最近,Analysis with Programming加入了Planet Python。做爲該網站的首批特約博客,我這裏來分享一下如何經過Python來開始數據分析。具體內容以下:html
這是很關鍵的一步,爲了後續的分析咱們首先須要導入數據。一般來講,數據是CSV格式,就算不是,至少也能夠轉換成CSV格式。在Python中,咱們的操做以下:python
1
2
3
4
5
6
7
8
|
import
pandas as pd
# Reading data locally
df
=
pd.read_csv(
'/Users/al-ahmadgaidasaad/Documents/d.csv'
)
# Reading data from web
data_url
=
"https://raw.githubusercontent.com/alstat/Analysis-with-Programming/master/2014/Python/Numerical-Descriptions-of-the-Data/data.csv"
df
=
pd.read_csv(data_url)
|
爲了讀取本地CSV文件,咱們須要pandas這個數據分析庫中的相應模塊。其中的read_csv函數可以讀取本地和web數據。git
既然在工做空間有了數據,接下來就是數據變換。統計學家和科學家們一般會在這一步移除分析中的非必要數據。咱們先看看數據:程序員
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
|
# Head of the data
print
df.head()
# OUTPUT
Abra Apayao Benguet Ifugao Kalinga
0
1243
2934
148
3300
10553
1
4158
9235
4287
8063
35257
2
1787
1922
1955
1074
4544
3
17152
14501
3536
19607
31687
4
1266
2385
2530
3315
8520
# Tail of the data
print
df.tail()
# OUTPUT
Abra Apayao Benguet Ifugao Kalinga
74
2505
20878
3519
19737
16513
75
60303
40065
7062
19422
61808
76
6311
6756
3561
15910
23349
77
13345
38902
2583
11096
68663
78
2623
18264
3745
16787
16900
|
對R語言程序員來講,上述操做等價於經過print(head(df))來打印數據的前6行,以及經過print(tail(df))來打印數據的後6行。固然Python中,默認打印是5行,而R則是6行。所以R的代碼head(df, n = 10),在Python中就是df.head(n = 10),打印數據尾部也是一樣道理。github
在R語言中,數據列和行的名字經過colnames和rownames來分別進行提取。在Python中,咱們則使用columns和index屬性來提取,以下:web
1
2
3
4
5
6
7
8
9
10
11
|
# Extracting column names
print
df.columns
# OUTPUT
Index([u
'Abra'
, u
'Apayao'
, u
'Benguet'
, u
'Ifugao'
, u
'Kalinga'
], dtype
=
'object'
)
# Extracting row names or the index
print
df.index
# OUTPUT
Int64Index([
0
,
1
,
2
,
3
,
4
,
5
,
6
,
7
,
8
,
9
,
10
,
11
,
12
,
13
,
14
,
15
,
16
,
17
,
18
,
19
,
20
,
21
,
22
,
23
,
24
,
25
,
26
,
27
,
28
,
29
,
30
,
31
,
32
,
33
,
34
,
35
,
36
,
37
,
38
,
39
,
40
,
41
,
42
,
43
,
44
,
45
,
46
,
47
,
48
,
49
,
50
,
51
,
52
,
53
,
54
,
55
,
56
,
57
,
58
,
59
,
60
,
61
,
62
,
63
,
64
,
65
,
66
,
67
,
68
,
69
,
70
,
71
,
72
,
73
,
74
,
75
,
76
,
77
,
78
], dtype
=
'int64'
)
|
數據轉置使用T方法,數組
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
|
# Transpose data
print
df.T
# OUTPUT
0
1
2
3
4
5
6
7
8
9
Abra
1243
4158
1787
17152
1266
5576
927
21540
1039
5424
Apayao
2934
9235
1922
14501
2385
7452
1099
17038
1382
10588
Benguet
148
4287
1955
3536
2530
771
2796
2463
2592
1064
Ifugao
3300
8063
1074
19607
3315
13134
5134
14226
6842
13828
Kalinga
10553
35257
4544
31687
8520
28252
3106
36238
4973
40140
...
69
70
71
72
73
74
75
76
77
Abra ...
12763
2470
59094
6209
13316
2505
60303
6311
13345
Apayao ...
37625
19532
35126
6335
38613
20878
40065
6756
38902
Benguet ...
2354
4045
5987
3530
2585
3519
7062
3561
2583
Ifugao ...
9838
17125
18940
15560
7746
19737
19422
15910
11096
Kalinga ...
65782
15279
52437
24385
66148
16513
61808
23349
68663
78
Abra
2623
Apayao
18264
Benguet
3745
Ifugao
16787
Kalinga
16900
Other transformations such as sort can be done using <code>sort<
/
code> attribute. Now let's extract a specific column. In Python, we do it using either <code>iloc<
/
code>
or
<code>ix<
/
code> attributes, but <code>ix<
/
code>
is
more robust
and
thus I prefer it. Assuming we want the head of the first column of the data, we have
|
其餘變換,例如排序就是用sort屬性。如今咱們提取特定的某列數據。Python中,可使用iloc或者ix屬性。可是我更喜歡用ix,由於它更穩定一些。假設咱們需數據第一列的前5行,咱們有:dom
1
2
3
4
5
6
7
8
9
|
print
df.ix[:,
0
].head()
# OUTPUT
0
1243
1
4158
2
1787
3
17152
4
1266
Name: Abra, dtype: int64
|
順便提一下,Python的索引是從0開始而非1。爲了取出從11到20行的前3列數據,咱們有:ide
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
|
print
df.ix[
10
:
20
,
0
:
3
]
# OUTPUT
Abra Apayao Benguet
10
981
1311
2560
11
27366
15093
3039
12
1100
1701
2382
13
7212
11001
1088
14
1048
1427
2847
15
25679
15661
2942
16
1055
2191
2119
17
5437
6461
734
18
1029
1183
2302
19
23710
12222
2598
20
1091
2343
2654
|
上述命令至關於df.ix[10:20, ['Abra', 'Apayao', 'Benguet']]
。函數
爲了捨棄數據中的列,這裏是列1(Apayao)和列2(Benguet),咱們使用drop屬性,以下:
1
2
3
4
5
6
7
8
9
|
print
df.drop(df.columns[[
1
,
2
]], axis
=
1
).head()
# OUTPUT
Abra Ifugao Kalinga
0
1243
3300
10553
1
4158
8063
35257
2
1787
1074
4544
3
17152
19607
31687
4
1266
3315
8520
|
axis
參數告訴函數到底捨棄列仍是行。若是axis
等於0,那麼就捨棄行。
下一步就是經過describe
屬性,對數據的統計特性進行描述:
1
2
3
4
5
6
7
8
9
10
11
12
|
print
df.describe()
# OUTPUT
Abra Apayao Benguet Ifugao Kalinga
count
79.000000
79.000000
79.000000
79.000000
79.000000
mean
12874.379747
16860.645570
3237.392405
12414.620253
30446.417722
std
16746.466945
15448.153794
1588.536429
5034.282019
22245.707692
min
927.000000
401.000000
148.000000
1074.000000
2346.000000
25
%
1524.000000
3435.500000
2328.000000
8205.000000
8601.500000
50
%
5790.000000
10588.000000
3202.000000
13044.000000
24494.000000
75
%
13330.500000
33289.000000
3918.500000
16099.500000
52510.500000
max
60303.000000
54625.000000
8813.000000
21031.000000
68663.000000
|
Python有一個很好的統計推斷包。那就是scipy裏面的stats。ttest_1samp實現了單樣本t檢驗。所以,若是咱們想檢驗數據Abra列的稻穀產量均值,經過零假設,這裏咱們假定整體稻穀產量均值爲15000,咱們有:
1
2
3
4
5
6
7
|
from
scipy
import
stats as ss
# Perform one sample t-test using 1500 as the true mean
print
ss.ttest_1samp(a
=
df.ix[:,
'Abra'
], popmean
=
15000
)
# OUTPUT
(
-
1.1281738488299586
,
0.26270472069109496
)
|
返回下述值組成的元祖:
經過上面的輸出,看到p值是0.267遠大於α等於0.05,所以沒有充分的證聽說平均稻穀產量不是150000。將這個檢驗應用到全部的變量,一樣假設均值爲15000,咱們有:
1
2
3
4
5
6
|
print
ss.ttest_1samp(a
=
df, popmean
=
15000
)
# OUTPUT
(array([
-
1.12817385
,
1.07053437
,
-
65.81425599
,
-
4.564575
,
6.17156198
]),
array([
2.62704721e
-
01
,
2.87680340e
-
01
,
4.15643528e
-
70
,
1.83764399e
-
05
,
2.82461897e
-
08
]))
|
第一個數組是t統計量,第二個數組則是相應的p值。
Python中有許多可視化模塊,最流行的當屬matpalotlib庫。稍加說起,咱們也可選擇bokeh和seaborn模塊。以前的博文中,我已經說明了matplotlib庫中的盒須圖模塊功能。
1
2
3
|
# Import the module for plotting
import
matplotlib.pyplot as plt
plt.show(df.plot(kind
=
'box'
))
|
如今,咱們能夠用pandas模塊中集成R的ggplot主題來美化圖表。要使用ggplot,咱們只須要在上述代碼中多加一行,
1
2
3
|
import
matplotlib.pyplot as plt
pd.options.display.mpl_style
=
'default'
# Sets the plotting display theme to ggplot2
df.plot(kind
=
'box'
)
|
這樣咱們就獲得以下圖表:
比matplotlib.pyplot主題簡潔太多。可是在本博文中,我更願意引入seaborn模塊,該模塊是一個統計數據可視化庫。所以咱們有:
1
2
3
4
|
# Import the seaborn library
import
seaborn as sns
# Do the boxplot
plt.show(sns.boxplot(df, widths
=
0.5
, color
=
"pastel"
))
|
多性感的盒式圖,繼續往下看。
1
|
plt.show(sns.violinplot(df, widths
=
0.5
, color
=
"pastel"
))
|
1
|
plt.show(sns.distplot(df.ix[:,
2
], rug
=
True
, bins
=
15
))
|
1
2
|
with sns.axes_style(
"white"
):
plt.show(sns.jointplot(df.ix[:,
1
], df.ix[:,
2
], kind
=
"kde"
))
|
1
|
plt.show(sns.lmplot(
"Benguet"
,
"Ifugao"
, df))
|
在Python中,咱們使用def函數來實現一個自定義函數。例如,若是咱們要定義一個兩數相加的函數,以下便可:
1
2
3
4
5
6
7
|
def
add_2int(x, y):
return
x
+
y
print
add_2int(
2
,
2
)
# OUTPUT
4
|
順便說一下,Python中的縮進是很重要的。經過縮進來定義函數做用域,就像在R語言中使用大括號{…}同樣。這有一個咱們以前博文的例子:
Python中,程序以下:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
|
import
numpy as np
import
scipy.stats as ss
def
case(n
=
10
, mu
=
3
, sigma
=
np.sqrt(
5
), p
=
0.025
, rep
=
100
):
m
=
np.zeros((rep,
4
))
for
i
in
range
(rep):
norm
=
np.random.normal(loc
=
mu, scale
=
sigma, size
=
n)
xbar
=
np.mean(norm)
low
=
xbar
-
ss.norm.ppf(q
=
1
-
p)
*
(sigma
/
np.sqrt(n))
up
=
xbar
+
ss.norm.ppf(q
=
1
-
p)
*
(sigma
/
np.sqrt(n))
if
(mu > low) & (mu < up):
rem
=
1
else
:
rem
=
0
m[i, :]
=
[xbar, low, up, rem]
inside
=
np.
sum
(m[:,
3
])
per
=
inside
/
rep
desc
=
"There are "
+
str
(inside)
+
" confidence intervals that contain "
"the true mean ("
+
str
(mu)
+
"), that is "
+
str
(per)
+
" percent of the total CIs"
return
{
"Matrix"
: m,
"Decision"
: desc}
|
上述代碼讀起來很簡單,可是循環的時候就很慢了。下面針對上述代碼進行了改進,這多虧了 Python專家,看我上篇博文的15條意見吧。
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
|
import
numpy as np
import
scipy.stats as ss
def
case2(n
=
10
, mu
=
3
, sigma
=
np.sqrt(
5
), p
=
0.025
, rep
=
100
):
scaled_crit
=
ss.norm.ppf(q
=
1
-
p)
*
(sigma
/
np.sqrt(n))
norm
=
np.random.normal(loc
=
mu, scale
=
sigma, size
=
(rep, n))
xbar
=
norm.mean(
1
)
low
=
xbar
-
scaled_crit
up
=
xbar
+
scaled_crit
rem
=
(mu > low) & (mu < up)
m
=
np.c_[xbar, low, up, rem]
inside
=
np.
sum
(m[:,
3
])
per
=
inside
/
rep
desc
=
"There are "
+
str
(inside)
+
" confidence intervals that contain "
"the true mean ("
+
str
(mu)
+
"), that is "
+
str
(per)
+
" percent of the total CIs"
return
{
"Matrix"
: m,
"Decision"
: desc}
|
那些對於本文ipython notebook版本感興趣的,請點擊這裏。這篇文章由Nuttens Claude負責轉換成 ipython notebook 。