`numpy`

Question

我有两个数据帧，一个是topic_，它是目标数据帧，tw是源数据帧。 topic_是按字矩阵的主题，其中每个单元格存储在特定主题中出现的单词的概率。我使用numpy.zeros将topic_数据框初始化为零。 tw数据框的样本 -

print(tw)
    topic_id                                     word_prob_pair
0          0  [(customer, 0.061703717964), (team, 0.01724444...
1          1  [(team, 0.0260560163563), (customer, 0.0247838...
2          2  [(customer, 0.0171786268847), (footfall, 0.012...
3          3  [(team, 0.0290787264225), (product, 0.01570401...
4          4  [(team, 0.0197917953222), (data, 0.01343226630...
5          5  [(customer, 0.0263740639141), (team, 0.0251677...
6          6  [(customer, 0.0289764173735), (team, 0.0249938...
7          7  [(client, 0.0265082412402), (want, 0.016477447...
8          8  [(customer, 0.0524006965405), (team, 0.0322975...
9          9  [(generic, 0.0373422774996), (product, 0.01834...
10        10  [(customer, 0.0305256248248), (team, 0.0241559...
11        11  [(customer, 0.0198707090364), (ad, 0.018516805...
12        12  [(team, 0.0159852971954), (customer, 0.0124540...
13        13  [(team, 0.033444510469), (store, 0.01961003290...
14        14  [(team, 0.0344793243818), (customer, 0.0210975...
15        15  [(team, 0.026416114692), (customer, 0.02041691...
16        16  [(campaign, 0.0486186973667), (team, 0.0236024...
17        17  [(customer, 0.0208270072145), (branch, 0.01757...
18        18  [(team, 0.0280889397541), (customer, 0.0127932...
19        19  [(team, 0.0297011415217), (customer, 0.0216007...

我的主题_数据框的大小为num_topics（即20）number_of_unique_words（在tw数据框中）

以下是我用来替换topic_数据框

中每个值的代码

for each_topic in range(num_topics):
    a = tw['word_prob_pair'].iloc[each_topic]
    for word, prob in a:
        topic_.set_value(each_topic, word, prob)

有没有更好的方法来完成这项任务？

Answer 1

您可以list comprehension使用DataFrame构造函数，最后将NaN替换为0 fillna：

df = pd.DataFrame({'word_prob_pair':[

[('customer', 0.061703717964), ('team', 0.01724444)],
[('team', 0.0260560163563), ('customer', 0.0247838)],
[('customer', 0.0171786268847), ('footfall', 0.012)],
[('team', 0.0290787264225), ('product', 0.01570401)],
[('team', 0.0197917953222), ('data', 0.01343226630)],
[('customer', 0.0263740639141), ('team', 0.0251677)],
[('customer', 0.0289764173735), ('team', 0.0249938)],
[('client', 0.0265082412402), ('want', 0.016477447)]
] })
print (df)
                                     word_prob_pair
0  [(customer, 0.061703717964), (team, 0.01724444)]
1  [(team, 0.0260560163563), (customer, 0.0247838)]
2  [(customer, 0.0171786268847), (footfall, 0.012)]
3  [(team, 0.0290787264225), (product, 0.01570401)]
4   [(team, 0.0197917953222), (data, 0.0134322663)]
5  [(customer, 0.0263740639141), (team, 0.0251677)]
6  [(customer, 0.0289764173735), (team, 0.0249938)]
7  [(client, 0.0265082412402), (want, 0.016477447)]

df1 = pd.DataFrame([dict(x) for x in df.word_prob_pair])
df1 = df1.fillna(0)
print (df1)
     client  customer      data  footfall   product      team      want
0  0.000000  0.061704  0.000000     0.000  0.000000  0.017244  0.000000
1  0.000000  0.024784  0.000000     0.000  0.000000  0.026056  0.000000
2  0.000000  0.017179  0.000000     0.012  0.000000  0.000000  0.000000
3  0.000000  0.000000  0.000000     0.000  0.015704  0.029079  0.000000
4  0.000000  0.000000  0.013432     0.000  0.000000  0.019792  0.000000
5  0.000000  0.026374  0.000000     0.000  0.000000  0.025168  0.000000
6  0.000000  0.028976  0.000000     0.000  0.000000  0.024994  0.000000
7  0.026508  0.000000  0.000000     0.000  0.000000  0.000000  0.016477

Answer 2

`numpy`

tid1 = df.topic_id.values
lens = [len(i) for i in df.word_prob_pair.values]
tid2 = tid1.repeat(lens)
cat, prob = np.concatenate(df.word_prob_pair.values).T
ucat, inv = np.unique(cat, return_inverse=True)
data = np.zeros((len(tid1), len(ucat)), dtype=float)
data[tid2, inv] = prob
pd.DataFrame(data, tid1, ucat)

时间

有效的方法来替换pandas数据帧中每个单元格的值

2 个答案:

`numpy`