Python 实用宝典

Question 1

Can anyone recommend a Python library that can do interactive graph visualization?

I specifically want something like d3.js but for python and ideally it would be 3D as well.

I have looked at:

NetworkX – it only does Matplotlib plots and those seem to be 2D. I didn’t see any sort of interactiveness, like one that d3.js gives, such as pulling nodes around.
graph-tool – it does only 2D plots and has very slow interactive graphs.

Question 2

You could use d3py a python module that generate xml pages embedding d3.js script. For example :

import d3py
import networkx as nx

import logging
logging.basicConfig(level=logging.DEBUG)

G = nx.Graph()
G.add_edge(1,2)
G.add_edge(1,3)
G.add_edge(3,2)
G.add_edge(3,4)
G.add_edge(4,2)

# use 'with' if you are writing a script and want to serve this up forever
with d3py.NetworkXFigure(G, width=500, height=500) as p:
    p += d3py.ForceLayout()
    p.show()

Question 3

Plotly supports interactive 2D and 3D graphing. Graphs are rendered with D3.js and can be created with a Python API, matplotlib, ggplot for Python, Seaborn, prettyplotlib, and pandas. You can zoom, pan, toggle traces on and off, and see data on the hover. Plots can be embedded in HTML, apps, dashboards, and IPython Notebooks. Below is a temperature graph showing interactivity. See the gallery of IPython Notebooks tutorials for more examples.

The docs provides examples of supported plot types and code snippets.

Specifically to your question, you can also make interactive plots from NetworkX.

For 3D plotting with Python, you can make 3D scatter, line, and surface plots that are similarly interactive. Plots are rendered with WebGL. For example, see a 3D graph of UK Swap rates.

Disclosure: I’m on the Plotly team.

Question 4

Have you looked at vincent? Vincent takes Python data objects and converts them to Vega visualization grammar. Vega is a higher-level visualization tool built on top of D3. As compared to D3py, the vincent repo has been updated more recently. Though the examples are all static D3.

more info:

The graphs can be viewed in Ipython, just add this code

vincent.core.initialize_notebook()

Or output to JSON where you can view the JSON output graph in the Vega online editor (http://trifacta.github.io/vega/editor/) or view them on your Python server locally. More info on viewing can be found in the pypi link above.

Not sure when, but the Pandas package should have D3 integration at some point. http://pandas.pydata.org/developers.html

Bokeh is a Python visualization library that supports interactive visualization. Its primary output backend is HTML5 Canvas and uses client/server model.

examples: http://continuumio.github.io/bokehjs/

Question 5

One recipe that I have used (described here: Co-Director Network Data Files in GEXF and JSON from OpenCorporates Data via Scraperwiki and networkx ) runs as follows:

generate a network representation using networkx
export the network as a JSON file
import that JSON into to d3.js. (networkx can export both the tree and graph/network representations that d3.js can import).

The networkx JSON exporter takes the form:

from networkx.readwrite import json_graph
import json
print json.dumps(json_graph.node_link_data(G))

Alternatively you can export the network as a GEXF XML file and then import this representation into the sigma.js Javascript visualisation library.

from xml.etree.cElementTree import tostring
writer=gf.GEXFWriter(encoding='utf-8',prettyprint=True,version='1.1draft')
writer.add_graph(G)
print tostring(writer.xml)

Question 6

Another option is bokeh which just went to version 0.3.

Question 7

For those who recommended pyd3, it is no longer under active development and points you to vincent. vincent is also no longer under active development and recommends using altair.

So if you want a pythonic d3, use altair.

Question 8

Try https://altair-viz.github.io/ – the successor of d3py and vincent. See also

Question 9

Check out python-nvd3. It is a python wrapper for nvd3. Looks cooler than d3.py and also has more chart options.

Question 10

I would suggest using mpld3 which combines D3js javascript visualizations with matplotlib of python.

The installation and usage is really simple and it has some cool plugins and interactive stuffs.

http://mpld3.github.io/

Question 11

Plotly can do some cool stuffs for you

https://plot.ly/

Produces highly interactive graphs that can be easily embedded withing the HTML pages for your private server or website using its off line API.

Update: I am note sure about its 3D plotting capabilities, for 2D graphs is awesome Thanks

Question 12

You can also choose to serialize your data and then visualize it in D3.js, as done here: Use Python & Pandas to Create a D3 Force Directed Network Diagram (It comes with a jupyter notebook as well!)

Here is the gist. You serialize your graph data in this format:

import json
json_data = {
  "nodes":[
    {"name":"Myriel","group":1},
    {"name":"Napoleon","group":1},
    {"name":"Mlle.Baptistine","group":1},
    {"name":"Mme.Magloire","group":1},
    {"name":"CountessdeLo","group":1},
  ],
  "links":[
    {"source":1,"target":0,"value":1},
    {"source":2,"target":0,"value":8},
    {"source":3,"target":0,"value":10},
    {"source":3,"target":2,"value":6},
    {"source":4,"target":0,"value":1},
    {"source":5,"target":0,"value":1},
  ]
}
filename_out = 'graph_data.json'
json_out = open(filename_out,'w')
json_out.write(json_data)
json_out.close()

Then you load the data in with d3.js:

d3.json("pcap_export.json", drawGraph);

For the routine drawGraph I refer you to the link, however.

Question 13

There is an interesting port of NetworkX to Javascript that might do what you want. See http://felix-kling.de/JSNetworkX/

Question 14

See:

Is there a good interactive 3D graph library out there?

The accepted answer suggests the following program, which apparently has python bindings: http://ubietylab.net/ubigraph/

Edit

I’m not sure about the interactivity of NetworkX, but you can definitely make 3D graphs. There is at least one example in the gallery:

http://networkx.lanl.gov/examples/drawing/edge_colormap.html

And another example in the ‘examples’. This one, however, requires that you have Mayavi.

http://networkx.lanl.gov/examples/3d_drawing/mayavi2_spring.html

Question 15

I’ve got a good example of automatically generating D3.js network diagrams using Python here: http://brandonrose.org/ner2sna

The cool thing is that you end up with auto-generated HTML and JS and can embed the interactive D3 chart in a notebook with an IFrame

Question 16

The library d3graph will build a force-directed d3-graph from within python. You can “break” the network based on the edge weight, and hover over the nodes for more information. Double click on a node will focus on the node and its connected edges.

pip install d3graph

Example:

source = ['node A','node F','node B','node B','node B','node A','node C','node Z']
target = ['node F','node B','node J','node F','node F','node M','node M','node A']
weight = [5.56, 0.5, 0.64, 0.23, 0.9,3.28,0.5,0.45]

# Import library
from d3graph import d3graph, vec2adjmat

# Convert to adjacency matrix
adjmat = vec2adjmat(source, target, weight=weight)
print(adjmat)
# target  node A  node B  node F  node J  node M  node C  node Z
# source                                                        
# node A    0.00     0.0    5.56    0.00    3.28     0.0     0.0
# node B    0.00     0.0    1.13    0.64    0.00     0.0     0.0
# node F    0.00     0.5    0.00    0.00    0.00     0.0     0.0
# node J    0.00     0.0    0.00    0.00    0.00     0.0     0.0
# node M    0.00     0.0    0.00    0.00    0.00     0.0     0.0
# node C    0.00     0.0    0.00    0.00    0.50     0.0     0.0
# node Z    0.45     0.0    0.00    0.00    0.00     0.0     0.0

# Example A: simple interactive network
out = d3graph(adjmat)

# Example B: Color nodes
out = d3graph(adjmat, node_color=adjmat.columns.values)

# Example C: include node size
node_size = [10,20,10,10,15,10,5]
out = d3graph(adjmat, node_color=adjmat.columns.values, node_size=node_size)

# Example D: include node-edge-size
out = d3graph(adjmat, node_color=adjmat.columns.values, node_size=node_size, node_size_edge=node_size[::-1], cmap='Set2')

# Example E: include node-edge color
out = d3graph(adjmat, node_color=adjmat.columns.values, node_size=node_size, node_size_edge=node_size[::-1], node_color_edge='#00FFFF')

# Example F: Change colormap
out = d3graph(adjmat, node_color=adjmat.columns.values, node_size=node_size, node_size_edge=node_size[::-1], node_color_edge='#00FFFF', cmap='Set2')

# Example H: Include directed links. Arrows are set from source -> target
out = d3graph(adjmat, node_color=adjmat.columns.values, node_size=node_size, node_size_edge=node_size[::-1], node_color_edge='#00FFFF', cmap='Set2', directed=True)

Interactive example from the titanic-case can be found here: https://erdogant.github.io/docs/d3graph/titanic_example/index.html https://erdogant.github.io/hnet/pages/html/Use%20Cases.html

Question 17

How can I plot the following 3 functions (i.e. sin, cos and the addition), on the domain t, in the same figure?

from numpy import *
import math
import matplotlib.pyplot as plt

t = linspace(0, 2*math.pi, 400)

a = sin(t)
b = cos(t)
c = a + b

Question 18

To plot multiple graphs on the same figure you will have to do:

from numpy import *
import math
import matplotlib.pyplot as plt

t = linspace(0, 2*math.pi, 400)
a = sin(t)
b = cos(t)
c = a + b

plt.plot(t, a, 'r') # plotting t, a separately 
plt.plot(t, b, 'b') # plotting t, b separately 
plt.plot(t, c, 'g') # plotting t, c separately 
plt.show()

Question 19

Perhaps a more pythonic way of doing so.

from numpy import *
import math
import matplotlib.pyplot as plt

t = linspace(0,2*math.pi,400)
a = sin(t)
b = cos(t)
c = a + b

plt.plot(t, a, t, b, t, c)
plt.show()

Question 20

Just use the function plot as follows

figure()
...
plot(t, a)
plot(t, b)
plot(t, c)

Question 21

How can one neatly represent a graph in Python? (Starting from scratch i.e. no libraries!)
What data structure (e.g. dicts/tuples/dict(tuples)) will be fast but also memory efficient?
One must be able to do various graph operations on it.

As pointed out, the various graph representations might help. How does one go about implementing them in Python?

As for the libraries, this question has quite good answers.

Question 22

Even though this is a somewhat old question, I thought I’d give a practical answer for anyone stumbling across this.

Let’s say you get your input data for your connections as a list of tuples like so:

[('A', 'B'), ('B', 'C'), ('B', 'D'), ('C', 'D'), ('E', 'F'), ('F', 'C')]

The data structure I’ve found to be most useful and efficient for graphs in Python is a dict of sets. This will be the underlying structure for our Graph class. You also have to know if these connections are arcs (directed, connect one way) or edges (undirected, connect both ways). We’ll handle that by adding a directed parameter to the Graph.__init__ method. We’ll also add some other helpful methods.

import pprint
from collections import defaultdict


class Graph(object):
    """ Graph data structure, undirected by default. """

    def __init__(self, connections, directed=False):
        self._graph = defaultdict(set)
        self._directed = directed
        self.add_connections(connections)

    def add_connections(self, connections):
        """ Add connections (list of tuple pairs) to graph """

        for node1, node2 in connections:
            self.add(node1, node2)

    def add(self, node1, node2):
        """ Add connection between node1 and node2 """

        self._graph[node1].add(node2)
        if not self._directed:
            self._graph[node2].add(node1)

    def remove(self, node):
        """ Remove all references to node """

        for n, cxns in self._graph.items():  # python3: items(); python2: iteritems()
            try:
                cxns.remove(node)
            except KeyError:
                pass
        try:
            del self._graph[node]
        except KeyError:
            pass

    def is_connected(self, node1, node2):
        """ Is node1 directly connected to node2 """

        return node1 in self._graph and node2 in self._graph[node1]

    def find_path(self, node1, node2, path=[]):
        """ Find any path between node1 and node2 (may not be shortest) """

        path = path + [node1]
        if node1 == node2:
            return path
        if node1 not in self._graph:
            return None
        for node in self._graph[node1]:
            if node not in path:
                new_path = self.find_path(node, node2, path)
                if new_path:
                    return new_path
        return None

    def __str__(self):
        return '{}({})'.format(self.__class__.__name__, dict(self._graph))

I’ll leave it as an “exercise for the reader” to create a find_shortest_path and other methods.

Let’s see this in action though…

>>> connections = [('A', 'B'), ('B', 'C'), ('B', 'D'),
                   ('C', 'D'), ('E', 'F'), ('F', 'C')]
>>> g = Graph(connections, directed=True)
>>> pretty_print = pprint.PrettyPrinter()
>>> pretty_print.pprint(g._graph)
{'A': {'B'},
 'B': {'D', 'C'},
 'C': {'D'},
 'E': {'F'},
 'F': {'C'}}

>>> g = Graph(connections)  # undirected
>>> pretty_print = pprint.PrettyPrinter()
>>> pretty_print.pprint(g._graph)
{'A': {'B'},
 'B': {'D', 'A', 'C'},
 'C': {'D', 'F', 'B'},
 'D': {'C', 'B'},
 'E': {'F'},
 'F': {'E', 'C'}}

>>> g.add('E', 'D')
>>> pretty_print.pprint(g._graph)
{'A': {'B'},
 'B': {'D', 'A', 'C'},
 'C': {'D', 'F', 'B'},
 'D': {'C', 'E', 'B'},
 'E': {'D', 'F'},
 'F': {'E', 'C'}}

>>> g.remove('A')
>>> pretty_print.pprint(g._graph)
{'B': {'D', 'C'},
 'C': {'D', 'F', 'B'},
 'D': {'C', 'E', 'B'},
 'E': {'D', 'F'},
 'F': {'E', 'C'}}

>>> g.add('G', 'B')
>>> pretty_print.pprint(g._graph)
{'B': {'D', 'G', 'C'},
 'C': {'D', 'F', 'B'},
 'D': {'C', 'E', 'B'},
 'E': {'D', 'F'},
 'F': {'E', 'C'},
 'G': {'B'}}

>>> g.find_path('G', 'E')
['G', 'B', 'D', 'C', 'F', 'E']

Question 23

NetworkX is an awesome Python graph library. You’ll be hard pressed to find something you need that it doesn’t already do.

And it’s open source so you can see how they implemented their algorithms. You can also add additional algorithms.

https://github.com/networkx/networkx/tree/master/networkx/algorithms

Question 24

First, the choice of classical list vs. matrix representations depends on the purpose (on what do you want to do with the representation). The well-known problems and algorithms are related to the choice. The choice of the abstract representation kind of dictates how it should be implemented.

Second, the question is whether the vertices and edges should be expressed only in terms of existence, or whether they carry some extra information.

From Python built-in data types point-of-view, any value contained elsewhere is expressed as a (hidden) reference to the target object. If it is a variable (i.e. named reference), then the name and the reference is always stored in (an internal) dictionary. If you do not need names, then the reference can be stored in your own container — here probably Python list will always be used for the list as abstraction.

Python list is implemented as a dynamic array of references, Python tuple is implemented as static array of references with constant content (the value of references cannot be changed). Because of that they can be easily indexed. This way, the list can be used also for implementation of matrices.

Another way to represent matrices are the arrays implemented by the standard module array — more constrained with respect to the stored type, homogeneous value. The elements store the value directly. (The list stores the references to the value objects instead). This way, it is more memory efficient and also the access to the value is faster.

Sometimes, you may find useful even more restricted representation like bytearray.

Question 25

There are two excellent graph libraries NetworkX and igraph. You can find both library source codes on GitHub. You can always see how the functions are written. But I prefer NetworkX because its easy to understand.
See their codes to know how they make the functions. You will get multiple ideas and then can choose how you want to make a graph using data structures.

Question 26

How do you trace the path of a Breadth-First Search, such that in the following example:

If searching for key 11, return the shortest list connecting 1 to 11.

[1, 4, 7, 11]

Question 27

You should have look at http://en.wikipedia.org/wiki/Breadth-first_search first.

Below is a quick implementation, in which I used a list of list to represent the queue of paths.

# graph is in adjacent list representation
graph = {
        '1': ['2', '3', '4'],
        '2': ['5', '6'],
        '5': ['9', '10'],
        '4': ['7', '8'],
        '7': ['11', '12']
        }

def bfs(graph, start, end):
    # maintain a queue of paths
    queue = []
    # push the first path into the queue
    queue.append([start])
    while queue:
        # get the first path from the queue
        path = queue.pop(0)
        # get the last node from the path
        node = path[-1]
        # path found
        if node == end:
            return path
        # enumerate all adjacent nodes, construct a new path and push it into the queue
        for adjacent in graph.get(node, []):
            new_path = list(path)
            new_path.append(adjacent)
            queue.append(new_path)

print bfs(graph, '1', '11')

Another approach would be maintaining a mapping from each node to its parent, and when inspecting the adjacent node, record its parent. When the search is done, simply backtrace according the parent mapping.

graph = {
        '1': ['2', '3', '4'],
        '2': ['5', '6'],
        '5': ['9', '10'],
        '4': ['7', '8'],
        '7': ['11', '12']
        }

def backtrace(parent, start, end):
    path = [end]
    while path[-1] != start:
        path.append(parent[path[-1]])
    path.reverse()
    return path


def bfs(graph, start, end):
    parent = {}
    queue = []
    queue.append(start)
    while queue:
        node = queue.pop(0)
        if node == end:
            return backtrace(parent, start, end)
        for adjacent in graph.get(node, []):
            if node not in queue :
                parent[adjacent] = node # <<<<< record its parent 
                queue.append(adjacent)

print bfs(graph, '1', '11')

The above codes are based on the assumption that there’s no cycles.

Question 28

I liked qiao’s first answer very much! The only thing missing here is to mark the vertexes as visited.

Why we need to do it?
Lets imagine that there is another node number 13 connected from node 11. Now our goal is to find node 13.
After a little bit of a run the queue will look like this:

[[1, 2, 6], [1, 3, 10], [1, 4, 7], [1, 4, 8], [1, 2, 5, 9], [1, 2, 5, 10]]

Note that there are TWO paths with node number 10 at the end.
Which means that the paths from node number 10 will be checked twice. In this case it doesn’t look so bad because node number 10 doesn’t have any children.. But it could be really bad (even here we will check that node twice for no reason..)
Node number 13 isn’t in those paths so the program won’t return before reaching to the second path with node number 10 at the end..And we will recheck it..

All we are missing is a set to mark the visited nodes and not to check them again..
This is qiao’s code after the modification:

graph = {
    1: [2, 3, 4],
    2: [5, 6],
    3: [10],
    4: [7, 8],
    5: [9, 10],
    7: [11, 12],
    11: [13]
}


def bfs(graph_to_search, start, end):
    queue = [[start]]
    visited = set()

    while queue:
        # Gets the first path in the queue
        path = queue.pop(0)

        # Gets the last node in the path
        vertex = path[-1]

        # Checks if we got to the end
        if vertex == end:
            return path
        # We check if the current node is already in the visited nodes set in order not to recheck it
        elif vertex not in visited:
            # enumerate all adjacent nodes, construct a new path and push it into the queue
            for current_neighbour in graph_to_search.get(vertex, []):
                new_path = list(path)
                new_path.append(current_neighbour)
                queue.append(new_path)

            # Mark the vertex as visited
            visited.add(vertex)


print bfs(graph, 1, 13)

The output of the program will be:

[1, 4, 7, 11, 13]

Without the unneccecery rechecks..

Question 29

Very easy code. You keep appending the path each time you discover a node.

graph = {
         'A': set(['B', 'C']),
         'B': set(['A', 'D', 'E']),
         'C': set(['A', 'F']),
         'D': set(['B']),
         'E': set(['B', 'F']),
         'F': set(['C', 'E'])
         }
def retunShortestPath(graph, start, end):

    queue = [(start,[start])]
    visited = set()

    while queue:
        vertex, path = queue.pop(0)
        visited.add(vertex)
        for node in graph[vertex]:
            if node == end:
                return path + [end]
            else:
                if node not in visited:
                    visited.add(node)
                    queue.append((node, path + [node]))

Question 30

I thought I’d try code this up for fun:

graph = {
        '1': ['2', '3', '4'],
        '2': ['5', '6'],
        '5': ['9', '10'],
        '4': ['7', '8'],
        '7': ['11', '12']
        }

def bfs(graph, forefront, end):
    # assumes no cycles

    next_forefront = [(node, path + ',' + node) for i, path in forefront if i in graph for node in graph[i]]

    for node,path in next_forefront:
        if node==end:
            return path
    else:
        return bfs(graph,next_forefront,end)

print bfs(graph,[('1','1')],'11')

# >>>
# 1, 4, 7, 11

If you want cycles you could add this:

for i, j in for_front: # allow cycles, add this code
    if i in graph:
        del graph[i]

Question 31

I like both @Qiao first answer and @Or’s addition. For a sake of a little less processing I would like to add to Or’s answer.

In @Or’s answer keeping track of visited node is great. We can also allow the program to exit sooner that it currently is. At some point in the for loop the current_neighbour will have to be the end, and once that happens the shortest path is found and program can return.

I would modify the the method as follow, pay close attention to the for loop

graph = {
1: [2, 3, 4],
2: [5, 6],
3: [10],
4: [7, 8],
5: [9, 10],
7: [11, 12],
11: [13]
}


    def bfs(graph_to_search, start, end):
        queue = [[start]]
        visited = set()

    while queue:
        # Gets the first path in the queue
        path = queue.pop(0)

        # Gets the last node in the path
        vertex = path[-1]

        # Checks if we got to the end
        if vertex == end:
            return path
        # We check if the current node is already in the visited nodes set in order not to recheck it
        elif vertex not in visited:
            # enumerate all adjacent nodes, construct a new path and push it into the queue
            for current_neighbour in graph_to_search.get(vertex, []):
                new_path = list(path)
                new_path.append(current_neighbour)
                queue.append(new_path)

                #No need to visit other neighbour. Return at once
                if current_neighbour == end
                    return new_path;

            # Mark the vertex as visited
            visited.add(vertex)


print bfs(graph, 1, 13)

The output and everything else will be the same. However, the code will take less time to process. This is especially useful on larger graphs. I hope this helps someone in the future.

Question 32

I have some nodes coming from a script that I want to map on to a graph. In the below, I want to use Arrow to go from A to D and probably have the edge colored too in (red or something).

This is basically, like a path from A to D when all other nodes are present. you can imagine each nodes as cities and traveling from A to D requires directions (with arrow heads).

This code below builds the graph

import networkx as nx
import numpy as np
import matplotlib.pyplot as plt

G = nx.Graph()
G.add_edges_from(
    [('A', 'B'), ('A', 'C'), ('D', 'B'), ('E', 'C'), ('E', 'F'),
     ('B', 'H'), ('B', 'G'), ('B', 'F'), ('C', 'G')])

val_map = {'A': 1.0,
           'D': 0.5714285714285714,
           'H': 0.0}

values = [val_map.get(node, 0.25) for node in G.nodes()]

nx.draw(G, cmap = plt.get_cmap('jet'), node_color = values)
plt.show()

but I want something like shown in the image.

Arrow heads of the first image and the edges in red color onto the second image.

Question 33

Fully fleshed out example with arrows for only the red edges:

import networkx as nx
import matplotlib.pyplot as plt

G = nx.DiGraph()
G.add_edges_from(
    [('A', 'B'), ('A', 'C'), ('D', 'B'), ('E', 'C'), ('E', 'F'),
     ('B', 'H'), ('B', 'G'), ('B', 'F'), ('C', 'G')])

val_map = {'A': 1.0,
           'D': 0.5714285714285714,
           'H': 0.0}

values = [val_map.get(node, 0.25) for node in G.nodes()]

# Specify the edges you want here
red_edges = [('A', 'C'), ('E', 'C')]
edge_colours = ['black' if not edge in red_edges else 'red'
                for edge in G.edges()]
black_edges = [edge for edge in G.edges() if edge not in red_edges]

# Need to create a layout when doing
# separate calls to draw nodes and edges
pos = nx.spring_layout(G)
nx.draw_networkx_nodes(G, pos, cmap=plt.get_cmap('jet'), 
                       node_color = values, node_size = 500)
nx.draw_networkx_labels(G, pos)
nx.draw_networkx_edges(G, pos, edgelist=red_edges, edge_color='r', arrows=True)
nx.draw_networkx_edges(G, pos, edgelist=black_edges, arrows=False)
plt.show()

Question 34

I only put this in for completeness. I’ve learned plenty from marius and mdml. Here are the edge weights. Sorry about the arrows. Looks like I’m not the only one saying it can’t be helped. I couldn’t render this with ipython notebook I had to go straight from python which was the problem with getting my edge weights in sooner.

import networkx as nx
import numpy as np
import matplotlib.pyplot as plt
import pylab

G = nx.DiGraph()

G.add_edges_from([('A', 'B'),('C','D'),('G','D')], weight=1)
G.add_edges_from([('D','A'),('D','E'),('B','D'),('D','E')], weight=2)
G.add_edges_from([('B','C'),('E','F')], weight=3)
G.add_edges_from([('C','F')], weight=4)


val_map = {'A': 1.0,
                   'D': 0.5714285714285714,
                              'H': 0.0}

values = [val_map.get(node, 0.45) for node in G.nodes()]
edge_labels=dict([((u,v,),d['weight'])
                 for u,v,d in G.edges(data=True)])
red_edges = [('C','D'),('D','A')]
edge_colors = ['black' if not edge in red_edges else 'red' for edge in G.edges()]

pos=nx.spring_layout(G)
nx.draw_networkx_edge_labels(G,pos,edge_labels=edge_labels)
nx.draw(G,pos, node_color = values, node_size=1500,edge_color=edge_colors,edge_cmap=plt.cm.Reds)
pylab.show()

Question 35

Instead of regular nx.draw you may want to use:

nx.draw_networkx(G[, pos, arrows, with_labels])

For example:

nx.draw_networkx(G, arrows=True, **options)

You can add options by initialising that ** variable like this:

options = {
    'node_color': 'blue',
    'node_size': 100,
    'width': 3,
    'arrowstyle': '-|>',
    'arrowsize': 12,
}

Also some functions support the directed=True parameter In this case this state is the default one:

G = nx.DiGraph(directed=True)

The networkx reference is found here.

Question 36

You need to use a directed graph instead of a graph, i.e.

G = nx.DiGraph()

Then, create a list of the edge colors you want to use and pass those to nx.draw (as shown by @Marius).

Putting this all together, I get the image below. Still not quite the other picture you show (I don’t know where your edge weights are coming from), but much closer! If you want more control of how your output graph looks (e.g. get arrowheads that look like arrows), I’d check out NetworkX with Graphviz.

Question 37

import networkx as nx
import matplotlib.pyplot as plt

g = nx.DiGraph()
g.add_nodes_from([1,2,3,4,5])
g.add_edge(1,2)
g.add_edge(4,2)
g.add_edge(3,5)
g.add_edge(2,3)
g.add_edge(5,4)

nx.draw(g,with_labels=True)
plt.draw()
plt.show()

This is just simple how to draw directed graph using python 3.x using networkx. just simple representation and can be modified and colored etc. See the generated graph here.

Note: It’s just a simple representation. Weighted Edges could be added like

g.add_edges_from([(1,2),(2,5)], weight=2)

and hence plotted again.

Question 38

import networkx as nx
import matplotlib.pyplot as plt

G = nx.DiGraph()
G.add_node("A")
G.add_node("B")
G.add_node("C")
G.add_node("D")
G.add_node("E")
G.add_node("F")
G.add_node("G")
G.add_edge("A","B")
G.add_edge("B","C")
G.add_edge("C","E")
G.add_edge("C","F")
G.add_edge("D","E")
G.add_edge("F","G")

print(G.nodes())
print(G.edges())

pos = nx.spring_layout(G)

nx.draw_networkx_nodes(G, pos)
nx.draw_networkx_labels(G, pos)
nx.draw_networkx_edges(G, pos, edge_color='r', arrows = True)

plt.show()

Question 39

I have an array of timestamps in the format (HH:MM:SS.mmmmmm) and another array of floating point numbers, each corresponding to a value in the timestamp array.

Can I plot time on the x axis and the numbers on the y-axis using Matplotlib?

I was trying to, but somehow it was only accepting arrays of floats. How can I get it to plot the time? Do I have to modify the format in any way?

Question 40

You must first convert your timestamps to Python datetime objects (use datetime.strptime). Then use date2num to convert the dates to matplotlib format.

Plot the dates and values using plot_date:

dates = matplotlib.dates.date2num(list_of_datetimes)
matplotlib.pyplot.plot_date(dates, values)

Question 41

You can also plot the timestamp, value pairs using pyplot.plot (after parsing them from their string representation). (Tested with matplotlib versions 1.2.0 and 1.3.1.)

Example:

import datetime
import random
import matplotlib.pyplot as plt

# make up some data
x = [datetime.datetime.now() + datetime.timedelta(hours=i) for i in range(12)]
y = [i+random.gauss(0,1) for i,_ in enumerate(x)]

# plot
plt.plot(x,y)
# beautify the x-labels
plt.gcf().autofmt_xdate()

plt.show()

Resulting image:

Here’s the same as a scatter plot:

import datetime
import random
import matplotlib.pyplot as plt

# make up some data
x = [datetime.datetime.now() + datetime.timedelta(hours=i) for i in range(12)]
y = [i+random.gauss(0,1) for i,_ in enumerate(x)]

# plot
plt.scatter(x,y)
# beautify the x-labels
plt.gcf().autofmt_xdate()

plt.show()

Produces an image similar to this:

Question 42

7 years later and this code has helped me. However, my times still were not showing up correctly.

Using Matplotlib 2.0.0 and I had to add the following bit of code from Editing the date formatting of x-axis tick labels in matplotlib by Paul H.

import matplotlib.dates as mdates
myFmt = mdates.DateFormatter('%d')
ax.xaxis.set_major_formatter(myFmt)

I changed the format to (%H:%M) and the time displayed correctly.

All thanks to the community.

Question 43

I had trouble with this using matplotlib version: 2.0.2. Running the example from above I got a centered stacked set of bubbles.

I “fixed” the problem by adding another line:

plt.plot([],[])

The entire code snippet becomes:

import datetime
import random
import matplotlib.pyplot as plt
import matplotlib.dates as mdates


# make up some data
x = [datetime.datetime.now() + datetime.timedelta(minutes=i) for i in range(12)]
y = [i+random.gauss(0,1) for i,_ in enumerate(x)]

# plot
plt.plot([],[])
plt.scatter(x,y)

# beautify the x-labels
plt.gcf().autofmt_xdate()
myFmt = mdates.DateFormatter('%H:%M')
plt.gca().xaxis.set_major_formatter(myFmt)

plt.show()
plt.close()

This produces an image with the bubbles distributed as desired.

Question 44

I have the following code:

import matplotlib.pyplot as plt

cdict = {
  'red'  :  ( (0.0, 0.25, .25), (0.02, .59, .59), (1., 1., 1.)),
  'green':  ( (0.0, 0.0, 0.0), (0.02, .45, .45), (1., .97, .97)),
  'blue' :  ( (0.0, 1.0, 1.0), (0.02, .75, .75), (1., 0.45, 0.45))
}

cm = m.colors.LinearSegmentedColormap('my_colormap', cdict, 1024)

plt.clf()
plt.pcolor(X, Y, v, cmap=cm)
plt.loglog()
plt.xlabel('X Axis')
plt.ylabel('Y Axis')

plt.colorbar()
plt.show()

So this produces a graph of the values ‘v’ on the axes X vs Y, using the specified colormap. The X and Y axes are perfect, but the colormap spreads between the min and max of v. I would like to force the colormap to range between 0 and 1.

I thought of using:

plt.axis(...)

To set the ranges of the axes, but this only takes arguments for the min and max of X and Y, not the colormap.

Edit:

For clarity, let’s say I have one graph whose values range (0 … 0.3), and another graph whose values (0.2 … 0.8).

In both graphs, I will want the range of the colorbar to be (0 … 1). In both graphs, I want this range of colour to be identical using the full range of cdict above (so 0.25 in both graphs will be the same colour). In the first graph, all colours between 0.3 and 1.0 won’t feature in the graph, but will in the colourbar key at the side. In the other, all colours between 0 and 0.2, and between 0.8 and 1 will not feature in the graph, but will in the colourbar at the side.

Question 45

Using vmin and vmax forces the range for the colors. Here’s an example:

import matplotlib as m
import matplotlib.pyplot as plt
import numpy as np

cdict = {
  'red'  :  ( (0.0, 0.25, .25), (0.02, .59, .59), (1., 1., 1.)),
  'green':  ( (0.0, 0.0, 0.0), (0.02, .45, .45), (1., .97, .97)),
  'blue' :  ( (0.0, 1.0, 1.0), (0.02, .75, .75), (1., 0.45, 0.45))
}

cm = m.colors.LinearSegmentedColormap('my_colormap', cdict, 1024)

x = np.arange(0, 10, .1)
y = np.arange(0, 10, .1)
X, Y = np.meshgrid(x,y)

data = 2*( np.sin(X) + np.sin(3*Y) )

def do_plot(n, f, title):
    #plt.clf()
    plt.subplot(1, 3, n)
    plt.pcolor(X, Y, f(data), cmap=cm, vmin=-4, vmax=4)
    plt.title(title)
    plt.colorbar()

plt.figure()
do_plot(1, lambda x:x, "all")
do_plot(2, lambda x:np.clip(x, -4, 0), "<0")
do_plot(3, lambda x:np.clip(x, 0, 4), ">0")
plt.show()

Question 46

Use the CLIM function (equivalent to CAXIS function in MATLAB):

plt.pcolor(X, Y, v, cmap=cm)
plt.clim(-4,4)  # identical to caxis([-4,4]) in MATLAB
plt.show()

Question 47

Not sure if this is the most elegant solution (this is what I used), but you could scale your data to the range between 0 to 1 and then modify the colorbar:

import matplotlib as mpl
...
ax, _ = mpl.colorbar.make_axes(plt.gca(), shrink=0.5)
cbar = mpl.colorbar.ColorbarBase(ax, cmap=cm,
                       norm=mpl.colors.Normalize(vmin=-0.5, vmax=1.5))
cbar.set_clim(-2.0, 2.0)

With the two different limits you can control the range and legend of the colorbar. In this example only the range between -0.5 to 1.5 is show in the bar, while the colormap covers -2 to 2 (so this could be your data range, which you record before the scaling).

So instead of scaling the colormap you scale your data and fit the colorbar to that.

Question 48

Using figure environment and .set_clim()

Could be easier and safer this alternative if you have multiple plots:

import matplotlib as m
import matplotlib.pyplot as plt
import numpy as np

cdict = {
  'red'  :  ( (0.0, 0.25, .25), (0.02, .59, .59), (1., 1., 1.)),
  'green':  ( (0.0, 0.0, 0.0), (0.02, .45, .45), (1., .97, .97)),
  'blue' :  ( (0.0, 1.0, 1.0), (0.02, .75, .75), (1., 0.45, 0.45))
}

cm = m.colors.LinearSegmentedColormap('my_colormap', cdict, 1024)

x = np.arange(0, 10, .1)
y = np.arange(0, 10, .1)
X, Y = np.meshgrid(x,y)

data = 2*( np.sin(X) + np.sin(3*Y) )
data1 = np.clip(data,0,6)
data2 = np.clip(data,-6,0)
vmin = np.min(np.array([data,data1,data2]))
vmax = np.max(np.array([data,data1,data2]))

fig = plt.figure()
ax = fig.add_subplot(131)
mesh = ax.pcolormesh(data, cmap = cm)
mesh.set_clim(vmin,vmax)
ax1 = fig.add_subplot(132)
mesh1 = ax1.pcolormesh(data1, cmap = cm)
mesh1.set_clim(vmin,vmax)
ax2 = fig.add_subplot(133)
mesh2 = ax2.pcolormesh(data2, cmap = cm)
mesh2.set_clim(vmin,vmax)
# Visualizing colorbar part -start
fig.colorbar(mesh,ax=ax)
fig.colorbar(mesh1,ax=ax1)
fig.colorbar(mesh2,ax=ax2)
fig.tight_layout()
# Visualizing colorbar part -end

plt.show()

A single colorbar

The best alternative is then to use a single color bar for the entire plot. There are different ways to do that, this tutorial is very useful for understanding the best option. I prefer this solution that you can simply copy and paste instead of the previous visualizing colorbar part of the code.

fig.subplots_adjust(bottom=0.1, top=0.9, left=0.1, right=0.8,
                    wspace=0.4, hspace=0.1)
cb_ax = fig.add_axes([0.83, 0.1, 0.02, 0.8])
cbar = fig.colorbar(mesh, cax=cb_ax)

P.S.

I would suggest using pcolormesh instead of pcolor because it is faster (more infos here ).

问题：相当于D3.js的Python

回答 0

回答 1

回答 2

回答 3

回答 4

回答 5

回答 6

回答 7

回答 8

回答 9

回答 10

回答 11

回答 12

回答 13

回答 14

问题：在Matplotlib中，如何在同一图形上绘制多个函数？

回答 0

回答 1

回答 2

问题：在Python中表示图（数据结构）

回答 0

回答 1

回答 2

回答 3

问题：如何在广度优先搜索中跟踪路径？

回答 0

回答 1

回答 2

回答 3

回答 4

问题：如何在python中使用networkx绘制有向图？

回答 0

回答 1

回答 2

回答 3

回答 4

回答 5

问题：使用Matplotlib在Python中绘制时间

回答 0

回答 1

回答 2

回答 3

问题：在matplotlib中设置颜色栏范围

回答 0

回答 1

回答 2

回答 3

使用图形环境和.set_clim（）

单个彩条

聚苯乙烯

Using figure environment and .set_clim()

A single colorbar

P.S.

问题：未定义DISPLAY时，使用matplotlib生成PNG

回答 0

回答 1

回答 2

回答 3

回答 4

回答 5

回答 6

回答 7

回答 8

回答 9

回答 10

回答 11

问题：Python图形库[关闭]

回答 0

回答 1

回答 2

回答 3

回答 4

回答 5

回答 6

回答 7

问题：您如何更改用matplotlib绘制的图形的大小？

回答 0

回答 1

回答 2