluster

import sys
from .. import Agent


[docs]class median_sklearnAgent(Agent): """ Agent that implements a k-medoid heuristic algorithm for the metric ambulance environment Methods: reset() : Clears data and call_locs which contain data on what has occurred so far in the environment update_config() : (UNIMPLEMENTED) pick_action(state, step) : Locations are chosen by finding the k-medoids in the accumulated arrival data, where k is the number of ambulances, using sci-kit learn's k-medoids algorithm Attributes: epLen: (int) number of time steps to run the experiment for data: (float list list) a list of all the states of the environment observed so far call_locs: (float list) the locations of all calls observed so far """
[docs] def __init__(self, epLen): """ Args: epLen: (int) number of time steps to run the experiment for """ self.epLen = epLen self.data = [] self.call_locs = []
[docs] def reset(self): # Resets data and call_locs arrays to be empty self.data = [] self.call_locs = []
[docs] def update_obs(self, obs, action, reward, newObs, timestep, info): '''Adds newObs, the most recently observed state, to data adds the most recent call arrival, found in info['arrival'] to call_locs.''' # Adds the most recent state obesrved in the environment to data self.data.append(newObs) # Adds the most recent arrival location observed to call_locs self.call_locs.append(info['arrival']) return
[docs] def update_policy(self, k): '''Update internal policy based upon records. Not used, because a greedy algorithm does not have a policy.''' # Greedy algorithm does not update policy self.greedy = self.greedy
[docs] def greedy(self, state, timestep, epsilon=0): """ For the first iteration, choose the starting state After that, choose locations for the ambulances that are most centrally located to the locations of previous calls using the k-medoids algorithm For more details about the k-medoids algorithm, see the readme document for the ambulance environment or the sci-kit learn documentation """ num_ambulance = len(self.data[0]) action = [] if len(self.call_locs) > num_ambulance: reshaped_call_locs = np.asarray(self.call_locs).reshape(-1, 1) clusters = sklearn_extra.cluster.KMedoids( n_clusters=num_ambulance, max_iter=50).fit(reshaped_call_locs) action = np.asarray(clusters.cluster_centers_).reshape(-1,) else: action = np.full(num_ambulance, np.median(self.call_locs)) return action
[docs] def pick_action(self, state, step): action = self.greedy(state, step) return action