issue_comments

Minor comment / nit: you don't really need the astype(np.float), the result should already be of dtype float since there are missing values after the rolling sum.

You're right. That is used for debuging the intermediate results.

Thanks again @keewis @max-sixty.

sounds good. Minor comment / nit: you don't really need the astype(np.float), the result should already be of dtype float since there are missing values after the rolling sum.

I don't know about performance but if it does what you asked for then that's good?

Hi @keewis, this is really a smart way, using rolling twice. I've refactored the code slightly as: ```python def continuous_meet(cond, count, dim): """ Continuously meet a given condition along a dimension. """ _found = cond.rolling(dim={dim:count}, center=True).sum().fillna(0).astype(np.float)

detected = (
    _found.rolling(dim={dim:count}, center=True) 
    .reduce(lambda a, axis: (a == count).any(axis=axis)) 
    .fillna(False) 
    .astype(bool) 
)

if count % 2 == 0:
    return detected.shift({dim:-1}).fillna(False)

return detected

sst = xr.DataArray( np.array( [0., 0., 0., 0., 0., 1., 1., 1., 1., 1., 0., 0., 1., 0., 0., 1., 1., 1., 1., 1., 1., 0., 0., 0.] ), dims="time", coords={"time": np.arange(24)}, name="sst", )

ElNino = continuous_meet(sst > 0.5, count=5, dim='time')

sst.plot.step(linewidth=3) sst.where(ElNino).plot.step(linewidth=2) `` Note that whencountis a even number, truly centeredrollingcannot be obtained. So we need toshift` the result by -1. Is this perfect? I didn't check the performance.

what about this? python In [86]: def detect(arr): ...: thresh = arr > 0.5 ...: n_found = thresh.rolling(dim={"x": 5}, center=True).sum().fillna(0) ...: detected = ( ...: n_found.rolling(dim={"x": 5}, center=True) ...: .reduce(lambda a, axis: (a == 5).any(axis=axis)) ...: .fillna(False) ...: .astype(bool) ...: ) ...: return detected ...: ...: arr = xr.DataArray( ...: np.array( ...: [0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 1, 0, 0, 1, 1, 1, 1, 1, 1, 0, 0, 0] ...: ), ...: dims="x", ...: coords={"x": np.arange(24)}, ...: name="sst", ...: ) ...: ( ...: arr.to_dataset() ...: .assign(mask=lambda ds: detect(ds.sst)) ...: .assign(selected=lambda ds: ds.sst.where(ds.mask)) ...: ) Out[86]: <xarray.Dataset> Dimensions: (x: 24) Coordinates: * x (x) int64 0 1 2 3 4 5 6 7 8 9 10 ... 14 15 16 17 18 19 20 21 22 23 Data variables: sst (x) int64 0 0 0 0 0 1 1 1 1 1 0 0 1 0 0 1 1 1 1 1 1 0 0 0 mask (x) bool False False False False False ... True False False False selected (x) float64 nan nan nan nan nan 1.0 ... 1.0 1.0 1.0 nan nan nan

OK, that's a bit harder but not impossible. You could take that, run array.astype(float).where(array).bfill(dim, limit=5), or similar? Others will probably have cleaner suggestions...

Hi @max-sixty, thanks for your kind help. But I found it works not as I expected. If the SST has the values [..., 0, 0, 1, 1, 1, 1, 1, 0, 0, ...], then the method you suggested will give [..., F, F, F, F, T, F, F, F, F, ...]. That is only one True for a 5 consecutive 1s. But I would expect five True like [..., F, F, T, T, T, T, T, F, F, ...]. Any suggestion?

IIUC, you could have, in order: - rolling 3 month SST - bool on whether that's above 0.5 - rolling 5 month count of that - bool on whether that's equal to 5