rsi function incorrect calculation in train_and_evaluate.ipynb

[max0x7ba]

Describe the current behavior rsi function in train_and_evaluate.ipynb calculation is incorrect by construction:

def rsi(price: 'pd.Series[pd.Float64Dtype]', period: float) -> 'pd.Series[pd.Float64Dtype]':
    r = price.diff()
    upside = np.minimum(r, 0).abs()
    downside = np.maximum(r, 0).abs()
    rs = upside.ewm(alpha=1 / period).mean() / downside.ewm(alpha=1 / period).mean()
    return 100*(1 - (1 + rs) ** -1)

In the code upside is actually the downside and vice versa. abs() applied without discrimination or good reason only masks the error in the calculation.

Describe the expected behavior

One fix is:

    upside = np.maximum(r, 0)
    downside = -np.minimum(r, 0) 

or, simpler:

    upside = r.clip(lower=0)
    downside = -r.clip(upper=0)

[rhamnett]

Fixed