matplotlib：如何防止x轴标签相互重叠

Question

我正在使用matplotlib生成条形图。这一切都很好但我无法弄清楚如何防止x轴的标签相互重叠。这是一个例子：

以下是postgres 9.1数据库的一些示例SQL：

drop table if exists mytable;
create table mytable(id bigint, version smallint, date_from timestamp without time zone);
insert into mytable(id, version, date_from) values

('4084036', '1', '2006-12-22 22:46:35'),
('4084938', '1', '2006-12-23 16:19:13'),
('4084938', '2', '2006-12-23 16:20:23'),
('4084939', '1', '2006-12-23 16:29:14'),
('4084954', '1', '2006-12-23 16:28:28'),
('4250653', '1', '2007-02-12 21:58:53'),
('4250657', '1', '2007-03-12 21:58:53')
;  

这是我的python脚本：

# -*- coding: utf-8 -*-
#!/usr/bin/python2.7
import psycopg2
import matplotlib.pyplot as plt
fig = plt.figure()

# for savefig()
import pylab

###
### Connect to database with psycopg2
###

try:
  conn_string="dbname='x' user='y' host='z' password='pw'"
  print "Connecting to database\n->%s" % (conn_string)

  conn = psycopg2.connect(conn_string)
  print "Connection to database was established succesfully"
except:
  print "Connection to database failed"

###
### Execute SQL query
###  

# New cursor method for sql
cur = conn.cursor()

# Execute SQL query. For more than one row use three '"'
try:
  cur.execute(""" 

-- In which year/month have these points been created?
-- Need 'yyyymm' because I only need Months with years (values are summeed up). Without, query returns every day the db has an entry.

SELECT to_char(s.day,'yyyymm') AS month
      ,count(t.id)::int AS count
FROM  (
   SELECT generate_series(min(date_from)::date
                         ,max(date_from)::date
                         ,interval '1 day'
          )::date AS day
   FROM   mytable t
   ) s
LEFT   JOIN mytable t ON t.date_from::date = s.day
GROUP  BY month
ORDER  BY month;

  """)

# Return the results of the query. Fetchall() =  all rows, fetchone() = first row
  records = cur.fetchall()
  cur.close()

except:
  print "Query could not be executed"

# Unzip the data from the db-query. Order is the same as db-query output
year, count = zip(*records)

###
### Plot (Barchart)
###

# Count the length of the range of the count-values, y-axis-values, position of axis-labels, legend-label
plt.bar(range(len(count)), count, align='center', label='Amount of created/edited points')

# Add database-values to the plot with an offset of 10px/10px
ax = fig.add_subplot(111)
for i,j in zip(year,count):
    ax.annotate(str(j), xy=(i,j), xytext=(10,10), textcoords='offset points')

# Rotate x-labels on the x-axis
fig.autofmt_xdate()

# Label-values for x and y axis
plt.xticks(range(len(count)), (year))

# Label x and y axis
plt.xlabel('Year')
plt.ylabel('Amount of created/edited points')

# Locate legend on the plot (http://matplotlib.org/users/legend_guide.html#legend-location)
plt.legend(loc=1)

# Plot-title
plt.title("Amount of created/edited points over time")

# show plot
pylab.show()

有什么方法可以防止标签相互重叠？理想情况下是以自动方式，因为我无法预测条形量。

Answer 1

我认为你对matplotlib处理日期的几点感到困惑。

目前你实际上并没有在绘制日期。您使用[0,1,2,...]在x轴上绘制事物，然后使用日期的字符串表示法手动标记每个点。

Matplotlib会自动定位滴答声。但是，你正在超越matplotlib的定位功能（使用xticks基本上是这样说：“我想要在这些位置上打勾”。）

目前，如果matplotlib自动定位，您将获得[10, 20, 30, ...]的刻度。但是，这些将对应于您用于绘制它们的值，而不是日期（绘制时未使用的日期）。

您可能希望使用日期来实际绘制内容。

目前，你正在做这样的事情：

import datetime as dt
import matplotlib.dates as mdates
import numpy as np
import matplotlib.pyplot as plt

# Generate a series of dates (these are in matplotlib's internal date format)
dates = mdates.drange(dt.datetime(2010, 01, 01), dt.datetime(2012,11,01), 
                      dt.timedelta(weeks=3))

# Create some data for the y-axis
counts = np.sin(np.linspace(0, np.pi, dates.size))

# Set up the axes and figure
fig, ax = plt.subplots()

# Make a bar plot, ignoring the date values
ax.bar(np.arange(counts.size), counts, align='center', width=1.0)

# Force matplotlib to place a tick at every bar and label them with the date
datelabels = mdates.num2date(dates) # Go back to a sequence of datetimes...
ax.set(xticks=np.arange(dates.size), xticklabels=datelabels) #Same as plt.xticks

# Make space for and rotate the x-axis tick labels
fig.autofmt_xdate()

plt.show()

相反，尝试这样的事情：

import datetime as dt
import matplotlib.dates as mdates
import numpy as np
import matplotlib.pyplot as plt

# Generate a series of dates (these are in matplotlib's internal date format)
dates = mdates.drange(dt.datetime(2010, 01, 01), dt.datetime(2012,11,01), 
                      dt.timedelta(weeks=3))

# Create some data for the y-axis
counts = np.sin(np.linspace(0, np.pi, dates.size))

# Set up the axes and figure
fig, ax = plt.subplots()

# By default, the bars will have a width of 0.8 (days, in this case) We want
# them quite a bit wider, so we'll make them them the minimum spacing between
# the dates. (To use the exact code below, you'll need to convert your sequence
# of datetimes into matplotlib's float-based date format.  
# Use "dates = mdates.date2num(dates)" to convert them.)
width = np.diff(dates).min()

# Make a bar plot. Note that I'm using "dates" directly instead of plotting
# "counts" against x-values of [0,1,2...]
ax.bar(dates, counts, align='center', width=width)

# Tell matplotlib to interpret the x-axis values as dates
ax.xaxis_date()

# Make space for and rotate the x-axis tick labels
fig.autofmt_xdate()

plt.show()

Answer 2

编辑2014-09-30

pandas现在有一个read_sql功能。你肯定想要使用它。

原始答案

以下是将日期字符串转换为实际日期时间对象的方法：

import numpy as np
import matplotlib.pyplot as plt
import matplotlib.dates as mdates
data_tuples = [
    ('4084036', '1', '2006-12-22 22:46:35'),
    ('4084938', '1', '2006-12-23 16:19:13'),
    ('4084938', '2', '2006-12-23 16:20:23'),
    ('4084939', '1', '2006-12-23 16:29:14'),
    ('4084954', '1', '2006-12-23 16:28:28'),
    ('4250653', '1', '2007-02-12 21:58:53'),
    ('4250657', '1', '2007-03-12 21:58:53')]
datatypes = [('col1', 'i4'), ('col2', 'i4'), ('date', 'S20')]
data = np.array(data_tuples, dtype=datatypes)
col1 = data['col1']
dates = mdates.num2date(mdates.datestr2num(data['date']))
fig, ax1 = plt.subplots()
ax1.bar(dates, col1)
fig.autofmt_xdate()

从数据库游标中获取一个简单的元组列表应该像......一样简单。

data_tuples = []
for row in cursor:
    data_tuples.append(row)

但是，我发布了一个函数版本，用于将db游标直接带到记录数组或pandas数据帧：How to convert SQL Query result to PANDAS Data Structure?

希望这也有帮助。

Answer 3

关于如何在xaxis上仅显示每个第4个刻度（例如）的问题，您可以这样做：

import matplotlib.ticker as mticker

myLocator = mticker.MultipleLocator(4)
ax.xaxis.set_major_locator(myLocator)

Answer 4

bool_and(department in ('Y','Z'))

日期刻度标签经常重叠：

-- employees with at least those two departments
select empid
from employee
where department in name in ('Y','Z')
group by empid
having count(distinct department) = 2

intersect

-- employees with exactly two departments
select empid
from employee
group by empid
having count(distinct department) = 2;

因此旋转它们并右对齐它们很有用。

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
# create a random dataframe with datetimeindex
date_range = pd.date_range('1/1/2011', '4/10/2011', freq='D')
df = pd.DataFrame(np.random.randint(0,10,size=(100, 1)), columns=['value'], index=date_range)