# http://pyrocko.org - GPLv3
#
# The Pyrocko Developers, 21st Century
# ---|P------/S----------~Lg----------
'''
NumPy support for :py:mod:`pyrocko.guts`.
'''
import numpy as num
from io import BytesIO
from base64 import b64decode, b64encode
import binascii
from .guts import TBase, Object, ValidationError, literal
try:
unicode
except NameError:
unicode = str
restricted_dtype_map = {
num.dtype('float64'): '<f8',
num.dtype('float32'): '<f4',
num.dtype('int64'): '<i8',
num.dtype('int32'): '<i4',
num.dtype('int16'): '<i2',
num.dtype('int8'): '<i1'}
restricted_dtype_map_rev = dict(
(v, k) for (k, v) in restricted_dtype_map.items())
def array_equal(a, b):
return a.dtype == b.dtype \
and a.shape == b.shape \
and num.all(a == b)
[docs]class Array(Object):
'''
Placeholder for :py:class:`numpy.ndarray`
Normally, no objects of this class should be instatiated. It is needed
by Guts' type system.
'''
dummy_for = num.ndarray
dummy_for_description = 'numpy.ndarray'
class __T(TBase):
def __init__(
self,
shape=None,
dtype=None,
serialize_as='table',
serialize_dtype=None,
*args, **kwargs):
TBase.__init__(self, *args, **kwargs)
self.shape = shape
self.dtype = dtype
assert serialize_as in (
'table', 'base64', 'list', 'npy',
'base64+meta', 'base64-compat')
self.serialize_as = serialize_as
self.serialize_dtype = serialize_dtype
def is_default(self, val):
if self._default is None:
return val is None
elif val is None:
return False
else:
return array_equal(self._default, val)
def regularize_extra(self, val):
if isinstance(val, str):
ndim = None
if self.shape:
ndim = len(self.shape)
if self.serialize_as == 'table':
val = num.loadtxt(
BytesIO(val.encode('utf-8')),
dtype=self.dtype, ndmin=ndim)
elif self.serialize_as == 'base64':
data = b64decode(val)
val = num.frombuffer(
data, dtype=self.serialize_dtype).astype(self.dtype)
elif self.serialize_as == 'base64-compat':
try:
data = b64decode(val)
val = num.frombuffer(
data,
dtype=self.serialize_dtype).astype(self.dtype)
except binascii.Error:
val = num.loadtxt(
BytesIO(val.encode('utf-8')),
dtype=self.dtype, ndmin=ndim)
elif self.serialize_as == 'npy':
data = b64decode(val)
try:
val = num.load(BytesIO(data), allow_pickle=False)
except TypeError:
# allow_pickle only available in newer NumPy
val = num.load(BytesIO(data))
elif isinstance(val, dict):
if self.serialize_as == 'base64+meta':
if not sorted(val.keys()) == ['data', 'dtype', 'shape']:
raise ValidationError(
'array in format "base64+meta" must have keys '
'"data", "dtype", and "shape"')
shape = val['shape']
if not isinstance(shape, list):
raise ValidationError('invalid shape definition')
for n in shape:
if not isinstance(n, int):
raise ValidationError('invalid shape definition')
serialize_dtype = val['dtype']
allowed = list(restricted_dtype_map_rev.keys())
if self.serialize_dtype is not None:
allowed.append(self.serialize_dtype)
if serialize_dtype not in allowed:
raise ValidationError(
'only the following dtypes are allowed: %s'
% ', '.join(sorted(allowed)))
data = val['data']
if not isinstance(data, str):
raise ValidationError(
'data must be given as a base64 encoded string')
data = b64decode(data)
dtype = self.dtype or \
restricted_dtype_map_rev[serialize_dtype]
val = num.frombuffer(
data, dtype=serialize_dtype).astype(dtype)
if val.size != num.prod(shape):
raise ValidationError('size/shape mismatch')
val = val.reshape(shape)
else:
val = num.asarray(val, dtype=self.dtype)
return val
def validate_extra(self, val):
if not isinstance(val, num.ndarray):
raise ValidationError(
'object %s is not of type numpy.ndarray: %s' % (
self.xname(), type(val)))
if self.dtype is not None and self.dtype != val.dtype:
raise ValidationError(
'array %s not of required type: need %s, got %s' % (
self.xname(), self.dtype, val.dtype))
if self.shape is not None:
la, lb = len(self.shape), len(val.shape)
if la != lb:
raise ValidationError(
'array %s dimension mismatch: need %i, got %i' % (
self.xname(), la, lb))
for a, b in zip(self.shape, val.shape):
if a is not None:
if a != b:
raise ValidationError(
'array %s shape mismatch: need %s, got: %s' % (
self.xname(), self.shape, val.shape))
def to_save(self, val):
if self.serialize_as == 'table':
out = BytesIO()
num.savetxt(out, val, fmt='%12.7g')
return literal(out.getvalue().decode('utf-8'))
elif self.serialize_as == 'base64' \
or self.serialize_as == 'base64-compat':
data = val.astype(self.serialize_dtype).tobytes()
return literal(b64encode(data).decode('utf-8'))
elif self.serialize_as == 'list':
if self.dtype == complex:
return [repr(x) for x in val]
else:
return val.tolist()
elif self.serialize_as == 'npy':
out = BytesIO()
try:
num.save(out, val, allow_pickle=False)
except TypeError:
# allow_pickle only available in newer NumPy
num.save(out, val)
return literal(b64encode(out.getvalue()).decode('utf-8'))
elif self.serialize_as == 'base64+meta':
serialize_dtype = self.serialize_dtype or \
restricted_dtype_map[val.dtype]
data = val.astype(serialize_dtype).tobytes()
return dict(
dtype=serialize_dtype,
shape=val.shape,
data=literal(b64encode(data).decode('utf-8')))
__all__ = ['Array']
