-
Notifications
You must be signed in to change notification settings - Fork 5
/
gf_mean.py
145 lines (115 loc) · 5.93 KB
/
gf_mean.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
# -*- coding: utf-8 -*-
# encoding = utf-8
'''
gf_mean.py
author：alvin
create dayno: 20210719
Function: Gaussian Filter + Phone Mean.
功能: 高斯滤波 + 平均路径。
History:
version contributor comment
v1.0 alvin 第一版
Reference:
1. 'Adaptive_gauss+phone_mean'(Petr B): https://www.kaggle.com/bpetrb/adaptive-gauss-phone-mean
'''
import numpy as np
import pandas as pd
from scipy.ndimage import gaussian_filter1d
from scipy.interpolate import interp1d
import optuna
import os
from pathlib import Path
def apply_gauss_smoothing(df, params):
'''Apply Gaussian Filter to smooth the data.'''
df = df.copy()
SZ_1 = params['sz_1']
SZ_2 = params['sz_2']
SZ_CRIT = params['sz_crit']
unique_paths = df[['collectionName', 'phoneName']].drop_duplicates().to_numpy()
for collection, phone in unique_paths:
cond = np.logical_and(df['collectionName'] == collection, df['phoneName'] == phone)
data = df[cond][['latDeg', 'lngDeg']].to_numpy()
lat_g1 = gaussian_filter1d(data[:, 0], np.sqrt(SZ_1))
lon_g1 = gaussian_filter1d(data[:, 1], np.sqrt(SZ_1))
lat_g2 = gaussian_filter1d(data[:, 0], np.sqrt(SZ_2))
lon_g2 = gaussian_filter1d(data[:, 1], np.sqrt(SZ_2))
lat_dif = data[1:,0] - data[:-1,0]
lon_dif = data[1:,1] - data[:-1,1]
lat_crit = np.append(np.abs(gaussian_filter1d(lat_dif, np.sqrt(SZ_CRIT)) / (1e-9 + gaussian_filter1d(np.abs(lat_dif), np.sqrt(SZ_CRIT)))),[0])
lon_crit = np.append(np.abs(gaussian_filter1d(lon_dif, np.sqrt(SZ_CRIT)) / (1e-9 + gaussian_filter1d(np.abs(lon_dif), np.sqrt(SZ_CRIT)))),[0])
df.loc[cond, 'latDeg'] = lat_g1 * lat_crit + lat_g2 * (1.0 - lat_crit)
df.loc[cond, 'lngDeg'] = lon_g1 * lon_crit + lon_g2 * (1.0 - lon_crit)
return df
def mean_with_other_phones(df):
df = df.copy()
collections_list = df[['collectionName']].drop_duplicates().to_numpy()
# Target for each colleciton. 针对每个collection
for collection in collections_list:
phone_list = df[df['collectionName'].to_list() == collection][['phoneName']].drop_duplicates().to_numpy()
phone_data = {}
corrections = {}
# Target for each phone. 针对每个phone
for phone in phone_list:
# Get the boolean of the none value. collection+phone的bool位置
cond = np.logical_and(df['collectionName'] == collection[0], df['phoneName'] == phone[0]).to_list()
phone_data[phone[0]] = df[cond][['millisSinceGpsEpoch', 'latDeg', 'lngDeg']].to_numpy()
# Choose a phone. 选择一个phone的数据
for current in phone_data:
correction = np.ones(phone_data[current].shape, dtype=np.float)
correction[:,1:] = phone_data[current][:,1:] # Load location info. 只载入经纬度，时间全变为1
# Telephones data don't complitely match by time, so - interpolate.
for other in phone_data:
if other == current:
continue
# Use other phone to interpolate. 用其它phone做插值
# x: timestamp; y: location
# x为时间phone_data[other][:,0]，y为经纬度phone_data[other][:,1:]
loc = interp1d(phone_data[other][:,0],
phone_data[other][:,1:],
axis=0,
kind='linear',
copy=False,
bounds_error=None,
fill_value='extrapolate',
assume_sorted=True)
# In the same collection, find out which points are the start point and stop point
# 找到同一个collection，哪个点最早和最晚
start_idx = 0
stop_idx = 0
for idx, val in enumerate(phone_data[current][:,0]):
if val < phone_data[other][0,0]:
start_idx = idx
if val < phone_data[other][-1,0]:
stop_idx = idx
if stop_idx - start_idx > 0:
correction[start_idx:stop_idx,0] += 1
correction[start_idx:stop_idx,1:] += loc(phone_data[current][start_idx:stop_idx,0])
# Mean the trajectorie of other phones. 现有机子和其它机子做平均
correction[:,1] /= correction[:,0]
correction[:,2] /= correction[:,0]
corrections[current] = correction.copy()
for phone in phone_list:
cond = np.logical_and(df['collectionName'] == collection[0], df['phoneName'] == phone[0]).to_list()
df.loc[cond, ['latDeg', 'lngDeg']] = corrections[phone[0]][:,1:]
return df
def calc_haversine(lat1, lon1, lat2, lon2):
RADIUS = 6_367_000
lat1, lon1, lat2, lon2 = map(np.radians, [lat1, lon1, lat2, lon2])
dlat = lat2 - lat1
dlon = lon2 - lon1
a = np.sin(dlat/2)**2 + \
np.cos(lat1) * np.cos(lat2) * np.sin(dlon/2)**2
dist = 2 * RADIUS * np.arcsin(a**0.5)
return dist
def compute_dist(pred_df, gt_df):
oof = pred_df.copy()
gt = gt_df.copy()
df = oof.merge(gt, on = ['phone','millisSinceGpsEpoch'])
dst_oof = calc_haversine(df.latDeg_x,df.lngDeg_x, df.latDeg_y, df.lngDeg_y)
scores = pd.DataFrame({'phone': df.phone,'dst': dst_oof})
scores_grp = scores.groupby('phone')
d50 = scores_grp.quantile(.50).reset_index()
d50.columns = ['phone','q50']
d95 = scores_grp.quantile(.95).reset_index()
d95.columns = ['phone','q95']
return (scores_grp.quantile(.50).mean() + scores_grp.quantile(.95).mean())/2, d50.merge(d95)