zip
实例汇总.zip
温度转换
TempStr = input("请输入带有符号的温度值: ")
if TempStr[-1] in ["F", "f"]:
C = (eval(TempStr[0:-1]) - 32) / 1.8
print("转换后的温度是{:.2f}C".format(C))
elif TempStr[-1] in ["C", "c"]:
F = 1.8 * eval(TempStr[0:-1]) + 32
print("转换后的温度是{:.2f}F".format(F))
else:
print("输入格式错误")
"""
请输入带有符号的温度值: 39C
转换后的温度是102.20F
"""
python 蟒蛇绘制
import turtle
turtle.setup(650, 350, 200, 200)
turtle.penup()
turtle.fd(-250)
turtle.pendown()
turtle.pensize(25)
turtle.pencolor("purple")
turtle.seth(-40)
for i in range(4):
turtle.circle(40, 80)
turtle.circle(-40, 80)
turtle.circle(40, 80 / 2)
turtle.fd(40)
turtle.circle(16, 180)
turtle.fd(40 * 2 / 3)
turtle.done()
天天向上的力量
dayup = pow(1.001, 365)
daydown = pow(0.999, 365)
print("向上:{:.2f},向下:{:.2f}".format(dayup, daydown))
# 向上:1.44,向下:0.69
dayfactor = 0.019
dayup = pow(1 + dayfactor, 365)
daydown = pow(1 - dayfactor, 365)
print("向上:{:.2f},向下:{:.2f}".format(dayup, daydown))
# 向上:962.89,向下:0.00
dayup = 1.0
dayfactor = 0.01
for i in range(365):
if i % 7 in [6, 0]:
dayup = dayup * (1 - dayfactor)
else:
dayup = dayup * (1 + dayfactor)
print("工作日的力量:{:.2f} ".format(dayup))
# 工作日的力量:4.63
def dayUP(df):
dayup = 1
for i in range(365):
if i % 7 in [6, 0]:
dayup = dayup * (1 - 0.01)
else:
dayup = dayup * (1 + df)
return dayup
dayfactor = 0.01
while dayUP(dayfactor) < 37.78:
dayfactor += 0.001
print("工作日的努力参数是:{:.3f} ".format(dayfactor))
# 工作日的努力参数是:0.019
文本进度条
import time
scale = 10
print("------执行开始------")
for i in range(scale + 1):
a = "*" * i
b = "." * (scale - i)
c = (i / scale) * 100
print("{:^3.0f}%[{}->{}]".format(c, a, b))
time.sleep(0.1)
print("------执行结束------")
"""
------执行开始------
0 %[->..........]
10 %[*->.........]
20 %[**->........]
30 %[***->.......]
40 %[****->......]
50 %[*****->.....]
60 %[******->....]
70 %[*******->...]
80 %[********->..]
90 %[*********->.]
100%[**********->]
------执行结束------
"""
import time
for i in range(101):
print("\r{:3}%".format(i), end="")
time.sleep(0.1)
# 100%
import time
scale = 50
print("执行开始".center(scale // 2, "-"))
start = time.perf_counter()
for i in range(scale + 1):
a = "*" * i
b = "." * (scale - i)
c = (i / scale) * 100
dur = time.perf_counter() - start
print("\r{:^3.0f}%[{}->{}]{:.2f}s".format(c, a, b, dur), end="")
time.sleep(0.1)
print("\n" + "执行结束".center(scale // 2, "-"))
"""
-----------执行开始----------
100%[**************************************************->]5.22s
-----------执行结束----------
"""
身体质量指数 BMI
height, weight = eval(input("请输入身高(米)和体重(公斤)[逗号隔开]: "))
bmi = weight / pow(height, 2)
print("BMI 数值为:{:.2f}".format(bmi))
who = ""
if bmi < 18.5:
who = "偏瘦"
elif 18.5 <= bmi < 25:
who = "正常"
elif 25 <= bmi < 30:
who = "偏胖"
else:
who = "肥胖"
print("BMI 指标为:国际'{0}'".format(who))
"""
请输入身高(米)和体重(公斤)[逗号隔开]: 1.66,66
BMI 数值为:23.95
BMI 指标为:国际'正常'
"""
height, weight = eval(input("请输入身高(米)和体重(公斤)[逗号隔开]: "))
bmi = weight / pow(height, 2)
print("BMI 数值为:{:.2f}".format(bmi))
nat = ""
if bmi < 18.5:
nat = "偏瘦"
elif 18.5 <= bmi < 24:
nat = "正常"
elif 24 <= bmi < 28:
nat = "偏胖"
else:
nat = "肥胖"
print("BMI 指标为:国内'{0}'".format(nat))
"""
请输入身高(米)和体重(公斤)[逗号隔开]: 1.66,66
BMI 数值为:23.95
BMI 指标为:国内'正常'
"""
height, weight = eval(input("请输入身高(米)和体重(公斤)[逗号隔开]: "))
bmi = weight / pow(height, 2)
print("BMI 数值为:{:.2f}".format(bmi))
who, nat = "", ""
if bmi < 18.5:
who, nat = "偏瘦", "偏瘦"
elif 18.5 <= bmi < 24:
who, nat = "正常", "正常"
elif 24 <= bmi < 25:
who, nat = "正常", "偏胖"
elif 25 <= bmi < 28:
who, nat = "偏胖", "偏胖"
elif 28 <= bmi < 30:
who, nat = "偏胖", "肥胖"
else:
who, nat = "肥胖", "肥胖"
print("BMI 指标为:国际'{0}', 国内'{1}'".format(who, nat))
"""
请输入身高(米)和体重(公斤)[逗号隔开]: 1.66,66
BMI 数值为:23.95
BMI 指标为:国际'正常', 国内'正常'
"""
圆周率的计算
pi = 0
N = 100
for k in range(N):
pi += (
1
/ pow(16, k)
* (4 / (8 * k + 1) - 2 / (8 * k + 4) - 1 / (8 * k + 5) - 1 / (8 * k + 6))
)
print("圆周率值是: {}".format(pi))
# 圆周率值是: 3.141592653589793
from random import random
from time import perf_counter
DARTS = 1000 * 1000
hits = 0.0
start = perf_counter()
for i in range(1, DARTS + 1):
x, y = random(), random()
dist = pow(x**2 + y**2, 0.5)
if dist <= 1.0:
hits = hits + 1
pi = 4 * (hits / DARTS)
print("圆周率值是: {}".format(pi))
print("运行时间是: {:.5f}s".format(perf_counter() - start))
"""
圆周率值是: 3.141568
运行时间是: 0.25680s
"""
七段数码管绘制
import turtle
def drawLine(draw): # 绘制单段数码管
turtle.pendown() if draw else turtle.penup()
turtle.fd(40)
turtle.right(90)
def drawDigit(digit): # 根据数字绘制七段数码管
drawLine(True) if digit in [2, 3, 4, 5, 6, 8, 9] else drawLine(False)
drawLine(True) if digit in [0, 1, 3, 4, 5, 6, 7, 8, 9] else drawLine(False)
drawLine(True) if digit in [0, 2, 3, 5, 6, 8, 9] else drawLine(False)
drawLine(True) if digit in [0, 2, 6, 8] else drawLine(False)
turtle.left(90)
drawLine(True) if digit in [0, 4, 5, 6, 8, 9] else drawLine(False)
drawLine(True) if digit in [0, 2, 3, 5, 6, 7, 8, 9] else drawLine(False)
drawLine(True) if digit in [0, 1, 2, 3, 4, 7, 8, 9] else drawLine(False)
turtle.left(180)
turtle.penup()
turtle.fd(20)
def drawDate(date): # 获得要输出的数字
for i in date:
drawDigit(eval(i)) # 通过eval()函数将数字变为整数
def main():
turtle.setup(800, 350, 200, 200)
turtle.penup()
turtle.fd(-300)
turtle.pensize(5)
drawDate("20181010")
turtle.hideturtle()
turtle.done()
main()
import turtle, time
def drawGap(): # 绘制数码管间隔
turtle.penup()
turtle.fd(5)
def drawLine(draw): # 绘制单段数码管
drawGap()
turtle.pendown() if draw else turtle.penup()
turtle.fd(40)
drawGap()
turtle.right(90)
def drawDigit(d): # 根据数字绘制七段数码管
drawLine(True) if d in [2, 3, 4, 5, 6, 8, 9] else drawLine(False)
drawLine(True) if d in [0, 1, 3, 4, 5, 6, 7, 8, 9] else drawLine(False)
drawLine(True) if d in [0, 2, 3, 5, 6, 8, 9] else drawLine(False)
drawLine(True) if d in [0, 2, 6, 8] else drawLine(False)
turtle.left(90)
drawLine(True) if d in [0, 4, 5, 6, 8, 9] else drawLine(False)
drawLine(True) if d in [0, 2, 3, 5, 6, 7, 8, 9] else drawLine(False)
drawLine(True) if d in [0, 1, 2, 3, 4, 7, 8, 9] else drawLine(False)
turtle.left(180)
turtle.penup()
turtle.fd(20)
def drawDate(date):
turtle.pencolor("red")
for i in date:
if i == "-":
turtle.write("年", font=("Arial", 18, "normal"))
turtle.pencolor("green")
turtle.fd(40)
elif i == "=":
turtle.write("月", font=("Arial", 18, "normal"))
turtle.pencolor("blue")
turtle.fd(40)
elif i == "+":
turtle.write("日", font=("Arial", 18, "normal"))
else:
drawDigit(eval(i))
def main():
turtle.setup(800, 350, 200, 200)
turtle.penup()
turtle.fd(-350)
turtle.pensize(5)
# drawDate('2018-10=10+')
drawDate(time.strftime("%Y-%m=%d+", time.gmtime()))
turtle.hideturtle()
turtle.done()
main()
科赫雪花小包裹
import turtle
def koch(size, n):
if n == 0:
turtle.fd(size)
else:
for angle in [0, 60, -120, 60]:
turtle.left(angle)
koch(size / 3, n - 1)
def main():
turtle.setup(800, 400)
turtle.penup()
turtle.goto(-300, -50)
turtle.pendown()
turtle.pensize(2)
koch(600, 3) # 0阶科赫曲线长度,阶数
turtle.hideturtle()
turtle.done()
main()
import turtle
def koch(size, n):
if n == 0:
turtle.fd(size)
else:
for angle in [0, 60, -120, 60]:
turtle.left(angle)
koch(size / 3, n - 1)
def main():
turtle.setup(600, 600)
turtle.penup()
turtle.goto(-200, 100)
turtle.pendown()
turtle.pensize(2)
level = 3 # 3阶科赫雪花,阶数
koch(400, level)
turtle.right(120)
koch(400, level)
turtle.right(120)
koch(400, level)
turtle.hideturtle()
turtle.done()
main()
基本统计值计算
def getNum(): # 获取用户不定长度的输入
nums = []
iNumStr = input("请输入数字(回车退出): ")
while iNumStr != "":
nums.append(eval(iNumStr))
iNumStr = input("请输入数字(回车退出): ")
return nums
def mean(numbers): # 计算平均值
s = 0.0
for num in numbers:
s = s + num
return s / len(numbers)
def dev(numbers, mean): # 计算方差
sdev = 0.0
for num in numbers:
sdev = sdev + (num - mean) ** 2
return pow(sdev / (len(numbers) - 1), 0.5)
def median(numbers): # 计算中位数
sorted(numbers)
size = len(numbers)
if size % 2 == 0:
med = (numbers[size // 2 - 1] + numbers[size // 2]) / 2
else:
med = numbers[size // 2]
return med
n = getNum() # 主体函数
m = mean(n)
print("平均值:{},方差:{:.2},中位数:{}.".format(m, dev(n, m), median(n)))
"""
请输入数字(回车退出): 1
请输入数字(回车退出): 2
请输入数字(回车退出): 3
请输入数字(回车退出): 4
请输入数字(回车退出): 5
请输入数字(回车退出):
平均值:3.0,方差:1.6,中位数:3.
"""
文本词频统计
def getText():
txt = open("hamlet.txt", "r").read()
txt = txt.lower()
for ch in '!"#$%&()*+,-./:;<=>?@[\\]^_‘{|}~':
txt = txt.replace(ch, " ") # 将文本中特殊字符替换为空格
return txt
hamletTxt = getText()
words = hamletTxt.split()
counts = {}
for word in words:
counts[word] = counts.get(word, 0) + 1
items = list(counts.items())
items.sort(key=lambda x: x[1], reverse=True)
for i in range(10):
word, count = items[i]
print("{0:<10}{1:>5}".format(word, count))
"""
the 1138
and 965
to 754
of 669
you 550
i 542
a 542
my 514
hamlet 462
in 436
"""
import jieba
txt = open("threekingdoms.txt", "r", encoding="utf-8").read()
words = jieba.lcut(txt)
counts = {}
for word in words:
if len(word) == 1:
continue
else:
counts[word] = counts.get(word, 0) + 1
items = list(counts.items())
items.sort(key=lambda x: x[1], reverse=True)
for i in range(15):
word, count = items[i]
print("{0:<10}{1:>5}".format(word, count))
"""
曹操 953
孔明 836
将军 772
却说 656
玄德 585
关公 510
丞相 491
二人 469
不可 440
荆州 425
玄德曰 390
孔明曰 390
不能 384
如此 378
张飞 358
"""
import jieba
excludes = {"将军", "却说", "荆州", "二人", "不可", "不能", "如此"}
txt = open("threekingdoms.txt", "r", encoding="utf-8").read()
words = jieba.lcut(txt)
counts = {}
for word in words:
if len(word) == 1:
continue
elif word == "诸葛亮" or word == "孔明曰":
rword = "孔明"
elif word == "关公" or word == "云长":
rword = "关羽"
elif word == "玄德" or word == "玄德曰":
rword = "刘备"
elif word == "孟德" or word == "丞相":
rword = "曹操"
else:
rword = word
counts[rword] = counts.get(rword, 0) + 1
for word in excludes:
del counts[word]
items = list(counts.items())
items.sort(key=lambda x: x[1], reverse=True)
for i in range(10):
word, count = items[i]
print("{0:<10}{1:>5}".format(word, count))
"""
曹操 1451
孔明 1383
刘备 1252
关羽 784
张飞 358
商议 344
如何 338
主公 331
军士 317
吕布 300
"""
自动轨迹绘制
import turtle as t
t.title("自动轨迹绘制")
t.setup(800, 600, 0, 0)
t.pencolor("red")
t.pensize(5)
# 数据读取
datals = []
f = open("data.txt")
for line in f:
line = line.replace("\n", "")
datals.append(list(map(eval, line.split(","))))
f.close()
# 自动绘制
for i in range(len(datals)):
t.pencolor(datals[i][3], datals[i][4], datals[i][5])
t.fd(datals[i][0])
if datals[i][1]:
t.rt(datals[i][2])
else:
t.lt(datals[i][2])
t.done()
政府工作报告词云
import jieba
import wordcloud
f = open("新时代中国特色社会主义.txt", "r", encoding="utf-8")
t = f.read()
f.close()
ls = jieba.lcut(t)
txt = " ".join(ls)
w = wordcloud.WordCloud(
width=1000, height=700, background_color="white", font_path="msyh.ttc"
)
w.generate(txt)
w.to_file("grwordcloud.png")
import jieba
import wordcloud
f = open("关于实施乡村振兴战略的意见.txt", "r", encoding="utf-8")
t = f.read()
f.close()
ls = jieba.lcut(t)
txt = " ".join(ls)
w = wordcloud.WordCloud(
width=1000, height=700, background_color="white", font_path="msyh.ttc"
)
w.generate(txt)
w.to_file("grwordcloud.png")
import jieba
import wordcloud
f = open("新时代中国特色社会主义.txt", "r", encoding="utf-8")
t = f.read()
f.close()
ls = jieba.lcut(t)
txt = " ".join(ls)
w = wordcloud.WordCloud(
width=1000, height=700, background_color="white", font_path="msyh.ttc", max_words=15
)
w.generate(txt)
w.to_file("grwordcloud.png")
import jieba
import wordcloud
f = open("关于实施乡村振兴战略的意见.txt", "r", encoding="utf-8")
t = f.read()
f.close()
ls = jieba.lcut(t)
txt = " ".join(ls)
w = wordcloud.WordCloud(
width=1000, height=700, background_color="white", font_path="msyh.ttc", max_words=15
)
w.generate(txt)
w.to_file("grwordcloud.png")
import jieba
import wordcloud
from imageio import imread
mask = imread("fivestart.png")
excludes = {}
f = open("新时代中国特色社会主义.txt", "r", encoding="utf-8")
t = f.read()
f.close()
ls = jieba.lcut(t)
txt = " ".join(ls)
w = wordcloud.WordCloud(
width=1000, height=700, background_color="white", font_path="msyh.ttc", mask=mask
)
w.generate(txt)
w.to_file("grwordcloudm.png")
import jieba
import wordcloud
from imageio import imread
mask = imread("fivestart.png")
excludes = {}
f = open("关于实施乡村振兴战略的意见.txt", "r", encoding="utf-8")
t = f.read()
f.close()
ls = jieba.lcut(t)
txt = " ".join(ls)
w = wordcloud.WordCloud(
width=1000, height=700, background_color="white", font_path="msyh.ttc", mask=mask
)
w.generate(txt)
w.to_file("grwordcloudm.png")
import jieba
import wordcloud
from imageio import imread
mask = imread("chinamap.jpg")
excludes = {}
f = open("新时代中国特色社会主义.txt", "r", encoding="utf-8")
t = f.read()
f.close()
ls = jieba.lcut(t)
txt = " ".join(ls)
w = wordcloud.WordCloud(
width=1000, height=700, background_color="white", font_path="msyh.ttc", mask=mask
)
w.generate(txt)
w.to_file("grwordcloudm.png")
import jieba
import wordcloud
from imageio import imread
mask = imread("chinamap.jpg")
excludes = {}
f = open("关于实施乡村振兴战略的意见.txt", "r", encoding="utf-8")
t = f.read()
f.close()
ls = jieba.lcut(t)
txt = " ".join(ls)
w = wordcloud.WordCloud(
width=1000, height=700, background_color="white", font_path="msyh.ttc", mask=mask
)
w.generate(txt)
w.to_file("grwordcloudm.png")
体育竞技分析
from random import random
def printIntro():
print("这个程序模拟两个选手A和B的某种竞技比赛")
print("程序运行需要A和B的能力值(以0到1之间的小数表示)")
def getInputs():
a = eval(input("请输入选手A的能力值(0-1): "))
b = eval(input("请输入选手B的能力值(0-1): "))
n = eval(input("模拟比赛的场次: "))
return a, b, n
def simNGames(n, probA, probB):
winsA, winsB = 0, 0
for i in range(n):
scoreA, scoreB = simOneGame(probA, probB)
if scoreA > scoreB:
winsA += 1
else:
winsB += 1
return winsA, winsB
def gameOver(a, b):
return a == 15 or b == 15
def simOneGame(probA, probB):
scoreA, scoreB = 0, 0
serving = "A"
while not gameOver(scoreA, scoreB):
if serving == "A":
if random() < probA:
scoreA += 1
else:
serving = "B"
else:
if random() < probB:
scoreB += 1
else:
serving = "A"
return scoreA, scoreB
def printSummary(winsA, winsB):
n = winsA + winsB
print("竞技分析开始,共模拟{}场比赛".format(n))
print("选手A获胜{}场比赛,占比{:0.1%}".format(winsA, winsA / n))
print("选手B获胜{}场比赛,占比{:0.1%}".format(winsB, winsB / n))
def main():
printIntro()
probA, probB, n = getInputs()
winsA, winsB = simNGames(n, probA, probB)
printSummary(winsA, winsB)
main()
"""
这个程序模拟两个选手A和B的某种竞技比赛
程序运行需要A和B的能力值(以0到1之间的小数表示)
请输入选手A的能力值(0-1): 0.5
请输入选手B的能力值(0-1): 0.5
模拟比赛的场次: 1000
竞技分析开始,共模拟1000场比赛
选手A获胜548场比赛,占比54.8%
选手B获胜452场比赛,占比45.2%
这个程序模拟两个选手A和B的某种竞技比赛
程序运行需要A和B的能力值(以0到1之间的小数表示)
请输入选手A的能力值(0-1): 0.5
请输入选手B的能力值(0-1): 0.5
模拟比赛的场次: 1000
竞技分析开始,共模拟1000场比赛
选手A获胜499场比赛,占比49.9%
选手B获胜501场比赛,占比50.1%
"""
第三方库安装脚本
import os
libs = {
"numpy",
"matplotlib",
"pillow",
"sklearn",
"requests",
"jieba",
"beautifulsoup4",
"wheel",
"networkx",
"sympy",
"pyinstaller",
"django",
"flask",
"werobot",
"pyqt5",
"pandas",
"pyopengl",
"pypdf2",
"docopt",
"pygame",
}
try:
for lib in libs:
os.system("pip3 install " + lib)
print("Successful")
except:
print("Failed Somehow")
霍兰德人格分析雷达图
import numpy as np
import matplotlib.pyplot as plt
import matplotlib
matplotlib.rcParams["font.family"] = "Hiragino Sans GB"
radar_labels = np.array(
["研究型(I)", "艺术型(A)", "社会型(S)", "企业型(E)", "常规型(C)", "现实型(R)", "研究型(I)"]
) # 雷达标签,增加一个起始元素以与 angles 数组长度匹配
nAttr = 6
data = np.array(
[
[0.40, 0.32, 0.35, 0.30, 0.30, 0.88],
[0.85, 0.35, 0.30, 0.40, 0.40, 0.30],
[0.43, 0.89, 0.30, 0.28, 0.22, 0.30],
[0.30, 0.25, 0.48, 0.85, 0.45, 0.40],
[0.20, 0.38, 0.87, 0.45, 0.32, 0.28],
[0.34, 0.31, 0.38, 0.40, 0.92, 0.28],
]
) # 数据值
data_labels = ("艺术家", "实验员", "工程师", "推销员", "社会工作者", "记事员")
angles = np.linspace(0, 2 * np.pi, nAttr, endpoint=False)
data = np.concatenate((data, [data[0]]))
angles = np.concatenate((angles, [angles[0]]))
fig = plt.figure(facecolor="white")
plt.subplot(111, polar=True)
plt.plot(angles, data, "o-", linewidth=1, alpha=0.2)
plt.fill(angles, data, alpha=0.25)
plt.thetagrids(angles * 180 / np.pi, radar_labels) # 移除 frac 参数
plt.figtext(0.52, 0.95, "霍兰德人格分析", ha="center", size=20)
legend = plt.legend(data_labels, loc=(0.94, 0.80), labelspacing=0.1)
plt.setp(legend.get_texts(), fontsize="large")
plt.grid(True)
plt.savefig("holland_radar.jpg")
plt.show()
玫瑰花绘制
import turtle as t
# 定义一个曲线绘制函数
def DegreeCurve(n, r, d=1):
for i in range(n):
t.left(d)
t.circle(r, abs(d))
# 初始位置设定
s = 0.2 # size
t.setup(450 * 5 * s, 750 * 5 * s)
t.pencolor("black")
t.fillcolor("red")
t.speed(100)
t.penup()
t.goto(0, 900 * s)
t.pendown()
# 绘制花朵形状
t.begin_fill()
t.circle(200 * s, 30)
DegreeCurve(60, 50 * s)
t.circle(200 * s, 30)
DegreeCurve(4, 100 * s)
t.circle(200 * s, 50)
DegreeCurve(50, 50 * s)
t.circle(350 * s, 65)
DegreeCurve(40, 70 * s)
t.circle(150 * s, 50)
DegreeCurve(20, 50 * s, -1)
t.circle(400 * s, 60)
DegreeCurve(18, 50 * s)
t.fd(250 * s)
t.right(150)
t.circle(-500 * s, 12)
t.left(140)
t.circle(550 * s, 110)
t.left(27)
t.circle(650 * s, 100)
t.left(130)
t.circle(-300 * s, 20)
t.right(123)
t.circle(220 * s, 57)
t.end_fill()
# 绘制花枝形状
t.left(120)
t.fd(280 * s)
t.left(115)
t.circle(300 * s, 33)
t.left(180)
t.circle(-300 * s, 33)
DegreeCurve(70, 225 * s, -1)
t.circle(350 * s, 104)
t.left(90)
t.circle(200 * s, 105)
t.circle(-500 * s, 63)
t.penup()
t.goto(170 * s, -30 * s)
t.pendown()
t.left(160)
DegreeCurve(20, 2500 * s)
DegreeCurve(220, 250 * s, -1)
# 绘制一个绿色叶子
t.fillcolor("green")
t.penup()
t.goto(670 * s, -180 * s)
t.pendown()
t.right(140)
t.begin_fill()
t.circle(300 * s, 120)
t.left(60)
t.circle(300 * s, 120)
t.end_fill()
t.penup()
t.goto(180 * s, -550 * s)
t.pendown()
t.right(85)
t.circle(600 * s, 40)
# 绘制另一个绿色叶子
t.penup()
t.goto(-150 * s, -1000 * s)
t.pendown()
t.begin_fill()
t.rt(120)
t.circle(300 * s, 115)
t.left(75)
t.circle(300 * s, 100)
t.end_fill()
t.penup()
t.goto(430 * s, -1070 * s)
t.pendown()
t.right(30)
t.circle(-600 * s, 35)
t.done()