def stationsInGeometry(geometry, stations):
intersect_stations = []
station_index = 0
intersect_stations.append(
["ID", "NAME", "LONG", "LAT", "ELEVATION", "STATIONID"])
geometry = shape(geometry[0]['geometry'])
print("Number of stations: " + str(len(stations["results"])))
for station in stations["results"]:
point = Point(station["longitude"], station["latitude"])
point.crs = {'init': '+proj=longlat +datum=WGS84 +no_defs'}
if geometry.contains(point):
station_index += 1
add_station = [
station_index, station["name"], station["longitude"],
station["latitude"], station["elevation"], station["id"]
]
intersect_stations.append(add_station)
print("Number of stations in geometry:" + str(station_index))
return intersect_stations
def do_KNN_and_naive_bayes(self):
new_p = self.new_p
# print(new_p)
# df= pd.read_excel('core/tweets_location.xls')
df = pd.read_csv('core/tweets_location.csv')
# new_p= [-81.33response020, 28.5380] #the point for which i need to know the zipcode
# KNN on lat long of the tweetzips.
n_neighbors = 5
samples = df.as_matrix(['longi', 'lat'])
neigh = NearestNeighbors(n_neighbors)
neigh.fit(samples)
NearestNeighbors(algorithm='auto', metric='haversine')
a, b = neigh.kneighbors([new_p])
candidates, tweetzip, candidate_loc, annotation, raw_tweets = self.most_frequent(
a, b, df)
# create new dataframe to match with the train file
new_point = pd.DataFrame(index=range(1))
new_point['predicted_k=%s' % str(n_neighbors)] = tweetzip
# make a shaply point
sp = Point(new_p[0], new_p[1])
sp.crs = {'init': 'epsg:4326'}
usps = gpd.GeoDataFrame.from_file('core/maps/usps_wgs84.shp')
esri = gpd.GeoDataFrame.from_file('core/maps/esri_wgs84.shp')
zcta = gpd.GeoDataFrame.from_file('core/maps/zcta_wgs84.shp')
map1 = gpd.GeoDataFrame.from_file('core/maps/map1_wgs84.shp')
usps.crs = {'init': 'epsg:4326'}
esri.crs = {'init': 'epsg:4326'}
zcta.crs = {'init': 'epsg:4326'}
map1.crs = {'init': 'epsg:4326'}
usps = usps.loc[usps.zipc > 30000]
esri = esri.loc[esri.ZIP_number > 30000]
zcta.ZCTA5CE10 = zcta.ZCTA5CE10.astype(int)
zcta = zcta.loc[zcta.ZCTA5CE10 > 30000]
map1.ZIP = map1.ZIP.astype(int)
map1 = map1.loc[map1.ZIP > 30000]
# intersect geodataframe of the point to the polygon geodataframe and get the zipcode
pnt = gpd.GeoDataFrame(geometry=[sp])
# print("here")
usps_int = gpd.sjoin(pnt,
usps[['zipc', 'geometry']],
how='left',
op='intersects')
new_point['tweet_usps'] = usps_int.zipc
esri_int = gpd.sjoin(pnt,
esri[['ZIP_number', 'geometry']],
how='left',
op='intersects')
zcta_int = gpd.sjoin(pnt,
zcta[['ZCTA5CE10', 'geometry']],
how='left',
op='intersects')
map1_int = gpd.sjoin(pnt,
map1[['ZIP', 'geometry']],
how='left',
op='intersects')
new_point['tweet_map'] = map1_int.ZIP
new_point['tweet_zcta'] = zcta_int.ZCTA5CE10
new_point['tweet_esri'] = esri_int.ZIP_number
new_point = new_point[[
'tweet_usps', 'tweet_map', 'tweet_zcta', 'tweet_esri',
'predicted_k=5'
]]
naive_train = pd.read_csv('core/k5_forNaive_allmap.csv')
X_train = naive_train[[
'tweet_map', 'tweet_zcta', 'tweet_esri', 'predicted_k=5'
]].values
y_train = naive_train[['tweet_usps']].values
X_test = new_point[[
'tweet_map', 'tweet_zcta', 'tweet_esri', 'predicted_k=5'
]].values
y_test = new_point[['tweet_usps']].values
# multinomial naive base prediction
clf = MultinomialNB().fit(X_train, y_train)
y_pred = clf.predict(new_point[[
'tweet_map', 'tweet_zcta', 'tweet_esri', 'predicted_k=5'
]])
zip_prob = pd.DataFrame(clf.predict_proba(X_test),
columns=clf.classes_)
# return int(y_pred)
# return zip_prob.to_json() ## to return all the zips with probabilities
# gaussian prediction
clf_gs = GaussianNB().fit(X_train, y_train.ravel())
y_pred_gs = clf_gs.predict(X_test)
zip_prob_gs = pd.DataFrame(clf_gs.predict_proba(X_test),
columns=clf_gs.classes_)
return {
"zip_with_probs": zip_prob_gs.to_json(),
"candidates": candidates,
"candidate_loc": candidate_loc,
"candidate_annotation": annotation,
"raw_tweets": raw_tweets
}
