如何根据python中另一个df的值过滤一个df

Question

I have 2x df's with data in them, and the index are dates

dfcarry = {'EUR3m3m': [1.5, 0.6, 1.7, 1.5, -1.2],
           'EUR6m3m': [2.0, 1.2, 1.3, 0.6, -1.7],
         'EUR6m3m3m': [1.3, 1.0, -1.4, 0.5, np.nan]}
dfcarry = pd.DataFrame(dfcarry, index=['26-09-2016','25-09-2016','24-09-2016','23-09-2016'])

and

dfflags = {'EUR3m3m': [1, 0, 1, 1, -1],
           'EUR6m3m': [1, 1, 1, 0, -1],
         'EUR6m3m3m': [1, 1, -1, 0, 0]}
dfflags = pd.DataFrame(dfflags, index=['26-09-2016','25-09-2016','24-09-2016','23-09-2016'])

Now, what I want to do is to limit the abs value of the sum of the numbers to 1, so for any given date, I can not have more than 2 flags in the same direction, ie a 1 cancels out a -1:

if abs(sum(dfflags['26-09-2016'])) > 1:
    then convert one of the flags to zero

then, the integer I want to get rid of is the value where the corresponding abs(carry) number is the least. If we have too many 1's, we get rid of the 1 where the carry number is least. If we have too many -1's in flags, we get rid of a -1 where the carry number is highest (least -ve)

In my carry df, i have some nans (on purpose).

How do I do this?

So to be clear, for the first date, '26-09-2016', the expected output is to keep a 1 at EUR6m3m, because I keep the 1 corresponsding to the highest absolute values of carry ( so I keep carry=2.0, and get rid of carry=1.5 and 1.3).

expected output overall is

dfflags = {'EUR3m3m': [0, 0, 1, 1, 0],
         'EUR6m3m':   [1, 1, 1, 0, -1],
         'EUR6m3m3m': [0, 0, -1, 0, 0]}

Thanks

Answer 1

EDIT: takes into account the ABS sum of flags as coldsteel pointed out

n = 2 #how many flags can be maximum equal to 1
df = abs(dfflags) * dfcarry
for col in df.columns: #for each column, set the flags
    if sum(dfflags[col]) > n:
        n_new = n + 2 * len(dfflags[dfflags[col] == -1])
        threshold = min(df[col].nlargest(n=n_new))
        df[col] = np.where(df[col] >= threshold,1,0)
    else:
        df[col] = abs(dfflags[col])
dfflags = dfflags[df == 1].fillna(0) #apply the 'filter df' to the dfflags, only keeping the top n 1 in each column

Result

_________ EUR3m3m EUR6m3m EUR6m3m3m 26-09-2016 0.0 1.0 0.0 25-09-2016 0.0 0.0 1.0 24-09-2016 1.0 1.0 -1.0 23-09-2016 1.0 0.0 0.0 22-09-2016 0.0 0.0 0.0

Open question: Sort carry by values as given or by absolute value?