BUG: MACD with signalperiod = 1 - data leak

[Egisess]

Hello everyone, I've encountered dangerous behavior of library while calculating MACD with signalperiod = 1:

How to reproduce:

np.random.seed(42)
test_data = np.random.randn(2000) * 100 + 1000 
fastperiod = 2 
slowperiod = 32
signalperiod = 1

macd_1, macdsignal_1, macdhist_1 = ta.MACD(
 real=np.append(test_data[:1500],-999999999), fastperiod=fastperiod, slowperiod=slowperiod, signalperiod=signalperiod
)
macd_2, macdsignal_2, macdhist_2 = ta.MACD(
 real=test_data[:1501], fastperiod=fastperiod, slowperiod=slowperiod, signalperiod=signalperiod
)
print(macdsignal_2[-3] - macdsignal_1[-3]) # Equals 606061258.6885501

As I am checking [-3] index in the array - it shouldn't know about the upcoming value in the end, but the calculated delta shows that TA-Lib takes this value into account.

Also I checked the correct parameters:

np.random.seed(42)
test_data = np.random.randn(2000) * 100 + 1000 
fastperiod = 2 
slowperiod = 32
signalperiod = 2

macd_1, macdsignal_1, macdhist_1 = ta.MACD(
 real=np.append(test_data[:1500],-999999999), fastperiod=fastperiod, slowperiod=slowperiod, signalperiod=signalperiod
)
macd_2, macdsignal_2, macdhist_2 = ta.MACD(
 real=test_data[:1501], fastperiod=fastperiod, slowperiod=slowperiod, signalperiod=signalperiod
)
print(macdsignal_2[-3] - macdsignal_1[-3]) # Equals 0.0

I think it is needed either to forbid signalperiod = 1 or to manually fill the macdhist array with zeros in this edge case.

[mrjbq7]

Yes, that's true.

Duplicate of #335.

I had reported it upstream (TA-Lib/ta-lib#48) but I think it hasn't been addressed yet.

[mario4tier]

This is now fixed in release 0.7.1