I have an array of tenors
Tenors = np.array(['10Y', '15Y', '1M', '1Y', '20Y', '2Y', '30Y', '3M', '5Y', '6M', '9M'])
where M stands for month and Y stands for years. The correctly sorted order (ascending) would then be
['1M', '3M', '6M', '9M', '1Y', '2Y', '5Y', '10Y', '15Y', '20Y', '30Y']
How do I achieve that using python with scipy/numpy? As the tenors originate from a pandas dataframe a solution based on pandas would be fine as well.
Approach #1 Here's a NumPy based approach using np.core.defchararray.replace -
repl = np.core.defchararray.replace
out = Tenors[repl(repl(Tenors,'M','00'),'Y','0000').astype(int).argsort()]
Approach #2 If you are working with strings like '18M', we need to do a bit more of work, like so -
def generic_case_vectorized(Tenors):
# Get shorter names for functions
repl = np.core.defchararray.replace
isalph = np.core.defchararray.isalpha
# Get scaling values
TS1 = Tenors.view('S1')
scale = repl(repl(TS1[isalph(TS1)],'Y','12'),'M','1').astype(int)
# Get the numeric values
vals = repl(repl(Tenors,'M',''),'Y','').astype(int)
# Finally scale numeric values and use sorted indices for sorting input arr
return Tenors[(scale*vals).argsort()]
Approach #3 Here's another approach, though a loopy one to again handle generic cases -
def generic_case_loopy(Tenors):
arr = np.array([[i[:-1],i[-1]] for i in Tenors])
return Tenors[(arr[:,0].astype(int)*((arr[:,1]=='Y')*11+1)).argsort()]
Sample run -
In [84]: Tenors
Out[84]:
array(['10Y', '15Y', '1M', '1Y', '20Y', '2Y', '30Y', '3M', '25M', '5Y',
'6M', '18M'],
dtype='|S3')
In [85]: generic_case_vectorized(Tenors)
Out[85]:
array(['1M', '3M', '6M', '1Y', '18M', '2Y', '25M', '5Y', '10Y', '15Y',
'20Y', '30Y'],
dtype='|S3')
In [86]: generic_case_loopy(Tenors)
Out[86]:
array(['1M', '3M', '6M', '1Y', '18M', '2Y', '25M', '5Y', '10Y', '15Y',
'20Y', '30Y'],
dtype='|S3')