如何检查一个 Pandas 时间序列是否存在于另一个长时间序列中？

Question

I have two very long time-series. I have to check if Series B is present(in the given order) in Series A.

Series A: 1,2,3,4,5,6,5,4,3. Series B: 3,4,5.

Result: True, with index where the small series first element found. Here, index:2 (as 3 is present at index 2 in Series A)

Note: The two series are quite big. let's say A contains 50000 elements and B contains 350.

Answer 1

a very slow solution is to convert series to list and check if first list is a subset of the main list in order

def is_series_a_subseries_in_order(main, sub):
    n = len(sub)
    main=main.tolist()
    sub=sub.tolist()
    return any((main[i:i+n] == sub) for i in range(len(main)-n+1))

will return True or False

Answer 2

A naive approach is to check for B(1) in A. In your example B(1) = A(3), so now you have to check if B(2) = A(4) and you continue till the end of your substring... If it's not correct, start with A(4) and continue till the end.

A better way to search for a substring is to apply Knuth-Morris-Pratt's algorithm. I'll let you search for more information about it!

Answer 3

Unluckily the rolling method of pandas does not allow being used as an iterator, even though implementation is planned in #11704 .
Thus we have to implement a rolling window for subset checking on our own.

ser_a = pd.Series(data=[1, 2, 3, 4, 5, 6, 5, 4, 3])
ser_b = pd.Series(data=[3, 4, 5])

slider_df = pd.concat(
    [ser_a.shift(-i)[:ser_b.size] for i in range(ser_a.size - ser_b.size + 1)], 
    axis=1).astype(ser_a.dtype).T
sub_series = (ser_b == slider_df).all(axis=1)
# if you want, you can extract only the indices where a subseries was found:
sub_series_startindex = sub_series.index[sub_series]

What I am doing here:

• [ser_a.shift(-i)[:ser_b.size] for i in range(ser_a.size - ser_b.size + 1)] : Create a "rolling window" by increased shifting of ser_a , limited to the size of the sub series ser_b to check for. Since shifts at the end will yield NaN , these are excluded in the range .
• pd.concat(..., axis=1) : Concatenate shifted Series, so that slider_df contains all shifts in the columns.
• .astype(ser_a.dtype) : is strictly optional. For large Series this may improve performance, for small Series it may degrade performance.
• .T : transpose df, so that sub-series-index are aligned by axis 0.
• sub_series = (ser_b == slider_df).all(axis=1) : Find where ser_b matches sub-series.
• sub_series.index[sub_series] : extract the indices, where a matching sub-series was found.