频率分布比较 Python答案

【问题标题】：Frequency Distribution Comparison Python频率分布比较 Python
【发布时间】：2015-11-09 12:04:32
【问题描述】：

我正在使用 python 和 nltk 研究一些文本，我想比较不同文本中词性的频率分布。

我可以为一个文本做到这一点：

from nltk import *

X_tagged = pos_tag(word_tokenize(open('/Users/X.txt').read()))

X_fd = FreqDist([tag for word, tag in X_tagged])
X_fd.plot(cumulative=True, title='Part of Speech Distribution in Corpus X')

我尝试添加另一个，但运气不佳。我有条件频率分布示例，用于比较多个文本中三个单词的计数，但我希望这些行代表四个不同的文本，y 轴代表计数，x 轴代表不同的文本词性。如何比较同一图表中的文本 Y 和 Z？

【问题讨论】：

标签： python nltk word-frequency frequency-distribution

【解决方案1】：

这是一个使用 matplotlib 的示例：

from matplotlib import pylab as plt
from nltk import *
import numpy as np

# you may use a tokenizer like nltk.tokenize.word_tokenize()
dist = {}
dist["win"] = FreqDist(tokenizer("first text"))
dist["draw"] =  FreqDist(tokenizer("second text"))
dist["lose"] =  FreqDist(tokenizer("third text"))
dist["mixed"] = FreqDist(tokenizer("fourth text"))

# sorted list of 50 most common terms in one of the texts
# (too many terms would be illegible in the graph)
most_common = [item for item, _ in dist["mixed"].most_common(50)] 

colors = ["green", "blue", "red", "turquoise"]

# loop over the dictionary keys to plot each distribution
for i, label in enumerate(dist):
    frequency = [dist[label][term] for term in most_common]
    color = colors[i]
    plt.plot(frequency, color=color, label=label)
plt.gca().grid(True)
plt.xticks(np.arange(0, len(most_common), 1), most_common, rotation=90)
plt.xlabel("Most common terms")
plt.ylabel("Frequency")
plt.legend(loc="upper right")
plt.show()

【讨论】：

【解决方案2】：

我想通了，如果有人感兴趣的话；您需要获取单独的频率分布并将它们输入到字典中，其中包含所有 FreqDist 共有的键和表示每个 FreqDist 结果的值元组，然后您需要绘制每个 FreqDist 的值并设置键作为 xvalues，按照您拉出它们的相同顺序。

win = FreqDist([tag for word, tag in win]) # 'win', 'draw', 'lose' and 'mixed' are already POS tagged (lists of tuples ('the', 'DT'))

draw = FreqDist([tag for word, tag in draw])

lose = FreqDist([tag for word, tag in lose])

mixed = FreqDist([tag for word, tag in mixed])

POS = [item for item in win] # list of common keys

results = {}
for key in POS:
    results[key] = tuple([win[key], draw[key], lose[key], mixed[key]]) # one key, tuple of values for each FreqDist (in order)

win_counts = [results[item][0] for item in results]

draw_counts = [results[item][1] for item in results]

lose_counts = [results[item][2] for item in results]

mixed_counts = [results[item][3] for item in results]

display = [item for item in results] # over-cautious, same as POS above

plt.plot(win_counts, color='green', label="win") # need to 'import pyplot as plt'
plt.plot(draw_counts, color='blue', label="draw")
plt.plot(lose_counts, color='red', label="lose")
plt.plot(mixed_counts, color='turquoise', label="mixed")
plt.gca().grid(True)
plt.xticks(np.arange(0, len(display), 1), display, rotation=45) # will put keys as x values
plt.xlabel("Parts of Speech")
plt.ylabel("Counts per 10,000 tweets")
plt.suptitle("Part of Speech Distribution across Pre-Win, Pre-Loss and Pre-Draw Corpora")
plt.legend(loc="upper right")
plt.show()

【讨论】：

【解决方案3】：

FreqDist.plot() 方法只是一种方便的方法。

您需要自己编写绘图逻辑（使用matplotlib）以在一个绘图中包含多个频率分布。

FreqDist 的绘图功能的source code 可能是让您入门的神点。 matplotlib 也有很好的tutorial 和初学者指南。

【讨论】：