stimmenfryslan/notebooks/Pronunciation map.ipynb

Geographical pronunciation statistics¶

In [ ]:

import pandas
import MySQLdb
import numpy
import json

db = MySQLdb.connect(user='root', passwd='Nmmxhjgt1@', db='stimmen', charset='utf8')

%matplotlib notebook
from matplotlib import pyplot
import folium
from IPython.display import display
from shapely.geometry import Polygon, MultiPolygon, shape, Point
from jupyter_progressbar import ProgressBar
from collections import defaultdict
from ipy_table import make_table
from html import escape

import numpy as np
import matplotlib.pyplot as plt
from matplotlib.colors import LogNorm
from sklearn import mixture
from skimage.measure import find_contours
from collections import Counter
from random import shuffle

In [ ]:

# Borders of Frysian municipalities

with open('Friesland_AL8.GeoJson') as f:
    gemeentes = json.load(f)

In [ ]:

shapes = [shape(feature['geometry']) for feature in gemeentes['features']]
gemeente_names = [feature['properties']['name'] for feature in gemeentes['features']]

def get_gemeente(point):
    for i, shape in enumerate(shapes):
        if shape.contains(point):
            return i
    return -1

In [ ]:

# Answers to how participants state a word should be pronounces.

answers = pandas.read_sql('''
SELECT prediction_quiz_id, user_lat, user_lng, question_text, answer_text
FROM       core_surveyresult as survey
INNER JOIN core_predictionquizresult as result ON survey.id = result.survey_result_id
INNER JOIN core_predictionquizresultquestionanswer as answer
    ON result.id = answer.prediction_quiz_id
''', db)

In [ ]:

# Takes approximately 2 minutes

gemeente_map = {
    (lng, lat): get_gemeente(Point(lng, lat))
    for lng, lat in set(zip(answers['user_lng'], answers['user_lat']))
}

answers['gemeente'] = [
    gemeente_map[(lng, lat)]
    for lat, lng in zip(answers['user_lat'], answers['user_lng'])
]

Mapping pronunciations¶

The idea is to plot each pronunciation as a point of a different color, now only seems to show participation density.

Slow, so started with the first question.

In [ ]:

# cmap = pyplot.get_cmap('gist_rainbow')

# std = (1.89, 1.35)

# for _, (question, rows) in zip(range(3), answers.groupby('question_text')):
#     plt.figure()
#     n_answers = len(rows.groupby('answer_text').count())
#     colors = cmap(range(256))[::256 // n_answers]
#     for (answer, rows_), color in zip(rows.groupby('answer_text'), colors):
#         if len(rows_) < 100:
#             continue
#         color = '#%02x%02x%02x' % tuple(int(c*255) for c in color[:3])
#         X = rows_[['user_lat', 'user_lng']].as_matrix()

#         clf = mixture.GaussianMixture(n_components=5, covariance_type='full')
#         clf.fit(X)
#         xlim = numpy.percentile(X[:, 0], [1, 99.5])
#         ylim = numpy.percentile(X[:, 1], [1, 99.5])
#         xlim = [2*xlim[0] - xlim[1], 2*xlim[1] - xlim[0]]
#         ylim = [2*ylim[0] - ylim[1], 2*ylim[1] - ylim[0]]
        
#         x = np.linspace(*xlim, 1000)
#         y = np.linspace(*ylim, 1000)
#         xx, yy = np.meshgrid(x, y)
#         xxyy = np.array([xx.ravel(), yy.ravel()]).T
#         z = np.exp(clf.score_samples(xxyy))
#         z = z.reshape(xx.shape)
        
#         z_sorted = sorted(z.ravel(), reverse=True)
#         z_sorted_cumsum = np.cumsum(z_sorted)
#         split = np.where(z_sorted_cumsum > (z_sorted_cumsum[-1] * 0.5))[0][0]
#         threshold = z_sorted[split]
#         threshold

# #         p = list(range(0, 100, 5))

#         p = [80]
#         plt.contour(xx, yy, z, levels=[threshold], colors=[color])
#         plt.plot(X[:, 0], X[:, 1], '.', c=color)
#         plt.xlim(*xlim)
#         plt.ylim(*ylim)

In [ ]:

zero_latlng_questions = {
    q
    for q, row in answers.groupby('question_text').agg('std').iterrows()
    if row['user_lat'] == 0 and row['user_lng'] == 0
}
answers_filtered = answers[answers['question_text'].map(lambda x: x not in zero_latlng_questions)]

In [ ]:

answers_filtered['question_text_url'] = answers_filtered['question_text'].map(
    lambda x: x.replace('"', '').replace('*', ''))

In [ ]:

def get_palette(n, no_black=True, no_white=True):
    with open('glasbey/{}_colors.txt'.format(n + no_black + no_white)) as f:
        return [
            '#%02x%02x%02x' % tuple(int(c) for c in line.replace('\n', '').split(','))
            for line in f
            if not no_black or line != '0,0,0\n'
            if not no_white or line != '255,255,255\n'
        ]

In [ ]:

options = [x[1] for x in sorted([
    (row['user_lng'], answer_text)
    for answer_text, row in rows.groupby('answer_text').agg({'user_lng': 'count'}).iterrows()
], reverse=True)]

groups = [options[:len(options) // 2], options[len(options) // 2:]]
groups

In [ ]:

80000 / 350

In [ ]:

import glob
with open('index.html', 'w') as f:
    f.write('<html><head></head><body>' + 
        '<br/>\n'.join(
            '\t<a href="http://herbertkruitbosch.com/pronunciation_maps/{}">{}<a>'.format(fn, fn[:-4].replace('_', ' '))
            for fn in sorted(
                glob.glob('*_all.html') +
                glob.glob('*_larger.html') +
                glob.glob('*_smaller.html')
            )
    ) + "</body></html>")

In [ ]:

# cmap = pyplot.get_cmap('gist_rainbow')
# colors = pyplot.get_cmap('tab20')
# colors = ['#e6194b', '#3cb44b', '#ffe119', '#0082c8', '#f58231', '#911eb4', '#46f0f0', '#f032e6', '#d2f53c', '#fabebe', '#008080', '#e6beff', '#aa6e28', '#fffac8', '#800000', '#aaffc3', '#808000', '#ffd8b1', '#000080', '#808080']

std = (1.89, 1.35)

for question, rows in answers_filtered.groupby('question_text_url'):
#     question = rows['question_text_url'][0]
    n_answers = len(rows.groupby('answer_text').count())
    
    
    options = [x[1] for x in sorted([
        (row['user_lng'], answer_text)
        for answer_text, row in rows.groupby('answer_text').agg({'user_lng': 'count'}).iterrows()
    ], reverse=True)]
    groups = [options]
    if n_answers > 6:
        groups.extend([options[:6], options[6:]])
    
    for group, group_name in zip(groups, ['all', 'larger', 'smaller']):
        m = folium.Map((rows['user_lat'].median(), rows['user_lng'].median()), tiles='stamentoner', zoom_start=9)
    #     colors = cmap(range(256))[::256 // n_answers]
        colors = get_palette(len(group))
        for answer, color in zip(group, colors):
            rows_ = rows[rows['answer_text'] == answer]
    #         color = '#%02x%02x%02x' % tuple(int(c*255) for c in color[:3])
            name = '<span style=\\"color:{}; \\">{} ({})'.format(color, escape(answer), len(rows_))

            group = folium.FeatureGroup(name=name)
            colormap[name] = color

            for point in zip(rows_['user_lat'], rows_['user_lng']):
                point = tuple(p + 0.01 * s * numpy.random.randn() for p, s in zip(point, std))
                folium.Circle(
                    point, color=None, fill_color=color,
                    radius=400*min(1, 100 / len(rows_)), fill_opacity=1 #1 - 0.5 * len(rows_) / len(rows)
                ).add_to(group)
            group.add_to(m)
        folium.map.LayerControl('topright', collapsed=False).add_to(m)
    
        print(group_name, question)
        if group_name == 'larger':
            display(m)
        m.save('{}_{}.html'.format(question, group_name))