simulator_v3.py

# coding: utf-8

# In[ ]:

import pdb
import numpy as np
import pandas as pd
from pathgenerator import PathGenerator
import geohelper as gh
import geohash2 as Geohash

TIMESTEP = 60
GEOHASH_PRECISION = 7
REJECT_DISTANCE = 7000
MIN_TRIPTIME = 1.0 # in meters
ASSIGNMENT_SPEED = 50 # km/h (grand circle distance)



######~~REWARD_PARAMETERS~~###### 

# SERVICE_REWARD = RIDE_REWARD + TRIP_REWARD * trip_time - WAIT_COST * wait_time
# REWARD IN CODE BELOW: (B1 * btn)RIDE_REWARD*num_pass + (B2 * ctn)TRIP_REWARD * trip_time - (B3 * dtnl)WAIT_COST * wait_time  - (B5 * Ht)HOP_REWARD*num_hops

#ORIGINALLY IN CODE: RIDE_REWARD = 10.0 #5,6,7,8,9,10
RIDE_REWARD = 5 #As In Paper
TRIP_REWARD = 1.0 #As In Paper
#ORIGINALLY IN CODE: WAIT_COST = 0.05
WAIT_COST = 3.5 #As In Paper
#ORIGINALLY IN CODE: HOP_REWARD = 3.5
HOP_REWARD = 2 #As In Paper 

######~~REWARD_PARAMETERS~~###### 




def index_pull(rows,trips_test_2):
    
    val= trips_test_2[(trips_test_2['plat']==rows['plat'].item()) & (trips_test_2['plon']==rows['plon'].item())].index.values       
    return val 


class FleetSimulator(object):
    """
    FleetSimulator is an environment in which fleets mobility, dispatch
    and passenger pickup / dropoff are simulated.
    
    Init Variables:     (type) Description
        router:         (PathGenerator) Allows for finding shortest paths for distance
        eta_model:      (h5) This model predicts eta given pickup and dropoff locations
        cycle:          (int) number of cycles **to be better understood
        max_action_time (int) Action update cycle **to be better understood
    
    List of Methods:    See Individual Signatures for each of the below
        reset(self, num_vehicles, dataset, dayofweek, minofday)
        update_time(self)
        step(self, actions=None)                    ** MAIN RUNNING METHOD
        get_requests(self, num_steps, offset=0)
        get_vehicles_dataframe(self)                ** calls get_state from vehicle class
        get_vehicles_location(self)                 ** calls get_location from vehicle class
        get_vehicles_score(self)                    ** calls get_score from vehicle class
        
    """

    def __init__(self, G, eta_model, cycle, max_action_time=15):
        self.router = PathGenerator(G)
        self.eta_model = eta_model
        self.cycle = cycle
        self.max_action_time = max_action_time


    def reset(self, num_vehicles, dataset, dayofweek, minofday):
        """
        This initiates the simulator environment taking the following arguments
        Args:
            self:                   see init
            num_vehicles:   (int)   number of "resources" to initiate. 
                                    Ultimately instantiates this number of vehicle objects.
            dataset:(pd.DataFrame)  chunks loaded from experiments.py
            dayofweek:      (int)   exactly what it sounds like
            minofday:       (int)   exactly what it sounds like
        """
        self.requests = dataset
        self.current_time = 0
        self.minofday = minofday
        self.dayofweek = dayofweek
        #randomly assign initial locations based on where the first N pickups are to be made.
        init_locations = self.requests[['plat', 'plon']].values[np.arange(num_vehicles) % len(self.requests)]
        self.vehicles = [Vehicle(i, init_locations[i]) for i in range(num_vehicles)]

    def update_time(self):
        self.current_time += TIMESTEP
        self.minofday += int(TIMESTEP / 60.0)
        if self.minofday >= 1440:
            self.minofday -= 1440
            self.dayofweek = (self.dayofweek + 1) % 7

    def step(self, actions=None):
        """
        step forward the environment by TIMESTEP.
        This is the main function that updates the state space, takes actions from dqn, and returns rewards(score).
        """
        num_steps = int(self.cycle * 60.0 / TIMESTEP)

        if actions:
            self.dispatch(actions)

        requests = self.get_requests(num_steps)
        wait, reject, gas,request_hop_zero = 0, 0, 0, 0
        for _ in range(num_steps):
            for vehicle in self.vehicles:
                #Is the only purpose of vehicle.transition to get gas usage?
                gas += vehicle.transition()
            X = self.get_vehicles_location()
            W = requests[(requests.second >= self.current_time)
                                 &(requests.second < self.current_time + TIMESTEP)]
            
            if len(W) == 0:
                print("WE GOT NO REQUESTS")
                pdb.set_trace()
            assignments = self.match(X, W)
            #pdb.set_trace()
            wait_,num_vehicles,num_passengers,request_hop_zero_ = self.assign(assignments)
            #######EDITED HERE########### setting the vehicle instance variable based on assigned passengers
            #vehicle.passengers = num_passengers
            #######EDITED HERE###########
            wait += wait_
            assignment_length = len(assignments)
            reject += len(W) - num_passengers
            request_hop_zero += request_hop_zero_
            self.update_time()

        vehicles = self.get_vehicles_dataframe()
        return vehicles, requests, wait, reject, gas, request_hop_zero

    def get_requests(self, num_steps, offset=0):
        requests = self.requests[(self.requests.second >= self.current_time + offset * TIMESTEP)
                                 &(self.requests.second < self.current_time + TIMESTEP * (num_steps + offset))]
        return requests

    def get_vehicles_dataframe(self):
        
        vehicles = [vehicle.get_state() for vehicle in self.vehicles]
        #######EDITED HERE########### added curr_passengers
        vehicles = pd.DataFrame(vehicles, columns=['id', 'available', 'geohash', 'dest_geohash',
                                                   'eta', 'status', 'reward', 'lat', 'lon', 'idle','eff_dist','act_dist','curr_passengers'])
        #######EDITED HERE###########
        return vehicles

    def get_vehicles_location(self):
        vehicles = [vehicle.get_location() for vehicle in self.vehicles]
        vehicles = pd.DataFrame(vehicles, columns=['id', 'lat', 'lon', 'available'])
        return vehicles

    def get_vehicles_score(self):
        vehicles = [vehicle.get_score() for vehicle in self.vehicles]
        vehicles = pd.DataFrame(vehicles, columns=['id', 'reward', 'service_time', 'idle_time'])
        return vehicles
    
    def match(self, resources, tasks):
        """""
        Greedy Matching Algorithm for cars and passengers within a zone. 
        """""
        #non_hop_tasks = tasks[tasks['hop_flag']==0]
        #non_hop_tasks_col = non_hop_tasks['plat','plon']
        #hop_tasks = tasks[tasks['hop_flag']==1]
        
        # Getting the locations of cars that are available
        R = resources[resources.available == 1]
        
        # Amongst the requests, group the ones with similar pickup locations and get counts
        tasks_uniq = tasks.groupby(['plat','plon']).count()
        
        # Reset Index for this new counts dataframe
        tasks_uniq = tasks_uniq.reset_index(level=['plat','plon'])
        
        # Only take the lat and lon columns
        tasks_uniq = tasks_uniq[['plat','plon']]
        
        
        # Reset the index of the counts dataframe by this apply function. Not sure what its significance is. If its important, i'll come back to this.
        
        ###### EDITED THIS#########
        tasks_uniq['index']=tasks_uniq.apply(lambda rows:tasks[(tasks['plat']==rows['plat'].item()) & (tasks['plon']==rows['plon'].item())].index.values  , axis=1)
        ###### EDITED THIS#########

        
        # Calculate distances from each of the cars and unique pick up locations
        d = gh.distance_in_meters(R.lat.values,
                                  R.lon.values,
                                  tasks_uniq.plat.values[:, None],
                                  tasks_uniq.plon.values[:, None])
        # The number of assignments to make is determined by whichever (R or tasks_uniq) is lower.
        # If there are two few cars in the area, more passengers will get rejected
        N = min(len(tasks_uniq), len(R))
        

        # Create a np zero array of the number of assignments to make
        vids = np.zeros(N, dtype=int)
        
        # For each assignment, pick the closest car considering the the reject distance 
        for i in range(N):
            vid = d[i].argmin()
            if d[i, vid] > REJECT_DISTANCE:
                vids[i] = -1 #assign -1 for a rejection
            else:
                vids[i] = vid
                d[:, vid] = float('inf') #updating the d array for some reason.
        
        
        # Getting indices of accepted tasks. Seems like its those which vid>0
        index_values = tasks_uniq.index[:N][vids >= 0]
        #trips_indices = tasks.loc[index_values]['index']
        #pdb.set_trace()

        # Zip to the accepted tasks (using indices) to the resources for which vid>0.
        assignments = list(zip(tasks_uniq.loc[index_values]['index'], R['id'].iloc[vids[vids >= 0]]))

        #assignments = zip(tasks.index[:N][vids >= 0], R['id'].iloc[vids[vids >= 0]])
        return assignments


    def assign(self, assignments):
        """
        Request to vehicle assignments are already made using match(self,tasks,resources)
        This method is more of a helper function that does assignments of requests in dataset to the vehicle class.
        """
        num_vehicles = len(assignments)
        request_hop_zero = 0 
        
        #Total Number of passengers assigned in the entire step.
        num_passengers = 0
        wait = 0
        effective_d = 0 
        actual_d = 0
        for r, v in assignments:
             
            vehicle = self.vehicles[v] # pointer to a Vehicle object
            request = self.requests.loc[r]
            num_present_vehicle_pass = len(request) 
            request_hop_zero += sum(request['hop_flag']==0)
            num_passengers += num_present_vehicle_pass
            request_sort = request.sort_values(by=['trip_time'])
            first_row = request_sort.iloc[0]
            last_row = request_sort.iloc[-1]
            #print (last_row)
            ploc = (first_row.plat, first_row.plon)
            dloc = (last_row.dlat, last_row.dlon)
            #dloc = (last_row.dlat, last_row.dlon)
            vloc = vehicle.location
            #print (vloc[0], vloc[1], ploc[0], ploc[1])

            d = gh.distance_in_meters(vloc[0], vloc[1], ploc[0], ploc[1])
            # wait_time = 1 + d / (ASSIGNMENT_SPEED * 1000 / 60)
            wait_time = (d * 2 / 1.414) / (ASSIGNMENT_SPEED * 1000 / 60)
            #print (d)
            ##calculation for the trip time , effective distance, total distance
            eff_distance = sum(request.trip_distance)
            
            request_sort_copy = request_sort.copy()
            request_sort.loc[:,'index_val'] = range(len(request_sort))
            request_sort_copy.loc[:,'index_val'] = request_sort['index_val'].values -1
            request_sort_copy = request_sort_copy[['dlat','dlon','index_val']]
            request_sort_join = pd.merge(request_sort, request_sort_copy, how='left', on=None, left_on='index_val', right_on='index_val',
                                         left_index=False, right_index=False, sort=False,suffixes=('_x', '_y'), 
                                         copy=True, indicator=False,validate=None)
            
            request_join_no_na = request_sort_join.dropna()
            #print (len(request_join_no_na))
            #print (request_)
            
            #request_join_no_na['dist_bw_dest'] = gh.distance_in_meters(request_join_no_na.dlat_x,request_join_no_na.dlon_x,
             #                                                          request_join_no_na.dlat_y,request_join_no_na.dlon_y)
            
            if len(request_join_no_na) > 0:
                request_join_no_na['dist_bw_dest'] = gh.distance_in_meters(request_join_no_na.dlat_x,request_join_no_na.dlon_x,request_join_no_na.dlat_y,request_join_no_na.dlon_y)
                
                actual_distance = first_row.trip_distance + (sum(request_join_no_na.dist_bw_dest))/1000
                trip_time = (actual_distance/ASSIGNMENT_SPEED)*60
                #print ('shared....')
                #print (type(actual_distance),type(trip_time))
                #print (trip_time)
            else:
                actual_distance = first_row.trip_distance
                trip_time = first_row.trip_time
                #print ('not shared....')
                #print (type(actual_distance),type(trip_time))
                #print (trip_time)
            
            #trip_time = request.trip_time
            vehicle.start_service(dloc, wait_time, trip_time,actual_distance,eff_distance,num_present_vehicle_pass)
            wait += wait_time
#             effective_d +=eff_distance
#             actual_d += actual_distance

        return wait,num_vehicles,num_passengers,request_hop_zero

    def dispatch(self, actions):
        cache = []
        distances = []
        vids, targets = zip(*actions)
        vlocs = [self.vehicles[vid].location for vid in vids]
        for vloc, tloc in zip(vlocs, targets):
            try:
                p, d, s, t = self.router.map_matching_shortest_path(vloc, tloc)
                cache.append((p, s, t))
                distances.append(d)
            except:
                start_lat = vloc[0]
                start_lon = vloc[1]
                dest_lat = tloc[0]
                dest_lon = tloc[1]
                d = gh.distance_in_meters(start_lat,start_lon, dest_lat,dest_lon)
                distances.append(d)

        N = len(vids)
        X = np.zeros((N, 7))
        X[:, 0] = self.dayofweek
        X[:, 1] = self.minofday / 60.0
        X[:, 2:4] = vlocs
        X[:, 4:6] = targets
        X[:, 6] = distances
        trip_times = self.eta_model.predict(X)

        for i, vid in enumerate(vids):
            if trip_times[i] > MIN_TRIPTIME:
                #p, s, t = cache[i]
                #step = distances[i] / (trip_times[i] * 60.0 / TIMESTEP)
                #trajectory = self.router.generate_path(vlocs[i], targets[i], step, p, s, t)
                eta = min(trip_times[i], self.max_action_time)
                self.vehicles[vid].route([], eta)
        return


class Vehicle(object):
    """
            Status      available   location    eta         storage_id
    WT:     waiting     1           real        0           0
    MV:     moving      1           real        >0          0
    SV:     serving     0           future      >0          0
    ST:     stored      0           real        0           >0
    CO:     carry-out   0           real        >0          r>0
    """
    def __init__(self, vehicle_id, location):
        self.id = vehicle_id
        self.status = 'WT'
        self.location = location
        self.zone = Geohash.encode(location[0], location[1], precision=GEOHASH_PRECISION)
        self.available = True
        self.trajectory = []
        self.eta = 0
        self.idle = 0
        self.total_idle = 0
        self.total_service = 0
        self.reward = 0
        self.effective_d =0
        self.actual_d = 0
        ####### EDITED HERE ########
        self.passengers = 0
        self.cap = 3
        ####### EDITED HERE ########

    def update_location(self, location):
        lat, lon = location
        self.location = (lat, lon)
        self.zone = Geohash.encode(lat, lon, precision=GEOHASH_PRECISION)

    def transition(self):
        cost = 0
        if self.status != 'SV':
            self.idle += TIMESTEP/60.0

        if self.eta > 0:
            time = min(TIMESTEP/60.0, self.eta)
            self.eta -= time
            # moving
            #if self.trajectory:
            if self.status == 'MV':
                #self.update_location(self.trajectory.pop(0))
                cost = time
                self.reward -= cost


        if self.eta <= 0:
            # serving -> waiting
            if self.status == 'SV':
                self.available = True
                self.status = 'WT'
            # moving -> waiting
            elif self.status == 'MV':
                self.status = 'WT'

        return cost
#vehicle.start_service(dloc, wait_time, trip_time,actual_distance,eff_distance,num_present_vehicle_pass)
    def start_service(self, destination, wait_time, trip_time,actual_distance,eff_distance,num_pass):
        #print (type(wait_time))
        #print (type(trip_time))
        #print type(destination)
        
        if not self.available:
            print ("The vehicle #%d is not available for service." % self.id)
            return False
        self.available = False
        self.update_location(destination)
        self.total_idle += self.idle + wait_time
        ######EDITED HERE########
        if actual_distance == 0:
            num_hops = actual_distance
        else:
            num_hops = eff_distance / actual_distance
        ######EDITED HERE########
        self.idle = 0
        self.eta = wait_time + trip_time
        self.total_service += trip_time
        ######### EDITED HERE #########
        self.passengers = num_pass
        ######### EDITED HERE #########
        #ORIGINAL : self.reward += RIDE_REWARD*num_pass + TRIP_REWARD * trip_time - WAIT_COST * wait_time  - HOP_REWARD*num_hops
        self.reward += RIDE_REWARD*num_pass - (TRIP_REWARD * trip_time) - (WAIT_COST * wait_time)  - (HOP_REWARD*num_hops)
        self.trajectory = []
        self.status = 'SV'
        self.effective_d += eff_distance
        self.actual_d +=actual_distance 
        return True

    def route(self, path, trip_time):
        if not self.available:
            print ("The vehicle #%d is not available for service." % self.id)
            return False
        self.eta = trip_time
        self.trajectory = path
        self.status = 'MV'
        return True


    def get_state(self):
        if self.trajectory:
            lat, lon = self.trajectory[-1]
            dest_zone = Geohash.encode(lat, lon, precision=GEOHASH_PRECISION)
        else:
            dest_zone = self.zone
        lat, lon = self.location
        #######EDITED HERE########### added self.passengers at the end
        return (self.id, int(self.available), self.zone, dest_zone,
                self.eta, self.status, self.reward, lat, lon, self.idle,self.effective_d,self.actual_d,self.passengers)
        #######EDITED HERE###########

    def get_location(self):
        lat, lon = self.location
        return (self.id, lat, lon, int(self.available))


    def get_score(self):
        return (self.id, self.reward, self.total_service, self.total_idle)