ctapipe.io contains functions and classes related to reading, writing, and
in-memory storage of event data
Reading Event Data¶
This module provides a set of event sources that are python
generators that loop through an input file or stream and fill in
Container classes, defined below. They are designed such that
ctapipe can be independent of the file format used for event data, and new
formats may be supported by simply adding a plug-in.
The underlying mechanism is a set of
EventSource sub-classes that
read data in various formats, with a common interface and automatic command-line
configuration parameters. These are generally constructed in a generic way by
EventSource(file_or_url) which will construct the
EventSource subclass based on the input file’s type.
EventSource then works like a python collection and can be
looped over, providing data for each subsequent event. If looped over
multiple times, each will start at the beginning of the file (except in
the case of streams that cannot be restarted):
with EventSource(input_url="file.simtel.gz") as source: for event in source: do_something_with_event(event)
If you need random access to events rather than looping over all events in
order, you can use the
EventSeeker class to allow random access by event
index or event_id. This may not be efficient for some EventSources if
the underlying file type does not support random access.
Creating a New EventSource Plugin¶
An example can be found in:
Event data that is intended to be read or written from files is stored
in subclasses of
Container, the structre of which is
defined in the
containers module (See reference API below). Each
element in the container is a
Field, containing the
default value, a description, and default unit if necessary. The
following rules should be followed when creating a
for new data:
Containers both provide a way to exchange data (in-memory) between parts of a code, as well as define the schema for any file output files to be written.
All items in a Container should be expected to be updated at the same frequency. Think of a Container as the column definitions of a table, therefore representing a single row in a table. For example, if the container has event-by-event info, it should not have an item in it that does not change between events (that should be in another container), otherwise it will be written out for each event and will waste space.
a Container should be filled in all at once, not at different times during the data processing (to allow for parallelization and to avoid difficulty in reading code).
Containers may contain a dictionary of metadata (in their
metadictionary), that will become headers in any output file (this data must not change per-event, etc)
Algorithms should not update values in a container that have already been filled in by another algorithm. Instead, prefer a new data item, or a second copy of the Container with the updated values.
Fields in a container should be one of the following:
numpy.NDarrayif the data are not scalar (use only simple dtypes that can be written to output files)
Containerclass (in the case a hierarchy is needed)
Fields that should not be in a container class:
Serialization of Containers:¶
TableReader base classes provide
an interface to implement subclasses that write/read Containers to/from
table-like data files. Currently the only implementation is for writing
HDF5 tables via the
HDF5TableWriter. The output that the
HDF5TableWriter produces can be read either one-row-at-a-time
HDF5TableReader, or more generically using the
pandas packages (note however any tables that have
array values in a column cannot be read into a
pandas.DataFrame, since it
only supports scalar values).
Writing Output Files:¶
DataWriter Component allows one to write a series of events (stored in
ctapipe.containers.ArrayEventContainer) to a standardized HDF5 format file
following the data model (see Data Models). This includes all related datasets
such as the instrument and simulation configuration information, simulated
shower and image information, observed images and parameters and reconstruction
information. It can be used in an event loop like:
with DataWriter(event_source=source, output_path="events.dl1.h5") as write_data: for event in source: calibrate(event) write_data(event)
Reading Output Tables:¶
In addition to using an
EventSource to read R0-DL1 data files, one can also access full tables for files that are in HDF5 format (e.g. DL1 files).
read_table function will load any table in an HDF5 table into an
astropy.table.QTable in memory,
while maintaining units, column descriptions, and other ctapipe metadata.
Astropy Tables can also be converted to Pandas tables via their
as long as the table does not contain any vector columns.
from ctapipe.io import read_table mctable = read_table("events.dl1.h5", "/simulation/event/subarray/shower") mctable['logE'] = np.log10(mc_table['energy']) mctable.write("output.fits")
Standard Metadata Headers¶
ctapipe.io.metadata package provides functions for generating standard CTA
metadata headers and attaching them to output files.
A very basic table writer that can take a container (or more than one) and write it to an HDF5 file.
Reader that reads a single row of an HDF5 table at once into a Container.
Base class for writing tabular data stored in
Base class for row-wise table readers.
Provides the functionality to seek through a
Parent class for EventSources.
Read events from a SimTelArray data file (in EventIO format).
Event source for files in the ctapipe DL1 format.
Enum of the different Data Levels
Serialize a sequence of events into a HDF5 DL1 file, in the correct format
Base functionality for reading and writing tabular data
Implementations of TableWriter and -Reader for HDF5 files
Management of CTA Reference Metadata, as defined in the CTA Top-Level Data Model document [ctatopleveldatamodel] , version 1A. This information is required to be attached to the header of any files generated.
The class Reference collects all required reference metadata, and can be turned into a
flat dictionary. The user should try to fill out all fields, or use a helper to fill
them (as in
ref = Reference( contact=Contact(name="Some User", email="email@example.com"), product=Product(format='hdf5', ...), process=Process(...), activity=Activity(...), instrument = Instrument(...) ) some_astropy_table.meta = ref.to_dict() some_astropy_table.write("output.ecsv")
All the reference Metadata required for a CTA output file, plus a way to turn it into a dict() for easy addition to the header of a file
Process (top-level workflow) information
Data product information
Activity (tool) information
Handles reading of different event/waveform containing files