Function random_dates

tf_quant_finance/datetime/date_tensor.py:694–759

Generates random dates between the supplied start and end dates. Generates specified number of random dates between the given start and end dates. The start and end dates are supplied as `DateTensor` objects. The dates uniformly distributed between the start date (inclusive) and end date (e

(*, start_date, end_date, size=1, seed=None, name=None)

def random_dates(*, start_date, end_date, size=1, seed=None, name=None):
  """Generates random dates between the supplied start and end dates.

  Generates specified number of random dates between the given start and end
  dates. The start and end dates are supplied as `DateTensor` objects. The dates
  uniformly distributed between the start date (inclusive) and end date
  (exclusive). Note that the dates are uniformly distributed over the calendar
  range, i.e. no holiday calendar is taken into account.

  Args:
    start_date: DateTensor of arbitrary shape. The start dates of the range from
      which to sample. The start dates are themselves included in the range.
    end_date: DateTensor of shape compatible with the `start_date`. The end date
      of the range from which to sample. The end dates are excluded from the
      range.
    size: Positive scalar int32 Tensor. The number of dates to draw between the
      start and end date.
      Default value: 1.
    seed: Optional seed for the random generation.
    name: Optional str. The name to give to the ops created by this function.
      Default value: 'random_dates'.

  Returns:
    A DateTensor of shape [size] + dates_shape where dates_shape is the common
    broadcast shape for (start_date, end_date).

  #### Example

  ```python
  # Note that the start and end dates need to be of broadcastable shape (though
  # not necessarily the same shape).
  # In this example, the start dates are of shape [2] and the end dates are
  # of a compatible but non-identical shape [1].
  start_dates = tff.datetime.dates_from_tuples([
    (2020, 5, 16),
    (2020, 6, 13)
  ])
  end_dates = tff.datetime.dates_from_tuples([(2021, 5, 21)])
  size = 3  # Generate 3 dates for each pair of (start, end date).
  sample = tff.datetime.random_dates(start_date=start_dates, end_date=end_dates,
                              size=size)
  # sample is a DateTensor of shape [3, 2]. The [3] is from the size and [2] is
  # the common broadcast shape of start and end date.
  ```
  """
  with tf.name_scope(name or "random_dates"):
    size = tf.reshape(
        tf.convert_to_tensor(size, dtype=tf.int32, name="size"), [-1])
    start_date = convert_to_date_tensor(start_date)
    end_date = convert_to_date_tensor(end_date)
    # Note that tf.random.uniform cannot deal with non scalar max value with
    # int dtypes. So we do this in float64 space and then floor. This incurs
    # some non-uniformity of the distribution but for practical purposes this
    # will be negligible.
    ordinal_range = tf.cast(
        end_date.ordinal() - start_date.ordinal(), dtype=tf.float64)
    sample_shape = tf.concat((size, tf.shape(ordinal_range)), axis=0)
    ordinal_sample = tf.cast(