ADX Method Divide by Zero Error (ta 0.3.7)

[palisadoes]

I've seen an issue the trs[i] value can equal zero giving this error:

/usr/local/lib/python3.6/dist-packages/ta/trend.py:170: RuntimeWarning: invalid value encountered in double_scalars
  dip[i] = 100 * (dip_mio[i]/trs[i])
/usr/local/lib/python3.6/dist-packages/ta/trend.py:174: RuntimeWarning: invalid value encountered in double_scalars
  din[i] = 100 * (din_mio[i]/trs[i])

I've checked the pandas series sent to adx, and there are no np.nan, or None values in them.

It needs to be handled in some way. The code doesn't crash, it just gives a warning.

ta version 0.3.7

[bizso09]

it's because this line is leaving a 0 at the end

for i in range(1, len(trs)-1):
    trs[i] = trs[i-1] - (trs[i-1]/float(n)) + tr[n+i]

I haven't looked into it more, but the resulting adx has no nan

[DavidAlmasan]

Is there a fix in plan for this? Or is there a workaround that we can use until a fix is in place?

[codeninja]

Bump for this as I'm running into the same problem using the latest master.

[codeninja]

@bukosabino I have fixed the error with this issue. @DavidAlmasan

solution
Trend.py lines ~140 - 180

    trs = np.zeros(len(close) - (n - 1))
    trs[0] = tr.dropna()[0:n].sum()
    tr = tr.reset_index(drop=True)
    for i in range(0, len(trs)-1):
        trs[i+1] = trs[i] - (trs[i]/float(n)) + tr[n+i]

    up = high - high.shift(1)
    dn = low.shift(1) - low
    pos = abs(((up > dn) & (up > 0)) * up)
    neg = abs(((dn > up) & (dn > 0)) * dn)

    dip_mio = np.zeros(len(close) - (n - 1))
    dip_mio[0] = pos.dropna()[0:n].sum()

    pos = pos.reset_index(drop=True)
    for i in range(0, len(dip_mio)-1):
        dip_mio[i+1] = dip_mio[i] - (dip_mio[i]/float(n)) + pos[n+i]

    din_mio = np.zeros(len(close) - (n - 1))
    din_mio[0] = neg.dropna()[0:n].sum()

    neg = neg.reset_index(drop=True)
    for i in range(0, len(din_mio)-1):
        din_mio[i+1] = din_mio[i] - (din_mio[i]/float(n)) + neg[n+i]

The issue is an off by one error by which the trs variable was not having it's final value in the list set (thus remaining 0) and this resulted in a divide by 0 error when it encountered

    for i in range(len(trs)):
        dip[i] = 100 * (dip_mio[i]/trs[i])

    din = np.zeros(len(trs))
    for i in range(len(trs)):
        din[i] = 100 * (din_mio[i]/trs[i])

I will be submitting a PR for this.

[bukosabino]

Hi @codeninja ,

Thank you for your effort. I accepted your PR, but, when I test the results using your implementation, the tests didn't pass :(

Could you take a look again?

Best,
Dario