tf_quant_finance.math.pde.grids.log_uniform_grid

Last updated: 2023-03-16.

tf_quant_finance.math.pde.grids.log_uniform_grid#

View source

Creates a grid spec for a uniform grid in a log-space.

tf_quant_finance.math.pde.grids.log_uniform_grid(
    minimums, maximums, sizes, dtype=None, validate_args=False, name=None
)

A log-uniform grid is characterized by having a constant gap between neighboring points along each axis in the log-space, i.e., the logarithm of output grid is the uniform grid.

Note that the shape of all three parameters must be fully defined and equal to each other. The shape is used to determine the dimension of the grid. Note that all the parameters are supplied and returned for the original space and not the log-space.

Examples#

dtype = np.float64
min_x, max_x, sizes = [0.1], [3.0], [5]
# Here min_x and max_x are in the original space and *not* in the log-space.
grid = log_uniform_grid(min_x, max_x, sizes,dtype=dtype)
with tf.Session() as sess:
  grid = sess.run(grid)
# Note that the minimum and maximum grid locations are the same as min_x and
# max_x.
print('locations: ', grid.locations)
# locations:  [array([ 0.1, 0.234, 0.548, 1.282, 3.0])]
print('grid: ', grid.grid)
# grid: array([[ 0.1], [0.234], [0.548], [1.282], [ 3.0]])
print('deltas: ', grid.deltas)
# deltas: [array([ 0.134, 0.314, 0.734, 1.718])]

Args:#

  • minimums: Real Tensor of rank 1 containing the lower end points of the output grid. Must have the same shape as those of maximums and sizes args.

  • maximums: Tensor of the same dtype and shape as minimums. The upper endpoints of the output grid.

  • sizes: Integer Tensor of the same shape as minimums. The size of the grid in each axis. Each entry must be greater than or equal to 2 (i.e. the sizes include the end points).

  • dtype: Optional tf.dtype. The default dtype to use for the grid.

  • validate_args: Python boolean indicating whether to validate the supplied arguments. The validation checks performed are (a) maximums > minimums (b) minimums > 0.0 (c) sizes >= 2.

  • name: Python str. The name prefixed to the ops created by this function. If not supplied, the default name ‘uniform_grid_spec’ is used.

Returns:#

The grid locations as projected along each axis. One Tensor of shape [..., n], where n is the number of points along that axis. The first dimensions are the batch shape. The grid itself can be seen as a cartesian product of the locations array.

Raises:#

ValueError if the shape of maximums, minimums and sizes are not fully defined or they are not identical to each other or they are not rank 1.

Contents