https://github.com/tensorly/tensorly
Tip revision: 4fbbde1c1a610aec38fd2b8bcb500c7f7f459147 authored by Jean Kossaifi on 19 November 2020, 00:12:03 UTC
validate_tucker_rank: opt for modes w/ fixed ranks
validate_tucker_rank: opt for modes w/ fixed ranks
Tip revision: 4fbbde1
numpy_backend.py
import numpy as np
from .core import Backend
class NumpyBackend(Backend):
backend_name = 'numpy'
@staticmethod
def context(tensor):
return {'dtype': tensor.dtype}
@staticmethod
def tensor(data, dtype=None):
return np.array(data, dtype=dtype)
@staticmethod
def is_tensor(tensor):
return isinstance(tensor, np.ndarray)
@staticmethod
def to_numpy(tensor):
return np.copy(tensor)
@staticmethod
def shape(tensor):
return tensor.shape
@staticmethod
def ndim(tensor):
return tensor.ndim
@staticmethod
def clip(tensor, a_min=None, a_max=None):
return np.clip(tensor, a_min, a_max)
@staticmethod
def dot(a, b):
return a.dot(b)
@staticmethod
def norm(tensor, order=2, axis=None):
# handle difference in default axis notation
if axis == ():
axis = None
if order == 'inf':
return np.max(np.abs(tensor), axis=axis)
if order == 1:
return np.sum(np.abs(tensor), axis=axis)
elif order == 2:
return np.sqrt(np.sum(tensor**2, axis=axis))
else:
return np.sum(np.abs(tensor)**order, axis=axis)**(1 / order)
def kr(self, matrices, weights=None, mask=None):
n_columns = matrices[0].shape[1]
n_factors = len(matrices)
start = ord('a')
common_dim = 'z'
target = ''.join(chr(start + i) for i in range(n_factors))
source = ','.join(i + common_dim for i in target)
operation = source + '->' + target + common_dim
if weights is not None:
matrices = [m if i else m*self.reshape(weights, (1, -1)) for i, m in enumerate(matrices)]
m = mask.reshape((-1, 1)) if mask is not None else 1
return np.einsum(operation, *matrices).reshape((-1, n_columns))*m
@property
def SVD_FUNS(self):
return {'numpy_svd': self.partial_svd,
'truncated_svd': self.partial_svd}
@staticmethod
def sort(tensor, axis, descending = False):
if descending:
return np.flip(np.sort(tensor, axis=axis), axis = axis)
else:
return np.sort(tensor, axis=axis)
for name in ['int64', 'int32', 'float64', 'float32', 'reshape', 'moveaxis',
'where', 'copy', 'transpose', 'arange', 'ones', 'zeros',
'zeros_like', 'eye', 'kron', 'concatenate', 'max', 'min',
'all', 'mean', 'sum', 'prod', 'sign', 'abs', 'sqrt', 'argmin',
'argmax', 'stack', 'conj', 'diag', 'einsum']:
NumpyBackend.register_method(name, getattr(np, name))
for name in ['solve', 'qr']:
NumpyBackend.register_method(name, getattr(np.linalg, name))