nbodykit.io.csv

Functions

make_partitions(filename, blocksize, config) Partition a CSV file into blocks, using the preferred blocksize in bytes, returning the partititions and number of rows in each partition
verify_data(path, names[, nrows]) Verify the data by reading the first few lines of the specified CSV file to determine the data type

Classes

CSVFile(path, names[, blocksize, dtype, …]) A file object to handle the reading of columns of data from a CSV file.
CSVPartition(filename, offset, blocksize, …) A simple class to convert byte strings of data from a CSV file to a pandas DataFrame on demand
class nbodykit.io.csv.CSVFile(path, names, blocksize=33554432, dtype={}, usecols=None, delim_whitespace=True, **config)[source]

A file object to handle the reading of columns of data from a CSV file.

Internally, this class partitions the CSV file into chunks, and data is only read from the relevant chunks of the file, using pandas.read_csv().

This setup provides a significant speed-up when reading from the end of the file, since the entirety of the data does not need to be read first.

The class supports any of the configuration keywords that can be passed to pandas.read_csv()

Warning

This assumes the delimiter for separate lines is the newline character and that all columns in the file represent data columns (no “index” column when using pandas)

Parameters:
  • path (str) – the name of the file to load
  • names (list of str) – the names of the columns of the csv file; this should give names of all the columns in the file – pass usecols to select a subset of columns
  • blocksize (int, optional) – the file will be partitioned into blocks of bytes roughly of this size
  • dtype (dict, str, optional) – if specified as a string, assume all columns have this dtype, otherwise; each column can have a dtype entry in the dict; if not specified, the data types will be inferred from the file
  • usecols (list, optional) – a pandas.read_csv; a subset of names to store, ignoring all other columns
  • delim_whitespace (bool, optional) – a pandas.read_csv keyword; if the CSV file is space-separated, set this to True
  • **config – additional keyword arguments that will be passed to pandas.read_csv(); see the documentation of that function for a full list of possible options
Attributes:
columns

A list of the names of the columns in the file.

dtype

A numpy.dtype object holding the data types of each column in the file.

ncol

The number of data columns in the file.

shape

The shape of the file, which defaults to (size, )

size

The size of the file, i.e., number of rows

Methods

asarray() Return a view of the file, where the fields of the structured array are stacked in columns of a single numpy array
get_dask(column[, blocksize]) Return the specified column as a dask array, which delays the explicit reading of the data until dask.compute() is called
keys() Aliased function to return columns
read(columns, start, stop[, step]) Read the specified column(s) over the given range
__getitem__(s)

This function provides numpy-like array indexing of the file object.

It supports:

  1. integer, slice-indexing similar to arrays
  2. string indexing using column names in keys()
  3. array-like indexing using integer lists or boolean arrays

Note

If a single column is being returned, a numpy array holding the data is returned, rather than a structured array with only a single field.

asarray()

Return a view of the file, where the fields of the structured array are stacked in columns of a single numpy array

Examples

Start with a file object with three named columns, ra, dec, and z

>>> ff.dtype
dtype([('ra', '<f4'), ('dec', '<f4'), ('z', '<f4')])
>>> ff.shape
(1000,)
>>> ff.columns
['ra', 'dec', 'z']
>>> ff[:3]
array([(235.63442993164062, 59.39099884033203, 0.6225500106811523),
       (140.36181640625, -1.162310004234314, 0.5026500225067139),
       (129.96627807617188, 45.970130920410156, 0.4990200102329254)],
      dtype=(numpy.record, [('ra', '<f4'), ('dec', '<f4'), ('z', '<f4')]))

Select a subset of columns and switch the ordering and convert output to a single numpy array

>>> x = ff[['dec', 'ra']].asarray()
>>> x.dtype
dtype('float32')
>>> x.shape
(1000, 2)
>>> x.columns
['dec', 'ra']
>>> x[:3]
array([[  59.39099884,  235.63442993],
       [  -1.16231   ,  140.36181641],
       [  45.97013092,  129.96627808]], dtype=float32)

Now, select only the first column (dec)

>>> dec = x[:,0]
>>> dec[:3]
array([ 59.39099884,  -1.16231   ,  45.97013092], dtype=float32)
Returns:a file object that will return a numpy array with the columns representing the fields
Return type:FileType
columns

A list of the names of the columns in the file.

This defaults to the named fields in the file’s dtype attribute, but differ from this if a view of the file has been returned with asarray()

dtype

A numpy.dtype object holding the data types of each column in the file.

get_dask(column, blocksize=None)

Return the specified column as a dask array, which delays the explicit reading of the data until dask.compute() is called

The dask array is chunked into blocks of size blocksize

Parameters:
  • column (str) – the name of the column to return
  • blocksize (int, optional) – the size of the chunks in the dask array
Returns:

the dask array holding the column, which computes the necessary functions to read the data, but delays evaluating until the user specifies

Return type:

dask.array.Array

keys()

Aliased function to return columns

ncol

The number of data columns in the file.

read(columns, start, stop, step=1)[source]

Read the specified column(s) over the given range

‘start’ and ‘stop’ should be between 0 and size, which is the total size of the file (in particles)

Parameters:
  • columns (str, list of str) – the name of the column(s) to return
  • start (int) – the row integer to start reading at
  • stop (int) – the row integer to stop reading at
  • step (int, optional) – the step size to use when reading; default is 1
Returns:

structured array holding the requested columns over the specified range of rows

Return type:

numpy.array

shape

The shape of the file, which defaults to (size, )

Multiple dimensions can be introduced into the shape if a view of the file has been returned with asarray()

size

The size of the file, i.e., number of rows

class nbodykit.io.csv.CSVPartition(filename, offset, blocksize, delimiter, **config)[source]

A simple class to convert byte strings of data from a CSV file to a pandas DataFrame on demand

The DataFrame is cached as value, so only a single call to pandas.read_csv() is used

Attributes:
value

Return the parsed btye string as a DataFrame

__init__(filename, offset, blocksize, delimiter, **config)[source]
Parameters:
  • filename (str) – the file to read data from
  • offset (int) – the offset in bytes to start reading at
  • blocksize (int) – the size of the bytes block to read
  • delimiter (byte str) – how to distinguish separate lines
  • **config – the configuration keywords passed to pandas.read_csv()
value

Return the parsed btye string as a DataFrame

nbodykit.io.csv.make_partitions(filename, blocksize, config, delimiter='\n')[source]

Partition a CSV file into blocks, using the preferred blocksize in bytes, returning the partititions and number of rows in each partition

This divides the input file into partitions with size roughly equal to blocksize, reads the bytes, and counts the number of delimiters to compute the size of each block

Parameters:
  • filename (str) – the name of the CSV file to load
  • blocksize (int) – the desired number of bytes per block
  • delimiter (str, optional) – the character separating lines; default is the newline character
  • config (dict) – any keyword options to pass to pandas.read_csv()
Returns:

  • partitions (list of CSVPartition) – list of objects storing the data content of each file partition, stored as a bytestring
  • sizes (list of int) – the list of the number of rows in each partition

nbodykit.io.csv.verify_data(path, names, nrows=10, **config)[source]

Verify the data by reading the first few lines of the specified CSV file to determine the data type

Parameters:
  • path (str) – the name of the CSV file to load
  • names (list of str) – the list of the names of the columns in the CSV file
  • nrows (int, optional) – the number of rows to read from the file in order to infer the data type; default is 10
  • **config (key, value pairs) – additional keywords to pass to pandas.read_csv()
Returns:

dtype – dictionary holding the dtype for each name in names

Return type:

dict