地理データ分析を使用して、市内のさまざまな場所での緊急通報の数を予測するにはどうすればよいですか

import pandas df_zones = pandas.read_csv('data/zones.csv', index_col='zone_id') df_zones.head() 

 import pickle import osmread from tqdm import tqdm_notebook LAT_MIN, LAT_MAX = 55.309397, 56.13526 LON_MIN, LON_MAX = 36.770379, 38.19270 osm_file = osmread.parse_file('osm/RU-MOS.osm.pbf') tagged_nodes = [ entry for entry in tqdm_notebook(osm_file, total=18976998) if isinstance(entry, osmread.Node) if len(entry.tags) > 0 if (LAT_MIN < entry.lat < LAT_MAX) and (LON_MIN < entry.lon < LON_MAX) ] #         with open('osm/tagged_nodes.pickle', 'wb') as fout: pickle.dump(tagged_nodes, fout, protocol=pickle.HIGHEST_PROTOCOL) #         with open('osm/tagged_nodes.pickle', 'rb') as fin: tagged_nodes = pickle.load(fin) 

 import folium fmap = folium.Map([55.753722, 37.620657]) #  /  for node in tagged_nodes: if node.tags.get('railway') == 'station': folium.CircleMarker([node.lat, node.lon], radius=3).add_to(fmap) #       calls_thresh = df_zones.calls_daily.quantile(.99) for _, row in df_zones.query('calls_daily > @calls_thresh').iterrows(): folium.features.RectangleMarker( bounds=((row.lat_bl, row.lon_bl), (row.lat_tr, row.lon_tr)), fill_color='red', ).add_to(fmap) #        fmap.save('map_demo.html') 

 import collections import math import numpy as np from sklearn.neighbors import NearestNeighbors kremlin_lat, kremlin_lon = 55.753722, 37.620657 def dist_calc(lat1, lon1, lat2, lon2): R = 6373.0 lat1 = math.radians(lat1) lon1 = math.radians(lon1) lat2 = math.radians(lat2) lon2 = math.radians(lon2) dlon = lon2 - lon1 dlat = lat2 - lat1 a = math.sin(dlat / 2)**2 + math.cos(lat1) * math.cos(lat2) * \ math.sin(dlon / 2)**2 c = 2 * math.atan2(math.sqrt(a), math.sqrt(1 - a)) return R * c df_features = collections.OrderedDict([]) df_features['distance_to_kremlin'] = df_zones.apply( lambda row: dist_calc(row.lat_c, row.lon_c, kremlin_lat, kremlin_lon), axis=1) #   ,      POINT_FEATURE_FILTERS = [ ('tagged', lambda node: len(node.tags) > 0), ('railway', lambda node: node.tags.get('railway') == 'station'), ('shop', lambda node: 'shop' in node.tags), ('public_transport', lambda node: 'public_transport' in node.tags), ] #      X_zone_centers = df_zones[['lat_c', 'lon_c']].as_matrix() for prefix, point_filter in POINT_FEATURE_FILTERS: #        coords = np.array([ [node.lat, node.lon] for node in tagged_nodes if point_filter(node) ]) #        neighbors = NearestNeighbors().fit(coords) #   "    R   " for radius in [0.001, 0.003, 0.005, 0.007, 0.01]: dists, inds = neighbors.radius_neighbors(X=X_zone_centers, radius=radius) df_features['{}_points_in_{}'.format(prefix, radius)] = \ np.array([len(x) for x in inds]) #   "   K " for n_neighbors in [3, 5, 10]: dists, inds = neighbors.kneighbors(X=X_zone_centers, n_neighbors=n_neighbors) df_features['{}_mean_dist_k_{}'.format(prefix, n_neighbors)] = \ dists.mean(axis=1) df_features['{}_max_dist_k_{}'.format(prefix, n_neighbors)] = \ dists.max(axis=1) df_features['{}_std_dist_k_{}'.format(prefix, n_neighbors)] = \ dists.std(axis=1) #   "   " df_features['{}_min'.format(prefix)] = dists.min(axis=1) #       features.csv df_features = pandas.DataFrame(df_features, index=df_zones.index) df_features.to_csv('data/features.csv') df_features.head() 

 from sklearn.model_selection import train_test_split from sklearn.ensemble import RandomForestRegressor df_zones_train = df_zones.query('is_test == 0 & is_target == 1') idx_train, idx_valid = train_test_split(df_zones_train.index, test_size=0.3) X_train = df_features.loc[idx_train, :] y_train = df_zones.loc[idx_train, 'calls_daily'] model = RandomForestRegressor(n_estimators=100, n_jobs=4) model.fit(X_train, y_train) 

 from scipy.stats import kendalltau X_valid = df_features.loc[idx_valid, :] y_valid = df_zones.loc[idx_valid, 'calls_daily'] y_pred = model.predict(X_valid) target_columns = ['calls_wd{}'.format(d) for d in range(7)] df_valid_target = df_zones.loc[idx_valid, target_columns] df_valid_predictions = pandas.DataFrame(collections.OrderedDict([ (column_name, y_pred) for column_name in target_columns ]), index=idx_valid) df_comparison = pandas.DataFrame({ 'target': df_valid_target.unstack(), 'prediction': df_valid_predictions.unstack(), }) valid_score = kendalltau(df_comparison['target'], df_comparison['prediction']).correlation print('Validation score:', valid_score) 

 idx_test = df_zones.query('is_test == 1').index X_test = df_features.loc[idx_test, :] y_pred = model.predict(X_test) df_test_predictions = pandas.DataFrame(collections.OrderedDict([ (column_name, y_pred) for column_name in target_columns ]), index=idx_test) df_test_predictions.to_csv('data/sample_submission.csv') df_test_predictions.head()