What happened?

I have lots (61) pickled pandas dataframes that I'm trying to convert from pickle/pandas to zarr/xarray. Since the dataframes are large (10000x2048) I can't load them all into memory. To get around this I'm (MCVE below) looping through the pickle files, reading them into dataframes, constructing DataArrays and then Datasets from the data, concatinating the dataset with the previous dataset and updating the dataset to point to this new concatenated dataset.

Since I didn't want to use up too much memory, I'm also periodically writing the Dataset to .zarr in the loop and reopening it (hoping to make use of dask storing data on disk?).

When I do this however, the final dataset ends up being all NaNs.

What did you expect to happen?

I expected the final dataset to contain all the concatenated data.

Minimal Complete Verifiable Example

```Python import pandas as pd import numpy as np import glob import xarray as xr from tqdm import tqdm

Creating pkl files

[pd.DataFrame(np.random.randint(0,10, (1000,500))).astype(object).to_pickle('df{}.pkl'.format(i)) for i in range(4)]

fnames = glob.glob('*.pkl')

df = pd.read_pickle(fnames[0]) df.columns = np.arange(0,500).astype(object) # the real pkl files contain all objects df.index = np.arange(0,1000).astype(object) df = df.astype(np.float32)

ds = xr.DataArray(df.values, dims=['fname', 'res_dim'], coords={'fname': df.index.values, 'res_dim': df.columns.values}) ds = ds.to_dataset(name='low_dim')

for idx, fname in enumerate(tqdm(fnames[1:])): df = pd.read_pickle(fname) df.columns = np.arange(0,500).astype(object) df.index = np.arange(0,1000).astype(object) df = df.astype(np.float32)

ds2 = xr.DataArray(df.values, dims=['fname', 'res_dim'],
              coords={'fname': df.index.values, 'res_dim': df.columns.values})
ds2 = ds2.to_dataset(name='low_dim')

ds = xr.concat([ds, ds2], dim='fname')
ds['fname'] = ds.fname.astype(str)
if (idx%2 == 0) & (idx !=0):
    ds.to_zarr('zarr_bug.zarr', mode='w')
    ds = xr.open_zarr('zarr_bug.zarr')

ds.to_zarr('zarr_bug.zarr', mode='w') ds = xr.open_zarr('zarr_bug.zarr')

print(ds.low_dim.values) ```

Relevant log output

Python [[nan nan nan ... nan nan nan] [nan nan nan ... nan nan nan] [nan nan nan ... nan nan nan] ... [nan nan nan ... nan nan nan] [nan nan nan ... nan nan nan] [nan nan nan ... nan nan nan]]

Anything else we need to know?

If I get rid of the loop saving, everything works normally.

Environment

INSTALLED VERSIONS

commit: None python: 3.9.11 (main, Mar 28 2022, 10:10:35) [GCC 7.5.0] python-bits: 64 OS: Linux OS-release: 5.11.0-27-generic machine: x86_64 processor: x86_64 byteorder: little LC_ALL: None LANG: None LOCALE: ('en_US', 'UTF-8') libhdf5: 1.12.0 libnetcdf: 4.7.4

xarray: 2022.3.0 pandas: 1.4.1 numpy: 1.21.0 scipy: 1.8.0 netCDF4: 1.5.8 pydap: installed h5netcdf: 1.0.0 h5py: 3.6.0 Nio: None zarr: 2.11.1 cftime: 1.6.0 nc_time_axis: None PseudoNetCDF: None rasterio: 1.2.10 cfgrib: 0.9.10.1 iris: None bottleneck: None dask: 2022.03.0 distributed: 2022.3.0 matplotlib: 3.5.1 cartopy: None seaborn: 0.11.2 numbagg: None fsspec: 2022.02.0 cupy: None pint: None sparse: None setuptools: 58.0.4 pip: 21.2.4 conda: None pytest: None IPython: 8.1.1 sphinx: None

What is your issue?

xr.where() and da.where() behave in seemingly opposite ways.

Example:

python da = xr.DataArray(np.arange(10) print(xr.where(da < 5, 0, da).values) print(da.where(da < 5, 0).values) [0 0 0 0 0 5 6 7 8 9] [0 1 2 3 4 0 0 0 0 0]

It seems like these two methods with the same name should have the same functionality, but they give inverse results.