""" Utilities for handling different array flavors in PyTables. :Author: Ivan Vilata i Balaguer :Contact: ivilata at carabos dot com :License: BSD :Created: December 30, 2006 :Revision: $Id: flavor.py 3309 2007-10-18 12:10:10Z ivilata $ Variables ========= `__docformat`__ The format of documentation strings in this module. `__version__` Repository version of this file. `internal_flavor` The flavor used internally by PyTables. `all_flavors` List of all flavors available to PyTables. `alias_map` Maps old flavor names to the most similar current flavor. `description_map` Maps flavors to short descriptions of their supported objects. `identifier_map` Maps flavors to functions that can identify their objects. The function associated with a given flavor will return a true value if the object passed to it can be identified as being of that flavor. See the `flavor_of()` function for a friendlier interface to flavor identification. `converter_map` Maps (source, destination) flavor pairs to converter functions. Converter functions get an array of the source flavor and return an array of the destination flavor. See the `array_of_flavor()` and `flavor_to_flavor()` functions for friendlier interfaces to flavor conversion. """ # Imports # ======= import warnings from tables.exceptions import FlavorError, FlavorWarning # Public variables # ================ __docformat__ = 'reStructuredText' """The format of documentation strings in this module.""" __version__ = '$Revision: 3309 $' """Repository version of this file.""" internal_flavor = 'numpy' """The flavor used internally by PyTables.""" # This is very slightly slower than a set for a small number of values # in terms of (infrequent) lookup time, but allows `flavor_of()` # (which may be called much more frequently) to check for flavors in # order, beginning with the most common one. all_flavors = [] # filled as flavors are registered """List of all flavors available to PyTables.""" alias_map = {} # filled as flavors are registered """Maps old flavor names to the most similar current flavor.""" description_map = {} # filled as flavors are registered """Maps flavors to short descriptions of their supported objects.""" identifier_map = {} # filled as flavors are registered """ Maps flavors to functions that can identify their objects. The function associated with a given flavor will return a true value if the object passed to it can be identified as being of that flavor. See the `flavor_of()` function for a friendlier interface to flavor identification. """ converter_map = {} # filled as flavors are registered """ Maps (source, destination) flavor pairs to converter functions. Converter functions get an array of the source flavor and return an array of the destination flavor. See the `array_of_flavor()` and `flavor_to_flavor()` functions for friendlier interfaces to flavor conversion. """ # Public functions # ================ def check_flavor(flavor): """Raise a ``FlavorError`` if the `flavor` is not valid.""" if flavor not in all_flavors: available_flavs = ", ".join(flav for flav in all_flavors) raise FlavorError( "flavor ``%s`` is unsupported or unavailable; " "available flavors in this system are: %s" % (flavor, available_flavs) ) def array_of_flavor2(array, src_flavor, dst_flavor): """ Get a version of the given `array` in a different flavor. The input `array` must be of the given `src_flavor`, and the returned array will be of the indicated `dst_flavor`. Both flavors may be the same, but it is not guaranteed that the returned array will be the same object as the input one in this case. If the conversion is not supported, a ``FlavorError`` is raised. """ convkey = (src_flavor, dst_flavor) if convkey not in converter_map: raise FlavorError( "conversion from flavor ``%s`` to flavor ``%s`` " "is unsupported or unavailable in this system" % (src_flavor, dst_flavor) ) convfunc = converter_map[convkey] return convfunc(array) def flavor_to_flavor(array, src_flavor, dst_flavor): """ Get a version of the given `array` in a different flavor. The input `array` must be of the given `src_flavor`, and the returned array will be of the indicated `dst_flavor` (see below for an exception to this). Both flavors may be the same, but it is not guaranteed that the returned array will be the same object as the input one in this case. If the conversion is not supported, a `FlavorWarning` is issued and the input `array` is returned as is. """ try: return array_of_flavor2(array, src_flavor, dst_flavor) except FlavorError, fe: warnings.warn( "%s; returning an object of the ``%s`` flavor instead" % (fe.args[0], src_flavor), FlavorWarning ) return array def internal_to_flavor(array, dst_flavor): """ Get a version of the given `array` in a different `dst_flavor`. The input `array` must be of the internal flavor, and the returned array will be of the given `dst_flavor`. See `flavor_to_flavor()` for more information. """ return flavor_to_flavor(array, internal_flavor, dst_flavor) def array_as_internal(array, src_flavor): """ Get a version of the given `array` in the internal flavor. The input `array` must be of the given `src_flavor`, and the returned array will be of the internal flavor. If the conversion is not supported, a ``FlavorError`` is raised. """ return array_of_flavor2(array, src_flavor, internal_flavor) def flavor_of(array): """ Identify the flavor of a given `array`. If the `array` can not be matched with any flavor, a ``TypeError`` is raised. """ for flavor in all_flavors: if identifier_map[flavor](array): return flavor type_name = type(array).__name__ supported_descs = "; ".join(description_map[fl] for fl in all_flavors) raise TypeError( "objects of type ``%s`` are not supported in this context, sorry; " "supported objects are: %s" % (type_name, supported_descs) ) def array_of_flavor(array, dst_flavor): """ Get a version of the given `array` in a different `dst_flavor`. The flavor of the input `array` is guessed, and the returned array will be of the given `dst_flavor`. If the conversion is not supported, a ``FlavorError`` is raised. """ return array_of_flavor2(array, flavor_of(array), dst_flavor) def restrict_flavors(keep=['python']): """ Disable all flavors except those in `keep`. Providing an empty `keep` sequence implies disabling all flavors (but the internal one). If the sequence is not specified, only optional flavors are disabled. .. Important:: Once you disable a flavor, it can not be enabled again. """ keep = set(keep).union([internal_flavor]) remove = set(all_flavors).difference(keep) for flavor in remove: _disable_flavor(flavor) # Flavor registration # =================== # # The order in which flavors appear in `all_flavors` determines the # order in which they will be tested for by `flavor_of()`, so place # most frequent flavors first. import numpy all_flavors.append('numpy') # this is the internal flavor all_flavors.append('python') # this is always supported try: import numarray import numarray.generic import numarray.strings import numarray.records except ImportError: pass else: all_flavors.append('numarray') try: import Numeric except ImportError: pass else: all_flavors.append('numeric') def _register_aliases(): """Register aliases of *available* flavors.""" for flavor in all_flavors: aliases = eval('_%s_aliases' % flavor) for alias in aliases: alias_map[alias] = flavor def _register_descriptions(): """Register descriptions of *available* flavors.""" for flavor in all_flavors: description_map[flavor] = eval('_%s_desc' % flavor) def _register_identifiers(): """Register identifier functions of *available* flavors.""" for flavor in all_flavors: identifier_map[flavor] = eval('_is_%s' % flavor) def _register_converters(): """Register converter functions between *available* flavors.""" def identity(array): return array for src_flavor in all_flavors: for dst_flavor in all_flavors: # Converters with the same source and destination flavor # are used when available, since they may perform some # optimizations on the resulting array (e.g. making it # contiguous). Otherwise, an identity function is used. convfunc = None try: convfunc = eval('_conv_%s_to_%s' % (src_flavor, dst_flavor)) except NameError: if src_flavor == dst_flavor: convfunc = identity if convfunc: converter_map[(src_flavor, dst_flavor)] = convfunc def _register_all(): """Register all *available* flavors.""" _register_aliases() _register_descriptions() _register_identifiers() _register_converters() def _deregister_aliases(flavor): """Deregister aliases of a given `flavor` (no checks).""" for (an_alias, a_flavor) in alias_map.items(): if a_flavor == flavor: del alias_map[an_alias] def _deregister_description(flavor): """Deregister description of a given `flavor` (no checks).""" del description_map[flavor] def _deregister_identifier(flavor): """Deregister identifier function of a given `flavor` (no checks).""" del identifier_map[flavor] def _deregister_converters(flavor): """Deregister converter functions of a given `flavor` (no checks).""" for flavor_pair in converter_map.keys(): if flavor in flavor_pair: del converter_map[flavor_pair] def _disable_flavor(flavor): """Completely disable the given `flavor` (no checks).""" _deregister_aliases(flavor) _deregister_description(flavor) _deregister_identifier(flavor) _deregister_converters(flavor) all_flavors.remove(flavor) # Implementation of flavors # ========================= _python_aliases = [ 'List', 'Tuple', 'Int', 'Float', 'String', 'VLString', 'Object' ] _python_desc = ( "homogeneous list or tuple, " "integer, float, complex or string" ) def _is_python(array): return isinstance(array, (tuple, list, int, float, complex, str)) _numpy_aliases = [] _numpy_desc = "NumPy array, record or scalar" def _is_numpy(array): return isinstance(array, (numpy.ndarray, numpy.generic)) _numarray_aliases = ['NumArray', 'CharArray'] _numarray_desc = "numarray array or record" def _is_numarray(array): na_array_or_record = (numarray.generic.NDArray, numarray.records.Record) return isinstance(array, na_array_or_record) _numeric_aliases = ['Numeric'] _numeric_desc = "Numeric array" def _is_numeric(array): return isinstance(array, Numeric.ArrayType) def _numpy_contiguous(convfunc): """Decorate `convfunc` to return a *contiguous* NumPy array.""" def conv_to_numpy(array): nparr = convfunc(array) if hasattr(nparr, 'flags') and not nparr.flags.contiguous: nparr = nparr.copy() # copying the array makes it contiguous return nparr conv_to_numpy.__name__ = convfunc.__name__ conv_to_numpy.__doc__ = convfunc.__doc__ return conv_to_numpy @_numpy_contiguous def _conv_numpy_to_numpy(array): # Passes contiguous arrays through and converts scalars into # scalar arrays. return numpy.asarray(array) @_numpy_contiguous def _conv_numarray_to_numpy(array): # Homogeneous arrays. if isinstance(array, (numarray.NumArray, numarray.strings.CharArray)): return numpy.asarray(array) # use the array protocol # Heterogeneous arrays and records. record = None if isinstance(array, numarray.records.Record): # Get a RecArray from a record record = array row = record.row array = record.array[row:row+1] if type(array) is numarray.records.RecArray: # Create a NestedRecArray array from the RecArray to easy the # conversion. This is sub-optimal and should be replaced by a # faster way to convert a plain RecArray into a numpy recarray. # F. Altet 2006-06-19 array = nra.array(array) nparray = numpy.ndarray( buffer=array._data, shape=array.shape, dtype=array.array_descr, offset=array._byteoffset ) if record: return nparray[row] # get the NumPy record return nparray @_numpy_contiguous def _conv_numeric_to_numpy(array): return numpy.asarray(array) # use the array protocol @_numpy_contiguous def _conv_python_to_numpy(array): return numpy.array(array) if 'numeric' in all_flavors: _numtype_from_nptype = { numpy.bool_: Numeric.UInt8, numpy.int8: Numeric.Int8, numpy.int16: Numeric.Int16, numpy.int32: Numeric.Int32, numpy.uint8: Numeric.UInt8, numpy.uint16: Numeric.UInt16, numpy.uint32: Numeric.UInt32, numpy.float32: Numeric.Float32, numpy.float64: Numeric.Float64, numpy.complex64: Numeric.Complex32, numpy.complex128: Numeric.Complex64 } if hasattr(Numeric, "Int64"): # Only defined for 64-bit platforms _numtype_from_nptype[numpy.int64] = Numeric.Int64 def _conv_numpy_to_numeric(array): kind = array.dtype.kind if kind == 'V': raise FlavorError( "the ``numeric`` flavor does not support " "heterogeneous arrays" ) # It seems that the array protocol in Numeric does leak. See # http://comments.gmane.org/gmane.comp.python.numeric.general/12563 # for more info on this issue. ##if kind != 'S': ## return Numeric.asarray(array) # use the array protocol shape = array.shape if kind == 'S': if array.itemsize > 1: # Numeric does not support character arrays with elements # with a size > 1. Simulate with an additional dimension. shape = shape + (array.itemsize,) typecode = 'c' else: # See the above note about the Numeric leak. typecode = _numtype_from_nptype[array.dtype.type] # Convert to a contiguous buffer (``tostring()`` is very efficient). arrstr = array.tostring() array = Numeric.fromstring(arrstr, typecode) array = Numeric.reshape(array, shape) return array if 'numarray' in all_flavors: from tables import nra def _conv_numpy_to_numarray(array): kind = array.dtype.kind if kind == 'S': # homogeneous string array # We can't use the array protocol to do this conversion if array.shape == (): array = array.item() return numarray.strings.array( buffer=array, shape=array.shape, itemsize=array.itemsize, padc='\x00' ) if kind != 'V': # homogeneous array # NumPy scalars are mishandled by numarray, see #98. This # case may be removed when the bug in numarray is fixed. if numpy.isscalar(array): # A problem with the ``item()`` conversion is the lose of # precise type information (see #125). natype = numpy.sctypeNA[type(array)] return numarray.array(array.item(), type=natype) # This works for regular homogeneous arrays and even for rank-0 arrays # Using asarray gives problems in some tests (I don't know exactly why) ##return numarray.asarray(array) # use the array protocol return numarray.array(array) # use the array protocol (with copy) # For the remaining heterogeneous arrays and records, leave the # task to ``nra``. return nra.fromnumpy(array) def _conv_numpy_to_python(array): if array.shape != (): # Lists are the default for returning multidimensional objects array = array.tolist() else: # 0-dim or scalar case array = array.item() return array # Now register everything related with *available* flavors. _register_all() # Main part # ========= def _test(): """Run ``doctest`` on this module.""" import doctest doctest.testmod() if __name__ == '__main__': _test()