问题是在Ubuntu重装了一下之后，之前的代码运行开始报错了；
大致分析一下报错信息，是关于matplotlib的；
Traceback (most recent call last):
File "/home/hazyparker/projects/RL-notebook/book_examples/Chapter2-MAB1.py", line 137, in
plot_results([epsilon_greedy_solver], ["EpsilonGreedy"])
File "/home/hazyparker/projects/RL-notebook/book_examples/Chapter2-MAB1.py", line 114, in plot_results
plt.plot(time_list, solver.regrets, label=solver_names[idx])
File "/home/hazyparker/.local/lib/python3.8/site-packages/matplotlib/pyplot.py", line 2728, in plot
return gca().plot(
File "/home/hazyparker/.local/lib/python3.8/site-packages/matplotlib/pyplot.py", line 2225, in gca
return gcf().gca()
File "/home/hazyparker/.local/lib/python3.8/site-packages/matplotlib/pyplot.py", line 830, in gcf
return figure()
File "/home/hazyparker/.local/lib/python3.8/site-packages/matplotlib/_api/deprecation.py", line 454, in wrapper
return func(*args, **kwargs)
File "/home/hazyparker/.local/lib/python3.8/site-packages/matplotlib/pyplot.py", line 771, in figure
manager = new_figure_manager(
File "/home/hazyparker/.local/lib/python3.8/site-packages/matplotlib/pyplot.py", line 346, in new_figure_manager
_warn_if_gui_out_of_main_thread()
File "/home/hazyparker/.local/lib/python3.8/site-packages/matplotlib/pyplot.py", line 336, in _warn_if_gui_out_of_main_thread
if (_get_required_interactive_framework(_get_backend_mod()) and
File "/home/hazyparker/.local/lib/python3.8/site-packages/matplotlib/pyplot.py", line 206, in _get_backend_mod
switch_backend(dict.__getitem__(rcParams, "backend"))
File "/home/hazyparker/.local/lib/python3.8/site-packages/matplotlib/pyplot.py", line 266, in switch_backend
canvas_class = backend_mod.FigureCanvas
AttributeError: module 'backend_interagg' has no attribute 'FigureCanvas'
总结就是，AttributeError: module 'backend_interagg' has no attribute 'FigureCanvas'
这个是个matplotlib的错，有的方法是加上matplotlib.use('TkAgg')，但也没用
直接python 路径+文件.py就正常运行，所以这到底是为啥？
# to realize MAB problem
# use Epsilon Greedy algorithm
# import libs
import numpy as np
import matplotlib.pylab as plt
# class to compose a Bernouli Bandit model, create bandit
class BernoulliBandit:
# init function
def __init__(self, K):
# K numbers to stand the possibility for each trail(K times)
self.probs = np.random.uniform(size = K)
# get the id number of the max possibility
self.best_id = np.argmax(self.probs)
# get the max possibility
self.max_prob = self.probs[self.best_id]
# get param K to the class
self.K = K
# function for each trail's final
def step(self, k):
if np.random.rand() < self.probs[k]:
# cautions, this should be less equal than probs[k]
return 1 # trail is a success
else:
return 0 # trail is a failure
# class to solver the Bernouli Bandit problem
# created bandit solver
class SolverBandit:
# init function
def __init__(self, bandit):
# get bandit instance from BernouliBandit
self.bandit = bandit
# set conut for each bandit's trial
self.counts = np.zeros(self.bandit.K)
# set regret for current step
self.regret = 0
# set a list to load regrets
self.regrets = []
# set a list to load actions
self.actions = []
# calculate cumulatice regret
def update_regret(self, k):
# update regret by the formula R(a) = Q* - Q(a)
self.regret = self.regret + self.bandit.max_prob - self.bandit.probs[k]
self.regrets.append(self.regret) # add it to the regrets list
# define for override
def run_one_step(self):
# In user defined base classes, abstract methods should raise this
# exception when they require derived classes to override the method,
# or while the class is being developed to indicate that the real
# implementation still needs to be added.
raise NotImplementedError
# main iteration process
def run(self, num_steps):
# num_steps means the iteration times
for _ in range (num_steps):
k = self.run_one_step()
self.counts[k] += 1
self.actions.append(k)
self.update_regret(k)
# define class for greedy algrothim, inherit from class SolverBandit
class EpsilonGreedy(SolverBandit):
# init function
def __init__(self, bandit, epsilon = 0.01, init_prob = 1.0):
# inherit from it's parent class
# there a good example telling about super()
# https://www.runoob.com/python/python-func-super.html
super().__init__(bandit)
self.epsilon = epsilon
# setting all options's prob as value of init_prob
self.estimates = np.array([init_prob] * self.bandit.K)
# epsilon greedy algorithm's main process
# be advised here, function with the same name of function in
# class SolverBandit
def run_one_step(self):
# the algorithm formula
if np.random.random() < self.epsilon:
# choose one number randomly form 0 to K
k = np.random.randint(0, self.bandit.K)
else:
# choose the max prob from estimated probs
k = np.argmax(self.estimates)
# update the estimates array
r = self.bandit.step(k) # acquire the final
self.estimates[k] += 1.0 / (self.counts[k] + 1) * (
r - self.estimates[k]) # use 1.0 to keep float
# why return k?
return k
# define plot function
def plot_results(solvers, solver_names):
# the use of enumerate can be referred here:
# https://www.geeksforgeeks.org/enumerate-in-python/
# this is set for different epsilons' situtions
for idx, solver in enumerate(solvers):
time_list = range(len(solver.regrets))
plt.plot(time_list, solver.regrets, label=solver_names[idx])
plt.xlabel('Time steps')
plt.ylabel('Cumulative regrets')
plt.title('%d-armed bandit' % solvers[0].bandit.K)
plt.legend()
plt.show()
np.random.seed(1) # make random numbers be the same
# set 10 arms for Bernouli Bandit
K = 10
bandit_10_arm = BernoulliBandit(K)
print("create %d arms Bernouli Bandit" % K)
print("the max probility is No. %d, with %.4f" %
(bandit_10_arm.best_id, bandit_10_arm.max_prob))
np.random.seed(1) # make random numbers be the same
# create greedy solver instance
epsilon_greedy_solver = EpsilonGreedy(bandit_10_arm, epsilon=0.01)
epsilon_greedy_solver.run(5000)
print('epsilon-greedy cumulative regret：',
epsilon_greedy_solver.regret)
plot_results([epsilon_greedy_solver], ["EpsilonGreedy"])
这个好说，先把代码拿到Windows环境，运行一哈，看看有什么问题，如果可运行，说明你的Ubuntu环境给整劈叉了，原来怎么配的在来一次就可以了。然后，pycharm安装一个，然后用pycharm打断点，仔细查看代码逻辑这些的，看是从哪开始报错的就行了啊。具体如何打断点，百度即可，没什么难度的。
两个pyuthon解释器不一样，导致环境不一样，所以会有差异