昨天做了一些前處理，今天把昨天的處理好的資料做進一步的整合。

資料過濾

首先是宜蘭縣鐵道(polyline)

import geopandas as gpd
rail_yilan=gpd.read_file('output/Rail_yilan.shp',encoding='utf-8')
rail_yilan

匯入車站資料，並進一步篩選通話中的車站

station=gpd.read_file('output/station.shp',encoding='utf-8')
station_list=['貢寮','福隆','大里','大溪','龜山','頭城','宜蘭','二結','羅東','冬山','新馬']
station=station[station['Station_Na'].isin(station_list)]
station=station.reset_index(drop=True)
station

最後的是對話資料

import pandas as pd 
talk=pd.read_csv('output/talk.csv',encoding='utf-8')
talk['time']=[row['time'][:-2]+"00" for idx,row in talk.iterrows()]
talk

正規化與整合

接下來，我們把原始的路線資料切成通訊對話逐字稿的模式，這裡會用到graph的資訊計算，

• 把polyline換算成graph
• 分別把車站設為起終點，計算路徑
• 例如在graph輸入宜蘭最近的node及羅東站最近的node,計算的得到這兩個站的路線
  會採用這樣的做法是臨時決定的>.< 說不定改天有更好的做法
  細節今天就先跳過，過幾天會講到，反正最終結果是要把polyline變成像是

福隆-貢寮 或是 宜蘭-二結的這種模式的線

## 這邊先手動整理對照表(對照上表)
routes=[[9,10],
[3,4],
[3,3],
[2,3],
[6,7],
[6,6],
[0,2],
[8,9] ,   
[1,0],
[4,5]]  

from shapely.geometry import LineString,Point
from s2g import ShapeGraph
import networkx as nx
sg = ShapeGraph(shapefile='output/Rail_yilan.shp', to_graph=True)
graph = sg.to_networkx()

line_gs=[]
line_name=[]
for route in routes:

    min_distance=999
    start_id=0
    for num in range(sg.nodes_counter):
        xy=sg.node_xy[num]
        if Point(xy).distance(station.at[route[0],'geometry'])<min_distance:
            min_distance=Point(xy).distance(station.at[route[0],'geometry'])
            start_id=num


    min_distance=999
    end_id=0
    for num in range(sg.nodes_counter):
        xy=sg.node_xy[num]
        if Point(xy).distance(station.at[route[1],'geometry'])<min_distance:
            min_distance=Point(xy).distance(station.at[route[1],'geometry'])
            end_id=num
            nearest_path= nx.shortest_path(graph, source=start_id, target=end_id)
    geoms=[]
    for item in nearest_path:
            geoms.append(sg.node_xy[item])
    try:
        line_gs.append(LineString(geoms))
        line_name.append("（地點："+station.at[route[0],'Station_Na']+"-"+station.at[route[1],'Station_Na']+")")
    except:pass
    

train_lines = gpd.GeoDataFrame(crs= {'init' :'epsg:4326'},geometry=line_gs)
train_lines['location']=line_name
train_lines

以下是我們要的路線整合結果：

最後整合對話與路線，

# 初始化
talk['geometry']=talk['location']
for i1,r1 in talk.iterrows():
    for i2,r2 in  train_lines.iterrows():
        if r1['location']==r2['location']:
            talk.at[i1,'geometry']=r2['geometry']
            break
        else:
            talk.at[i1,'geometry']=talk.at[0,'geometry']

            
            
st=talk.at[0,'time']
temp_text=""         
geoms=[]
temp_location=talk.at[0,'location']
for i1,r1 in talk.iterrows():
    if temp_location!=talk.at[i1,'location']:
        geoms.append([talk.at[i1-1,'geometry'],[st,talk.at[i1-1,'time']],talk.at[i1-1,'location'],temp_text])
       
        ##release
        temp_location=talk.at[i1,'location']
        temp_text="" 
        st=talk.at[i1,'time']
    else:
        temp_text+=talk.at[i1,'content']+";"
geoms.append([talk.at[i1,'geometry'],[st,talk.at[i1,'time']],talk.at[i1,'location'],temp_text])      
train_lines_talk= gpd.GeoDataFrame(geoms)
train_lines_talk.columns=['geometry','time','location','text']
train_lines_talk

最終成果：