就地舍入和覆盖 xarray 数据

Question

I want to round all of the values in a xarray to 2 decimal places.

I have tried the following:

def round_dim(data_arr):
   data_arr.data = data_arr.data.round(decimals = 2)
   return data_arr

I call it as such:

data_values = <xarray.core.dataarray.DataArray>
data_values = round_dim(data_values)

I would expect data_values to now how the same values but rounded to 2 decimal places but this does not happen.

EDIT

When I look at the data it is as follows:

def round_dim(data_arr):
   #(data_arr.values) is [-50.406578, -50.415337, -50.42315]
   data_arr.data = data_arr.data.round(decimals = 2)
   #(data_arr.values) is [-50.41, -50.42, -50.42]
   return data_arr

As can be seen after data_arr.data = data_arr.data.round(decimals = 2) I do get the correct rounding.

However, doing:

data_values = <xarray.core.dataarray.DataArray>
data_values = round_dim(data_values)

print(data_values.values)

I do not get the correct rounded values, instead:

[-50.409999, -50.419998, -50.419998]

I would expect the data array to look like [-50.41, -50.42, -50.42]

Answer 1

If you want to apply round(decimals = 2) to your whole dataset, you can use xarray.Dataset.map .

It returns a copy of the original dataset with all its data rounded, it does not work in place.

import numpy as np
import xarray

dataset = xarray.Dataset()
dataset["a"] = xarray.DataArray(np.linspace(0, 1, 10))
dataset["b"] = xarray.DataArray(np.arange(10))
print(dataset)
# <xarray.Dataset>
# ...
#    a        (dim_0) float64 0.0 0.1111 0.2222 0.3333 ... 0.7778 0.8889 1.0
#    b        (dim_0) int32 0 1 2 3 4 5 6 7 8 9

def round_dim(dataset):
    return dataset.map(lambda a: np.round(a, decimals=2))

dataset = round_dim(dataset)
print(dataset)
# <xarray.Dataset>
# ...
#     a        (dim_0) float64 0.0 0.11 0.22 0.33 0.44 0.56 0.67 0.78 0.89 1.0
#     b        (dim_0) int32 0 1 2 3 4 5 6 7 8 9

Answer 2

So the issue was to do with the dtype of the data. Apparently dtype of float32 does not do well with .round .

To fix the issue, I converted the data to dtype float64 using .astype('float64') and the rounding works as expected.