xarray_fundamentals.py

import pandas
import xarray

ds = xarray.Dataset(
  {'x': ([None], [1,2,3] ),
   'y': ([None], [4,5,6] )},
)

# *is equivalent to*

df = pandas.DataFrame(
  {'x': [1,2,3],
   'y': [4,5,6]})
)

# check:
df.to_xarray() == ds
ds.to_dataframe() == df

### More Usefully: ###

ds = xarray.Dataset(
  {'x': (['space'], [1,2]),
   'y': (['time'], [3,4,5])})

ds.to_dataframe()

            x  y
space time      
0     0     1  3
      1     1  4
      2     1  5
1     0     2  3
      1     2  4
      2     2  5

# Or, working from scratch:

ds = xarray.Dataset(
  {'x': xarray.DataArray([1,2,3], coords=[['a','b','c']], dims=['index']), 
   'y': xarray.DataArray([4,5,6], coords=[['a','b','c']], dims=['index'])}
)

ds = xarray.Dataset(
  {'x': (['index'], [1,2,3]),
   'y': (['index'], [4,5,6])}
)

# *is equivalent to*

df = pandas.DataFrame(
  {'x': [1,2,3],
   'y': [4,5,6]},
  index=pandas.Index(['a','b','c'], name='index')
)



# Examples of operations:

ds['z'] = xarray.DataArray([7,8,9], coords=[['a','b','c']], dims=['index'])

# *is* #

df['z'] = pandas.Series([7,8,9], pandas.Index(['a','b','c'], name='index'))


# Computation:

ds.sum(dim='index')

Dimensions:  ()
Data variables:
    x        int64 6
    y        int64 15

df.sum(axis=0)

x     6
y    15
dtype: int64

# In case of .apply, we go over variables,
# i.e., equivalent to df.apply(lambda col: col, axis=0)

ds.apply(lambda var: var**2)

Dimensions:  (index: 3)
Coordinates:
  * index    (index) <U1 'a' 'b' 'c'
Data variables:
    x        (index) int64 1 4 9
    y        (index) int64 16 25 36
    
# is #

df.apply(lambda col: col**2, axis=0)

       x   y
index       
a      1  16
b      4  25
c      9  36

# Another example:

ds.apply(lambda var: var.sum())

Dimensions:  ()
Data variables:
    x        int64 6
    y        int64 15

df.apply(lambda col: col.sum(), axis=0)

x     6
y    15
dtype: int64

  
  
# However, in xarray, broadcasting and alignment across dimensions happen automatically:

xarray.Dataset(
  {'x': xarray.DataArray([1,2,3], coords=[['a','b','c']], dims=['index']),
   'y': xarray.DataArray([4,5], coords=[['a','b']], dims=['index2'])}).to_dataframe()

              x  y
index index2      
a     a       1  4
      d       1  5
b     a       2  4
      d       2  5
c     a       3  4
      d       3  5


# And, here is some professional useage example:

import pandas
import xarray

# Create some dimensions
x = pandas.np.linspace(-10,10,10)
y = pandas.np.linspace(-20,20,20)
(yy, xx) = pandas.np.meshgrid(y,x)

# Make two different DataArrays with equal dimensions
var1 = xarray.DataArray(pandas.np.random.randn(len(x),len(y)), coords=[x, y], dims=['x','y'])
var2 = xarray.DataArray(-xx**2+yy**2, coords=[x, y], dims=['x','y'])

# Save one DataArray as dataset
ds = var1.to_dataset(name='var1')

# Add second DataArray to existing dataset (ds)
ds['var2'] = var2


                           var1        var2
x          y                               
-10.000000 -20.000000  0.250051  300.000000
           -17.894737 -0.136954  220.221607
           -15.789474 -0.156218  149.307479
           -13.684211 -0.214556   87.257618
           -11.578947 -1.011103   34.072022
...
            13.684211 -0.430594   87.257618
            15.789474  0.792872  149.307479
            17.894737 -0.648614  220.221607
            20.000000 -0.271610  300.000000

[200 rows x 2 columns]