In [1]:
%matplotlib inline
import matplotlib.pyplot as plt
In [2]:
import pandas as pd
使用pandas清洗泰坦尼克数据¶
In [3]:
# Import the Pandas library
import pandas as pd
# Load the train and test datasets to create two DataFrames
train_url = "http://s3.amazonaws.com/assets.datacamp.com/course/Kaggle/train.csv"
train = pd.read_csv(train_url)
test_url = "http://s3.amazonaws.com/assets.datacamp.com/course/Kaggle/test.csv"
test = pd.read_csv(test_url)
#Print the `head` of the train and test dataframes
train.head()
Out[3]:
In [3]:
test.head()
Out[3]:
In [3]:
train.to_csv('../data/tatanic_train.csv')
test.to_csv('../data/tatanic_test.csv')
You can easily explore a DataFrame
- .describe() summarizes the columns/features of the DataFrame, including the count of observations, mean, max and so on.
- Another useful trick is to look at the dimensions of the DataFrame. This is done by requesting the .shape attribute of your DataFrame object. (ex. your_data.shape)
从本机读取数据¶
In [19]:
import pandas as pd
train = pd.read_csv('../data/tatanic_train.csv',\
sep = ",", header=0)
test = pd.read_csv('../data/tatanic_test.csv',\
sep = ",", header=0)
In [4]:
train.head()
Out[4]:
In [5]:
train.describe()
Out[5]:
In [8]:
train.shape#, len(train)
#train.columns
Out[8]:
In [9]:
# Passengers that survived vs passengers that passed away
train["Survived"]
Out[9]:
In [27]:
# Passengers that survived vs passengers that passed away
train["Survived"].value_counts()
Out[27]:
In [9]:
# As proportions
train["Survived"].value_counts(normalize = True)
Out[9]:
In [10]:
train['Sex'].value_counts()
Out[10]:
In [14]:
train[train['Sex']=='female']#[train['Pclass'] == 3]
Out[14]:
In [15]:
# Males that survived vs males that passed away
train["Survived"][train["Sex"] == 'male']
Out[15]:
In [15]:
# Males that survived vs males that passed away
train["Survived"][train["Sex"] == 'male'].value_counts()
Out[15]:
In [31]:
# Females that survived vs Females that passed away
train["Survived"][train["Sex"] == 'female'].value_counts()
Out[31]:
In [32]:
# Normalized male survival
train["Survived"][train["Sex"] == 'male'].value_counts(normalize = True)
Out[32]:
In [33]:
# Normalized female survival
train["Survived"][train["Sex"] == 'female'].value_counts(normalize = True)
Out[33]:
In [16]:
# Create the column Child, and indicate whether child or not a child. Print the new column.
train["Child"] = float('NaN')
train.Child[train.Age < 5] = 1
train.Child[train.Age >= 5] = 0
print(train.Child[:3])
In [12]:
# Normalized Survival Rates for under 18
train.Survived[train.Child == 1].value_counts(normalize = True)
Out[12]:
In [24]:
# Normalized Survival Rates for over 18
train.Survived[train.Child == 0].value_counts(normalize = True)
Out[24]:
In [21]:
# Create a copy of test: test_one
test_one = test
# Initialize a Survived column to 0
test_one['Survived'] = 0
# Set Survived to 1 if Sex equals "female" and print the `Survived` column from `test_one`
test_one.Survived[test_one.Sex =='female'] = 1
print(test_one.Survived[:3])
In [26]:
#Convert the male and female groups to integer form
train["Sex"][train["Sex"] == "male"] = 0
train["Sex"][train["Sex"] == "female"] = 1
#Impute the Embarked variable
train["Embarked"] = train["Embarked"].fillna('S')
#Convert the Embarked classes to integer form
train["Embarked"][train["Embarked"] == "S"] = 0
train["Embarked"][train["Embarked"] == "C"] = 1
train["Embarked"][train["Embarked"] == "Q"] = 2
分析天涯回帖数据¶
In [22]:
df = pd.read_csv('../data/tianya_bbs_threads_list.txt',\
sep = "\t", names = ['title','link', \
'author','author_page',\
'click','reply','time'])
df[:2]
Out[22]:
In [24]:
da = pd.read_csv('../data/tianya_bbs_threads_author_info.txt',
sep = "\t", names = ['author_page','followed_num',\
'fans_num','post_num', \
'comment_num'])
da[:2]
Out[24]:
In [25]:
data = pd.concat([df,da], axis=1)
len(data)
Out[25]:
In [26]:
data[:3]
Out[26]:
Time¶
In [27]:
type(data.time[0])
Out[27]:
In [28]:
# extract date from datetime
# date = map(lambda x: x[:10], data.time)
date = [i[:10] for i in data.time]
#date = [i[:10] for i in data.time]
data['date'] = pd.to_datetime(date)
In [32]:
# convert str to datetime format
data.time = pd.to_datetime(data.time)
data['month'] = data.time.dt.month
data['year'] = data.time.dt.year
data['day'] = data.time.dt.day
type(data.time[0])
Out[32]:
In [34]:
data.describe()
#data.head()
Out[34]:
Statsmodels¶
http://statsmodels.sourceforge.net/
Statsmodels is a Python module that allows users to explore data, estimate statistical models, and perform statistical tests.
An extensive list of descriptive statistics, statistical tests, plotting functions, and result statistics are available for different types of data and each estimator.
Researchers across fields may find that statsmodels fully meets their needs for statistical computing and data analysis in Python.
Features include:¶
- Linear regression models
- Generalized linear models
- Discrete choice models
- Robust linear models
- Many models and functions for time series analysis
- Nonparametric estimators
- A collection of datasets for examples
- A wide range of statistical tests
- Input-output tools for producing tables in a number of formats and for reading Stata files into NumPy and Pandas.
- Plotting functions
- Extensive unit tests to ensure correctness of results
- Many more models and extensions in development
In [35]:
import statsmodels.api as sm
In [36]:
' '.join(dir(sm))
Out[36]:
In [37]:
' '.join(dir(sm.stats))
Out[37]:
Describe¶
In [38]:
data.describe()
Out[38]:
In [39]:
import numpy as np
np.mean(data.click), np.std(data.click), np.sum(data.click)
Out[39]:
In [40]:
# 不加权的变量描述
d1 = sm.stats.DescrStatsW(data.click, \
weights=[1 for i in data.click])
d1.mean, d1.var, d1.std, d1.sum
Out[40]:
In [41]:
# 加权的变量描述
d1 = sm.stats.DescrStatsW(data.click, weights=data.reply)
d1.mean, d1.var, d1.std, d1.sum
Out[41]:
In [163]:
np.median(data.click) # np.percentile
Out[163]:
In [42]:
plt.hist(data.click)
plt.show()
In [43]:
plt.hist(data.reply, color = 'green')
plt.show()
In [45]:
plt.hist(np.log(data.click+1), color='green')
plt.hist(np.log(data.reply+1), color='red')
plt.show()
In [46]:
# Plot the height and weight to see
plt.boxplot([np.log(data.click+1)])
plt.show()
In [47]:
# Plot the height and weight to see
plt.boxplot([data.click, data.reply])
plt.show()
In [48]:
def transformData(dat):
results = []
for i in dat:
if i != 'na':
results.append( int(i))
else:
results.append(0)
return results
In [49]:
data.fans_num = transformData(data.fans_num)
data.followed_num = transformData(data.followed_num )
data.post_num = transformData(data.post_num )
data.comment_num = transformData(data.comment_num )
data.describe()
Out[49]:
In [50]:
import numpy as np
# Plot the height and weight to see
plt.boxplot([np.log(data.click+1), np.log(data.reply+1),
np.log(data.fans_num+1),\
np.log(data.followed_num + 1)],
labels = ['$Click$', '$Reply$', '$Fans$',\
'$Followed$'])
plt.show()
Pandas自身已经包含了boxplot的功能¶
In [51]:
fig = plt.figure(figsize=(12,4))
data.boxplot(return_type='dict')
plt.yscale('log')
plt.show()
In [42]:
' '.join(dir(data)[200:])
Out[42]:
In [52]:
from pandas.tools import plotting
# fig = plt.figure(figsize=(10, 10))
plotting.scatter_matrix(data[['click', 'reply',\
'post_num','comment_num']])
plt.show()
In [450]:
' '.join(dir(plotting))
Out[450]:
更多使用pandas.plotting绘图的操作见：¶
http://pandas.pydata.org/pandas-docs/version/0.15.0/visualization.html
In [53]:
import seaborn # conda install seaborn
seaborn.pairplot(data, vars=['click', 'reply', \
'post_num', 'comment_num'],
kind='reg')
Out[53]:
In [54]:
seaborn.pairplot(data, vars=['click', 'reply', 'post_num'],
kind='reg', hue='year')
Out[54]:
In [55]:
seaborn.lmplot(y='reply', x='click', data=data,
size = 5)
plt.show()
values_counts¶
In [56]:
data.year.value_counts()
Out[56]:
In [57]:
d = data.year.value_counts()
dd = pd.DataFrame(d)
dd = dd.sort_index(axis=0, ascending=True)
dd
Out[57]:
In [58]:
dd.index
Out[58]:
In [64]:
dd_date_str = list(map(lambda x: str(x) +'-01-01', dd.index))
dd_date_str
Out[64]:
In [65]:
dd_date = pd.to_datetime(list(dd_date_str))
dd_date
Out[65]:
In [66]:
plt.plot(dd_date, dd.year, 'r-o')
plt.show()
In [67]:
ds = dd.cumsum()
ds
Out[67]:
In [70]:
d = data.year.value_counts()
dd = pd.DataFrame(d)
dd = dd.sort_index(axis=0, ascending=True)
ds = dd.cumsum()
def getDate(dat):
dat_date_str = list(map(lambda x: str(x) +'-01-01', dat.index))
dat_date = pd.to_datetime(dat_date_str)
return dat_date
ds.date = getDate(ds)
dd.date = getDate(dd)
plt.plot(ds.date, ds.year, 'g-s', label = '$Cumulative\: Number\:of\: Threads$')
plt.plot(dd.date, dd.year, 'r-o', label = '$Yearly\:Number\:of\:Threads$')
plt.legend(loc=2,numpoints=1,fontsize=13)
plt.show()
groupby¶
In [72]:
dg = data.groupby('year').sum()
dg
Out[72]:
In [73]:
dgs = dg.cumsum()
dgs
Out[73]:
In [74]:
def getDate(dat):
dat_date_str = list(map(lambda x: str(x) +'-01-01', dat.index))
dat_date = pd.to_datetime(dat_date_str)
return dat_date
dg.date = getDate(dg)
In [75]:
fig = plt.figure(figsize=(12,5))
plt.plot(dg.date, dg.click, 'r-o', label = '$Yearly\:Number\:of\:Clicks$')
plt.plot(dg.date, dg.reply, 'g-s', label = '$Yearly\:Number\:of\:Replies$')
plt.plot(dg.date, dg.fans_num, 'b->', label = '$Yearly\:Number\:of\:Fans$')
plt.yscale('log')
plt.legend(loc=4,numpoints=1,fontsize=13)
plt.show()
In [76]:
data.groupby('year')['click'].sum()
Out[76]:
In [275]:
data.groupby('year')['click'].mean()
Out[275]:
常用的统计分析方法¶
- t检验
- 卡方检验
- 相关
- 回归
RQ: 转载的文章的点击量是否显著地小于原创的文章？¶
In [62]:
repost = []
for i in df.title:
if u'转载' in i:
repost.append(1)
else:
repost.append(0)
In [63]:
df['repost'] = repost
In [64]:
df.groupby('repost').mean()
Out[64]:
In [48]:
df['click'][df['repost']==0][:5]
Out[48]:
In [49]:
df['click'][df['repost']==1][:5]
Out[49]:
In [323]:
from scipy import stats
stats.ttest_ind(data.click, data.repost)
Out[323]:
In [324]:
sm.stats.ttest_ind(data.click, data.reply)
# test statistic, pvalue and degrees of freedom
Out[324]:
A chi-squared test¶
https://en.wikipedia.org/wiki/Chi-squared_test
- also referred to as χ² test (or chi-square test), is any statistical hypothesis test in which the sampling distribution of the test statistic is a chi-square distribution when the null hypothesis is true.
- A chi-squared test can then be used to
reject the null hypothesis that the data are independent. - Test statistics that follow a chi-squared distribution arise from an
assumption of independent normally distributed data, which is valid in many cases due tothe central limit theorem. - Chi-squared tests are often constructed from a
sum of squared errors, or through thesample variance.
Suppose there is a city of 1 million residents with
four neighborhoods: A, B, C, and D.A random sample of 650 residents of the city is taken and their occupation is recorded as "blue collar", "white collar", or "no collar".
The null hypothesisis that each person's neighborhood of residence is independent of the person's occupational classification. The data are tabulated as:
| A | B | C | D | Total | |
|---|---|---|---|---|---|
| White collar | 90 | 60 | 104 | 95 | 349 |
| Blue collar | 30 | 50 | 51 | 20 | 151 |
| No coloar | 30 | 40 | 45 | 35 | 150 |
| Total | 150 | 150 | 200 | 150 | 650 |
- Let us take the sample living in neighborhood A, 150/650, to estimate what proportion of the whole 1 million people live in neighborhood A.
- Similarly we take 349/650 to estimate what proportion of the 1 million people are white-collar workers.
- By the assumption of independence under the hypothesis we should "expect" the number of
white-collar workers in neighborhood Ato be
$ \frac{150}{650} \frac{349}{650} 650 = 80.54 $
Then in that "cell" of the table, we have
$\frac{(\text{observed}-\text{expected})^2}{\text{expected}} = \frac{(90-80.54)^2}{80.54}$.
The sum of these quantities over all of the cells is the test statistic.
Under the null hypothesis, it has approximately a chi-square distribution whose number of degrees of freedom are
$ (\text{number of rows}-1)(\text{number of columns}-1) = (3-1)(4-1) = 6. $
If the test statistic is improbably large according to that chi-square distribution, then one rejects the null hypothesis of independence.
scipy.stats.chisquare(f_obs, f_exp=None, ddof=0, axis=0)[source]¶
- Calculates a one-way chi square test.
- The chi square test tests the null hypothesis that the categorical data has the given frequencies.
Parameters:
- f_obs : array_like Observed frequencies in each category.
- f_exp : array_like, optional Expected frequencies in each category. By default the categories are assumed to be equally likely.
- ddof : int, optional
In [428]:
from scipy.stats import chisquare
chisquare([16, 18, 16, 14, 12, 12], \
f_exp=[16, 16, 16, 16, 16, 8])
Out[428]:
In [90]:
from scipy.stats import chi2
# p_value = chi2.sf(chi_statistic, df)
chi2.sf(3.5, 5), 1 - chi2.cdf(3.5,5)
Out[90]:
Correlation¶
In [65]:
print np.corrcoef(data.click, data.reply)
print np.corrcoef(np.log(data.click+1), \
np.log(data.reply+1))
In [383]:
data.corr()
Out[383]:
In [13]:
plt.plot(df.click, df.reply, 'r-o')
plt.show()
In [16]:
plt.plot(df.click, df.reply, 'gs')
plt.xlabel('$Clicks$', fontsize = 20)
plt.ylabel('$Replies$', fontsize = 20)
plt.xscale('log')
plt.yscale('log')
plt.title('$Allowmetric\,Law$', fontsize = 20)
plt.show()
Regression¶
In [66]:
import numpy as np
import statsmodels.api as sm
import statsmodels.formula.api as smf
In [67]:
# Load data
dat = sm.datasets.get_rdataset("Guerry", "HistData").data
# Fit regression model (using the natural log of one of the regressors)
results = smf.ols('Lottery ~ Literacy + np.log(Pop1831)', \
data=dat).fit()
有些使用windows的同学无法运行上述代码：
在spyder中打开terminal¶
输入: pip install -U patsy¶
https://groups.google.com/forum/#!topic/pystatsmodels/KcSzNqDxv-Q
In [21]:
# Inspect the results
print results.summary()
In [68]:
reg = smf.ols('reply ~ click + followed_num', \
data=data).fit()
In [138]:
print reg.summary()
In [60]:
reg1 = smf.ols('np.log(reply+1) ~ np.log(click+1) \
+np.log(followed_num+1)+month', data=data).fit()
print reg1.summary()
In [209]:
fig = plt.figure(figsize=(12,8))
fig = sm.graphics.plot_partregress_grid(reg1, fig = fig)
plt.show()
In [429]:
import statsmodels.api as sm
from statsmodels.formula.api import ols
moore = sm.datasets.get_rdataset("Moore", "car",
cache=True) # load data
data = moore.data
data = data.rename(columns={"partner.status" :
"partner_status"}) # make name pythonic
In [434]:
data[:5]
Out[434]:
In [430]:
moore_lm = ols('conformity ~ C(fcategory, Sum)*C(partner_status, Sum)',
data=data).fit()
In [432]:
fig = plt.figure(figsize=(12,8))
fig = sm.graphics.plot_partregress_grid(moore_lm, fig = fig)
plt.show()
In [431]:
table = sm.stats.anova_lm(moore_lm, typ=2) # Type 2 ANOVA DataFrame
print table
参考文献¶
- scipy-lectures http://www.scipy-lectures.org/index.html