Trajectory Full Scanning¶

import pyscan
import matplotlib.pyplot as plt
import random
import math
import statistics
import csv
import itertools
import numpy as np

def plot_points(ax, pts, c):
    xs = []
    ys = []
    for pt in pts:
        xs.append(pt[0] )
        ys.append(pt[1])
    ax.scatter(xs, ys, color=c)

def plot_points_traj(ax, pts, c):
    xs = []
    ys = []
    for pt in pts:
        xs.append(pt[0])
        ys.append(pt[1])
    ax.plot(xs, ys, color=c)

def plot_approx(ax, traj_pts, core_set_pts):
    ax.set_xlim([-.01, 1.01])
    ax.set_ylim([-.01, 1.01])
    plot_points_traj(ax, traj_pts, "g")
    plot_points(ax, core_set_pts, "b")
    ax.set_axis_off()

def plot_full_trajectories(red, blue, ax):
    plot_set = [(reg, True) for reg in blue] + [(reg, False) for reg in red]
    random.shuffle(plot_set)

    for traj, is_blue in plot_set:
        if is_blue:
            plot_points_traj(ax, traj, "b")
        else:
            plot_points_traj(ax, traj, "r")


def remove_long_trajectories(trajectories, percent=.9):
    def toTraj(pts):
        return pyscan.Trajectory([pyscan.Point(p[0], p[1], 1.0) for p in pts])

    ltraj = sorted(toTraj(traj).get_length() for traj in trajectories)
    perc_len_traj = ltraj[int(percent * len(trajectories))]
    del ltraj
    return [traj for traj in trajectories if toTraj(traj).get_length() <= perc_len_traj]


def normalize(pt, mxx, mnx, mxy, mny):
    return (pt[0] - mnx) / (mxx - mnx), (pt[1] - mny) / (mxy - mny)


def read_csv(fname, filter_long=True):
    traj_id_set = []
    with open(fname) as f:
        curr_id = None
        mnx = float("inf")
        mxx = -float("inf")
        mny = float("inf")
        mxy = -float("inf")

        for row in csv.reader(f, delimiter=" "):
            try:
                if row[0] != curr_id:
                    curr_id = row[0]
                    traj_id_set.append([])
                if math.isnan(float(row[1])) or math.isinf(float(row[1])):
                    continue
                if math.isnan(float(row[2])) or math.isinf(float(row[2])):
                    continue

                x, y = (float(row[1]), float(row[2]))
                mnx = min(mnx, x)
                mny = min(mny, y)
                mxx = max(mxx, x)
                mxy = max(mxy, y)
                traj_id_set[-1].append((float(row[1]), float(row[2])))

            except ValueError:
                continue
        norm_traces = []
        while traj_id_set:
            norm_trace = []
            trace = traj_id_set.pop()
            for pt in trace:
                norm_trace.append(normalize(pt, mxx, mnx, mxy, mny))
            norm_traces.append(norm_trace)
        if filter_long:
            return remove_long_trajectories(norm_traces)
        else:
            return norm_traces

# Ok we finally have the trajectories scaled to be in a [0,1]x[0,1] box with the extremely
#long trajectories removed
trajectories = read_csv("./bjtaxi_samples_10k.tsv")

r = .1
p = .5
q = .2
alpha = 1 / 100
max_r = 1 / 50
n = 100
s = 4000
min_r = alpha