Python数据框：随机排列行

Question

What is the best way to shuffle a group of rows in a data frame? Need this for a shuffled train set of the model.

For example, shuffle every 10 rows as a separate group or have some logical condition to create separate groups and shuffle them as a group.

Answer 1

If you create a new column with the index you're grouping on, you could do something like:

groups = [df.sample(frac=1) for _, df in df.groupby('index_to_group_on')]
return pandas.concat(groups)

If, for example, you want to shuffle every group of 10 rows, you could create this index via:

df['group_of_ten'] = numpy.arange(len(df)/10)

If you're trying to do cross validation, you can look into scikit-learn's train_test_split : http://scikit-learn.org/stable/modules/generated/sklearn.model_selection.train_test_split.html

Answer 2

There may be other ways too, one way may be using shuffle from sklearn . You can slice the n rows that you want to shuffle and append remaining other rows using .append to the result of shuffled rows.

from sklearn.utils import shuffle

# if df is the dataframe to then:
n = 10 # number of rows to shuffle
shuffled_df = shuffle(df[:n]).append(df[n:])

Answer 3

What you can do is - create a column which identifies the group and then group by that column, followed by a random shuffle of each group.

df = pd.DataFrame(np.random.randint(0,100,size=(100, 4)), columns=list('ABCD'))
df['group_id'] = np.arange(df.shape[0]) // 10  # // is integer division in python3, won't work in python2
shuffled_groups = [v.drop(['group_id'], axis=1).sample(frac=1).reset_index(drop=True) for k, v in df.groupby('group_id')]