python數據可視化pygal模擬擲骰子實現示例

可視化包Pygal生成可縮放矢量圖形文件

可以在尺寸不同的屏幕上自動縮放,顯示圖表

#安裝pygal
pip install pygal
'''
想要瞭解Pygal可生成什麼樣的圖表,可訪問http://www.pygal.org/
單擊document,點擊chart types,每個示例都包含源代碼
'''
from random import randint
#創建一個骰子的類
class Die():
    def __init__(self,num_sides = 6):
        self.num_sides = num_sides
    def roll(self):
        #返回一個位於1和骰子面數之間的隨機值
        return randint(1, self.num_sides)
#擲骰子
die = Die()
#創建一個列表,將結果存儲在一個列表中
results = []
 #投100次   
for roll_num in range(100):
    result = die.roll()
    results.append(result)
print(results)
[3, 4, 2, 2, 6, 5, 5, 5, 2, 3, 1, 4, 3, 2, 1, 2, 3, 6, 6, 5, 5, 3, 2, 3, 1, 1, 4, 1, 4, 6, 1, 6, 2, 3, 4, 6, 2, 5, 5, 1, 6, 1, 5, 4, 3, 3, 4, 5, 6, 3, 5, 1, 4, 3, 5, 6, 6, 6, 4, 6, 5, 6, 5, 4, 6, 3, 1, 4, 1, 4, 2, 1, 1, 4, 4, 4, 2, 3, 1, 4, 6, 2, 1, 5, 6, 2, 2, 6, 6, 3, 6, 2, 6, 6, 4, 4, 2, 1, 1, 6]

分析結果,計算每個點數出現的次數

frequencies = []
for value in range(1, die.num_sides+1):
    frequency = results.count(value)
    frequencies.append(frequency)
print(frequencies)
[10, 23, 13, 9, 26, 19]

繪制直方圖

import pygal
hist = pygal.Bar()
hist.title = 'results of rolling one d6 100 times'
hist.x_lables = ['1', '2', '3', '4', '5', '6']
hist.x_title = 'result'
hist.y_title = 'frequency of result'
hist.add('d6', frequencies)
#將圖渲染為SVG文件,需要打開瀏覽器,才能查看生成的直方圖
hist.render_to_file('die_visual.svg')

同時投擲兩個骰子

from random import randint
#創建一個骰子的類
class Die():
    def __init__(self,num_sides = 6):
        self.num_sides = num_sides
    def roll(self):
        #返回一個位於1和骰子面數之間的隨機值
        return randint(1, self.num_sides)
#擲骰子
die1 = Die()
die2 = Die()
#創建一個列表,將結果存儲在一個列表中
results = []
 #投100次   
for roll_num in range(100):
    result = die1.roll() + die2.roll()
    results.append(result)
print(results)
#分析結果,計算每個點數出現的次數
frequencies = []
max_result = die1.num_sides + die2.num_sides
for value in range(1, max_result+1):
    frequency = results.count(value)
    frequencies.append(frequency)
print(frequencies)
#繪制直方圖
import pygal
hist = pygal.Bar()
hist.title = 'results of rolling one d6 dice 100 times'
hist.x_lables = ['2', '3', '4', '5', '6', '7', '8', '9', '10', '11', '12']
hist.x_title = 'result'
hist.y_title = 'frequency of result'
hist.add('d6 + d6', frequencies)
#將圖渲染為SVG文件,需要打開瀏覽器,才能查看生成的直方圖
hist.render_to_file('die_visual.svg')
[4, 7, 4, 5, 8, 4, 3, 6, 8, 9, 8, 11, 9, 11, 8, 8, 5, 6, 10, 5, 11, 7, 4, 3, 12, 12, 7, 2, 4, 9, 9, 5, 7, 10, 4, 7, 4, 6, 5, 6, 7, 2, 7, 9, 7, 6, 11, 5, 9, 6, 11, 4, 8, 10, 7, 9, 5, 4, 3, 7, 4, 10, 5, 7, 2, 6, 4, 2, 2, 5, 5, 9, 6, 3, 6, 10, 12, 7, 4, 11, 8, 6, 10, 5, 7, 5, 5, 7, 9, 4, 11, 6, 7, 8, 6, 11, 6, 4, 3, 12]
[0, 5, 5, 14, 13, 13, 15, 8, 9, 6, 8, 4]

同時投擲兩個面數不同骰子

from random import randint
#創建一個骰子的類
class Die():
    def __init__(self,num_sides = 6):
        self.num_sides = num_sides
    def roll(self):
        #返回一個位於1和骰子面數之間的隨機值
        return randint(1, self.num_sides)
#擲骰子
die1 = Die()
die2 = Die(10)
#創建一個列表,將結果存儲在一個列表中
results = []
 #投100次   
for roll_num in range(100):
    result = die1.roll() + die2.roll()
    results.append(result)
print(results)
#分析結果,計算每個點數出現的次數
frequencies = []
max_result = die1.num_sides + die2.num_sides
for value in range(1, max_result+1):
    frequency = results.count(value)
    frequencies.append(frequency)
print(frequencies)
#繪制直方圖
import pygal
hist = pygal.Bar()
hist.title = 'results of rolling one d10 dice 100 times'
hist.x_lables = ['2', '3', '4', '5', '6', '7', '8', '9', '10', '11', '12', '13', '14','15','16']
hist.x_title = 'result'
hist.y_title = 'frequency of result'
hist.add('d6 + d10', frequencies)
#將圖渲染為SVG文件,需要打開瀏覽器,才能查看生成的直方圖
hist.render_to_file('die_visual.svg')
[5, 3, 6, 13, 8, 9, 10, 11, 11, 4, 5, 14, 11, 10, 11, 8, 14, 12, 16, 8, 9, 11, 7, 11, 9, 2, 8, 9, 9, 10, 7, 8, 12, 11, 8, 12, 9, 9, 10, 11, 8, 14, 10, 12, 10, 7, 12, 5, 4, 8, 6, 7, 7, 11, 9, 16, 6, 13, 6, 10, 6, 7, 16, 9, 14, 5, 7, 12, 8, 9, 11, 11, 6, 11, 5, 8, 11, 16, 4, 10, 5, 10, 13, 4, 9, 9, 11, 9, 11, 13, 7, 13, 13, 5, 5, 4, 5, 3, 12, 14]
[0, 1, 2, 5, 9, 6, 8, 10, 13, 9, 15, 7, 6, 5, 0, 4]

以上就是python數據可視化pygal模擬擲骰子實現示例的詳細內容,更多關於python pygal模擬擲骰子的資料請關註WalkonNet其它相關文章!

推薦閱讀: