There are 2 issues with your code example:
1. You don't request enough data. Resampling 100 daily prices to monthly gives you only 6 values but you want rolling windows of 20 and 50 which return only NaNs. Request more days in data.history, something like 1200.
2. The symbols object is different in research and IDE. In research it's and equity object but in IDE it's a list of equity objects. In your code

weeklySPY[symbols('SPY')]  

is a series in research but a data frame in IDE. To get the series in IDE use symbol without 's' at the end (this is not available in research).
Your code should work like this:

    pxTLT = np.log(data.history(context.tlt, 'price', 1200, '1d'))  
    weeklyTLT = pxTLT.resample('1M', closed='right', label='right').last().ffill()

    pxSPY = np.log(data.history(context.spy, 'price', 1200, '1d'))  
    weeklySPY = pxSPY.resample('1M', closed='right', label='right').last().ffill()  
    px = (weeklySPY[symbol('SPY')] - weeklyTLT[symbol('TLT')])#.dropna()

    fastMA = px.rolling(window = 20).mean().dropna()  
    slowMA = px.rolling(window = 50).mean().dropna()  
    osc = (fastMA - slowMA)/ slowMA

    signal = np.mean(osc[-3:], axis=0)  
    lag = np.mean(osc[-6: -4], axis=0)

I assume you have defined context.tlt as symbols('TLT'). You should keep it like that if you only make those 2 changes. Alternatively you could shorten it a bit like this:

    tlt = symbol('TLT')  
    spy = symbol('SPY')

    hist = np.log(data.history([tlt, spy], 'price', 1200, '1d'))  
    weekly = hist.resample('1M', closed='right', label='right').last().ffill()  
    px = weekly[spy] - weekly[tlt]

    fastMA = px.rolling(20).mean()  
    slowMA = px.rolling(50).mean()  
    osc = (fastMA - slowMA) / slowMA

    signal = osc[-3:].mean()  
    lag = osc[-6: -4].mean()

Thanks Tentor, that was very clear and helpful!