Data Tables (astropy.table)

Introduction

astropy.table provides functionality for storing and manipulating heterogeneous tables of data in a way that is familiar to numpy users. A few notable capabilities of this package are:

• Initialize a table from a wide variety of input data structures and types.

• Modify a table by adding or removing columns, changing column names, or adding new rows of data.

• Handle tables containing missing values.

• Include table and column metadata as flexible data structures.

• Specify a description, units, and output formatting for columns.

• Interactively scroll through long tables similar to using more.

• Create a new table by selecting rows or columns from a table.

• Perform Table Operations like database joins, concatenation, and binning.

• Maintain a table index for fast retrieval of table items or ranges.

• Manipulate multidimensional and structured array columns.

• Handle non-native (mixin) column types within table.

• Methods for Reading and Writing Table Objects to files.

• Hooks for Subclassing Table and its component classes.

Getting Started

The basic workflow for creating a table, accessing table elements, and modifying the table is shown below. These examples demonstrate a concise case, while the full astropy.table documentation is available from the Using table section.

First create a simple table with columns of data named a, b, c, and d. These columns have integer, float, string, and Quantity values respectively:

>>> from astropy.table import QTable
>>> import astropy.units as u
>>> import numpy as np

>>> a = np.array([1, 4, 5], dtype=np.int32)
>>> b = [2.0, 5.0, 8.5]
>>> c = ['x', 'y', 'z']
>>> d = [10, 20, 30] * u.m / u.s

>>> t = QTable([a, b, c, d],
...            names=('a', 'b', 'c', 'd'),
...            meta={'name': 'first table'})

Comments:

• Column a is a numpy.ndarray with a specified dtype of int32. If the data type is not provided, the default type for integers is int64 on Mac and Linux and int32 on Windows.

• Column b is a list of float values, represented as float64.

• Column c is a list of str values, represented as unicode. See Bytestring Columns for more information.

• Column d is a Quantity array. Since we used QTable, this stores a native Quantity within the table and brings the full power of Units and Quantities (astropy.units) to this column in the table.

Note

If the table data have no units or you prefer to not use Quantity, then you can use the Table class to create tables. The only difference between QTable and Table is the behavior when adding a column that has units. See Quantity and QTable and Columns with Units for details on the differences and use cases.

There are many other ways of Constructing a Table, including from a list of rows (either tuples or dicts), from a numpy structured or 2D array, by adding columns or rows incrementally, or even converting from a SkyCoord or a pandas.DataFrame.

There are a few ways of Accessing a Table. You can get detailed information about the table values and column definitions as follows:

>>> t
<QTable length=3>
  a      b     c      d
                    m / s
int32 float64 str1 float64
----- ------- ---- -------
    1     2.0    x    10.0
    4     5.0    y    20.0
    5     8.5    z    30.0

You can get summary information about the table as follows:

>>> t.info
<QTable length=3>
name  dtype   unit  class
---- ------- ----- --------
   a   int32         Column
   b float64         Column
   c    str1         Column
   d float64 m / s Quantity

From within a Jupyter notebook, the table is displayed as a formatted HTML table (details of how it appears can be changed by altering the astropy.table.conf.default_notebook_table_class item in the Configuration System (astropy.config):

Or you can get a fancier notebook interface with in-browser search, and sort using show_in_notebook():

If you print the table (either from the notebook or in a text console session) then a formatted version appears:

>>> print(t)
 a   b   c    d
            m / s
--- --- --- -----
  1 2.0   x  10.0
  4 5.0   y  20.0
  5 8.5   z  30.0

If you do not like the format of a particular column, you can change it through the ‘info’ property:

>>> t['b'].info.format = '7.3f'
>>> print(t)
 a     b     c    d
                m / s
--- ------- --- -----
  1   2.000   x  10.0
  4   5.000   y  20.0
  5   8.500   z  30.0

For a long table you can scroll up and down through the table one page at time:

>>> t.more()  

You can also display it as an HTML-formatted table in the browser:

>>> t.show_in_browser()  

Or as an interactive (searchable and sortable) javascript table:

>>> t.show_in_browser(jsviewer=True)  

Now examine some high-level information about the table:

>>> t.colnames
['a', 'b', 'c', 'd']
>>> len(t)
3
>>> t.meta
{'name': 'first table'}

Access the data by column or row using familiar numpy structured array syntax:

>>> t['a']       # Column 'a'
<Column name='a' dtype='int32' length=3>
1
4
5

>>> t['a'][1]    # Row 1 of column 'a'
4

>>> t[1]         # Row 1 of the table
<Row index=1>
  a      b     c      d
                    m / s
int32 float64 str1 float64
----- ------- ---- -------
    4   5.000    y    20.0


>>> t[1]['a']    # Column 'a' of row 1
4

You can retrieve a subset of a table by rows (using a slice) or by columns (using column names), where the subset is returned as a new table:

>>> print(t[0:2])      # Table object with rows 0 and 1
 a     b     c    d
                m / s
--- ------- --- -----
  1   2.000   x  10.0
  4   5.000   y  20.0


>>> print(t['a', 'c'])  # Table with cols 'a' and 'c'
 a   c
--- ---
  1   x
  4   y
  5   z

Modifying a Table in place is flexible and works as you would expect:

>>> t['a'][:] = [-1, -2, -3]    # Set all column values in place
>>> t['a'][2] = 30              # Set row 2 of column 'a'
>>> t[1] = (8, 9.0, "W", 4 * u.m / u.s) # Set all values of row 1
>>> t[1]['b'] = -9              # Set column 'b' of row 1
>>> t[0:2]['b'] = 100.0         # Set column 'b' of rows 0 and 1
>>> print(t)
 a     b     c    d
                m / s
--- ------- --- -----
 -1 100.000   x  10.0
  8 100.000   W   4.0
 30   8.500   z  30.0

Replace, add, remove, and rename columns with the following:

>>> t['b'] = ['a', 'new', 'dtype']   # Replace column 'b' (different from in-place)
>>> t['e'] = [1, 2, 3]               # Add column 'e'
>>> del t['c']                       # Delete column 'c'
>>> t.rename_column('a', 'A')        # Rename column 'a' to 'A'
>>> t.colnames
['A', 'b', 'd', 'e']

Adding a new row of data to the table is as follows. Note that the unit value is given in cm / s but will be added to the table as 0.1 m / s in accord with the existing unit.

>>> t.add_row([-8, 'string', 10 * u.cm / u.s, 10])
>>> t['d']
<Quantity [10. ,  4. , 30. ,  0.1] m / s>

Tables can be used for data with missing values:

>>> from astropy.table import MaskedColumn
>>> a_masked = MaskedColumn(a, mask=[True, True, False])
>>> t = QTable([a_masked, b, c], names=('a', 'b', 'c'),
...            dtype=('i4', 'f8', 'U1'))
>>> t
<QTable length=3>
  a      b     c
int32 float64 str1
----- ------- ----
   --     2.0    x
   --     5.0    y
    5     8.5    z

In addition to Quantity, you can include certain object types like Time, SkyCoord, and NdarrayMixin in your table. These “mixin” columns behave like a hybrid of a regular Column and the native object type (see Mixin Columns). For example:

>>> from astropy.time import Time
>>> from astropy.coordinates import SkyCoord
>>> tm = Time(['2000:002', '2002:345'])
>>> sc = SkyCoord([10, 20], [-45, +40], unit='deg')
>>> t = QTable([tm, sc], names=['time', 'skycoord'])
>>> t
<QTable length=2>
         time          skycoord
                       deg,deg
         Time          SkyCoord
--------------------- ----------
2000:002:00:00:00.000 10.0,-45.0
2002:345:00:00:00.000  20.0,40.0

Now let us compute the interval since the launch of the Chandra X-ray Observatory aboard STS-93 and store this in our table as a Quantity in days:

>>> dt = t['time'] - Time('1999-07-23 04:30:59.984')
>>> t['dt_cxo'] = dt.to(u.d)
>>> t['dt_cxo'].info.format = '.3f'
>>> print(t)
         time          skycoord   dt_cxo
                       deg,deg      d
--------------------- ---------- --------
2000:002:00:00:00.000 10.0,-45.0  162.812
2002:345:00:00:00.000  20.0,40.0 1236.812

Using table

The details of using astropy.table are provided in the following sections:

Construct Table

• Constructing a Table
  • Examples
  • Initialization Details
  • Copy versus Reference
  • Column and TableColumns Classes
  • Subclassing Table
  • Columns and Quantities
  • Table-like Objects

Access Table

• Accessing a Table
  • Basics
  • Details

Modify Table

• Modifying a Table
  • Quick Overview
  • Caveats
  • In-Place Versus Replace Column Update

Table Operations

• Table Operations
  • Grouped Operations
  • Binning
  • Stack Vertically
  • Stack Horizontally
  • Stack Depth-Wise
  • Join
  • Merging Details
  • Unique Rows
  • Set Difference
  • Table Diff

Indexing

• Table Indexing
  • Creating an Index
  • Row Retrieval using Indices
  • Effects on Performance
  • Index Modes
  • Updating Rows using Indices
  • Retrieving the Location of Rows using Indices
  • Engines

Masking

• Masking and Missing Values
  • Table Creation
  • Table Access
  • Masking and Filling

I/O with Tables

• Reading and Writing Table Objects
  • Getting Started
  • Supported Formats
• Interfacing with the Pandas Package
  • Example

Mixin Columns

• Mixin Columns
  • Basic Example
  • Quantity and QTable
  • Mixin Attributes
  • Details and Caveats
  • Mixin Protocol
  • Example: ArrayWrapper
  • Registering array-like objects as mixin columns

Implementation

• Table Implementation Details

Performance Tips

Constructing Table objects row by row using add_row() can be very slow:

>>> from astropy.table import Table
>>> t = Table(names=['a', 'b'])
>>> for i in range(100):
...     t.add_row((1, 2))

If you do need to loop in your code to create the rows, a much faster approach is to construct a list of rows and then create the Table object at the very end:

>>> rows = []
>>> for i in range(100):
...     rows.append((1, 2))
>>> t = Table(rows=rows, names=['a', 'b'])

Writing a Table with MaskedColumn to .ecsv using write() can be very slow:

>>> from astropy.table import Table
>>> import numpy as np
>>> x = np.arange(10000, dtype=float)
>>> tm = Table([x], masked=True)
>>> tm.write('tm.ecsv', overwrite=True)

If you want to write .ecsv using write(), then use serialize_method='data_mask'. This uses the non-masked version of data and it is faster:

>>> tm.write('tm.ecsv', overwrite=True, serialize_method='data_mask')

Read FITS with memmap=True

By default read() will read the whole table into memory, which can take a lot of memory and can take a lot of time, depending on the table size and file format. In some cases, it is possible to only read a subset of the table by choosing the option memmap=True.

For FITS binary tables, the data is stored row by row, and it is possible to read only a subset of rows, but reading a full column loads the whole table data into memory:

>>> import numpy as np
>>> from astropy.table import Table
>>> tbl = Table({'a': np.arange(1e7),
...              'b': np.arange(1e7, dtype=float),
...              'c': np.arange(1e7, dtype=float)})
>>> tbl.write('test.fits', overwrite=True)
>>> table = Table.read('test.fits', memmap=True)  # Very fast, doesn't actually load data
>>> table2 = tbl[:100]  # Fast, will read only first 100 rows
>>> print(table2)  # Accessing column data triggers the read
 a    b    c
---- ---- ----
0.0  0.0  0.0
1.0  1.0  1.0
2.0  2.0  2.0
...  ...  ...
98.0 98.0 98.0
99.0 99.0 99.0
Length = 100 rows
>>> col = table['my_column']  # Will load all table into memory

read() does not support memmap=True for the HDF5 and ASCII file formats.

Reference/API

astropy.table Package

Functions

hstack(tables[, join_type, uniq_col_name, ...])	Stack tables along columns (horizontally)
join(left, right[, keys, join_type, ...])	Perform a join of the left table with the right table on specified keys.
represent_mixins_as_columns(tbl[, ...])	Represent input Table tbl using only Column or MaskedColumn objects.
setdiff(table1, table2[, keys])	Take a set difference of table rows.
unique(input_table[, keys, silent, keep])	Returns the unique rows of a table.
vstack(tables[, join_type, metadata_conflicts])	Stack tables vertically (along rows)
dstack(tables[, join_type, metadata_conflicts])	Stack columns within tables depth-wise
join_skycoord(distance[, distance_func])	Helper function to join on SkyCoord columns using distance matching.
join_distance(distance[, kdtree_args, ...])	Helper function to join table columns using distance matching.

Classes

BST(data, row_index[, unique])	A basic binary search tree in pure Python, used as an engine for indexing.
Column([data, name, dtype, shape, length, ...])	Define a data column for use in a Table object.
ColumnGroups(parent_column[, indices, keys])
ColumnInfo([bound])	Container for meta information like name, description, format.
Conf()	Configuration parameters for astropy.table.
JSViewer([use_local_files, display_length])	Provides an interactive HTML export of a Table.
MaskedColumn([data, name, mask, fill_value, ...])	Define a masked data column for use in a Table object.
NdarrayMixin(obj, *args, **kwargs)	Mixin column class to allow storage of arbitrary numpy ndarrays within a Table.
QTable([data, masked, names, dtype, meta, ...])	A class to represent tables of heterogeneous data.
Row(table, index)	A class to represent one row of a Table object.
SCEngine(data, row_index[, unique])	Fast tree-based implementation for indexing, using the sortedcontainers package.
SerializedColumn	Subclass of dict used to serialize mixin columns.
SortedArray(data, row_index[, unique])	Implements a sorted array container using a list of numpy arrays.
StringTruncateWarning	Warning class for when a string column is assigned a value that gets truncated because the base (numpy) string length is too short.
Table([data, masked, names, dtype, meta, ...])	A class to represent tables of heterogeneous data.
TableAttribute([default])	Descriptor to define a custom attribute for a Table subclass.
TableColumns([cols])	OrderedDict subclass for a set of columns.
TableFormatter()
TableGroups(parent_table[, indices, keys])
TableMergeError
TableReplaceWarning	Warning class for cases when a table column is replaced via the Table.__setitem__ syntax e.g.
PprintIncludeExclude([default])	Maintain tuple that controls table column visibility for print output.

Class Inheritance Diagram