# Copyright 2025 Daniil Shmelev
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# =========================================================================
from typing import Union
import numpy as np
import torch
from .param_checks import check_type, check_non_neg, check_log_sig_method, check_n_jobs
from .error_codes import err_msg
from .dtypes import (CPSIG_SIG_TO_LOG_SIG_BACKPROP,
CUSIG_SIG_TO_LOG_SIG_BACKPROP_CUDA,
CPSIG_LOG_SIG_FROM_PATH_BACKPROP, CUSIG_LOG_SIG_FROM_PATH_BACKPROP_CUDA)
from .sig_length import sig_length, log_sig_length, aug_dim, _infer_scalar_term
from .data_handlers import SigOutputHandler, SigInputHandler, PathInputHandler, PathOutputHandler
[docs]
def sig_to_log_sig_backprop(
sig : Union[np.ndarray, torch.tensor],
log_sig_derivs : Union[np.ndarray, torch.tensor],
dimension : int,
degree : int,
*,
time_aug : bool = False,
lead_lag : bool = False,
method : int = 1,
n_jobs : int = 1
) -> Union[np.ndarray, torch.tensor]:
"""
Backpropagates through the ``pysiglib.sig_to_log_sig`` function.
Given the derivatives of a scalar function :math:`F` with respect to the
log signature, :math:`\\partial F / \\partial \\log(S(x))`, returns the
derivatives of :math:`F` with respect to the signature,
:math:`\\partial F / \\partial S(x)`.
:param sig: The signature or batch of signatures, given as a `numpy.ndarray` or `torch.tensor`.
For a single signature, this must be of shape ``sig_length``. For a batch of paths, this must
be of shape ``(batch_size, sig_length)``.
:type sig: numpy.ndarray | torch.tensor
:param log_sig_derivs: Derivatives of the scalar function :math:`F` with respect to the log signature(s),
:math:`\\partial F / \\partial S(x)`. This must be an array of the same shape as the
log signature(s).
:type log_sig_derivs: numpy.ndarray | torch.tensor
:param dimension: Dimension of the underlying path(s).
:type dimension: int
:param degree: Truncation degree of the (log) signature(s).
:type degree: int
:param time_aug: Whether the signatures were computed with ``time_aug=True``.
:type time_aug: bool
:param lead_lag: Whether the signatures were computed with ``lead_lag=True``.
:type lead_lag: bool
:param method: Method used for the log signature computation (`0`, `1` or `2`).
:type method: int
:param n_jobs: Number of threads to run in parallel.
If n_jobs = 1, the computation is run serially. If set to -1, all available threads
are used. For n_jobs below -1, (max_threads + 1 + n_jobs) threads are used. For example
if n_jobs = -2, all threads but one are used.
:type n_jobs: int
:return: Derivatives of the scalar function :math:`F` with respect to the signature(s),
:math:`\\partial F / \\partial S(x)`.
This is an array of the same shape as the provided signature(s).
:rtype: numpy.ndarray | torch.tensor
Example:
---------
.. code-block:: python
import torch
import pysiglib
batch, length, dimension, degree = 10, 100, 5, 4
pysiglib.prepare_log_sig(dimension, degree, time_aug=True, method=1)
path = torch.rand((batch, length, dimension))
sigs = pysiglib.sig(path, degree, time_aug=True)
log_sigs = pysiglib.sig_to_log_sig(sigs, dimension, degree, time_aug=True, method=1)
log_sig_derivs = torch.ones_like(log_sigs)
sig_derivs = pysiglib.sig_to_log_sig_backprop(
sigs, log_sig_derivs, dimension, degree,
time_aug=True, method=1,
)
print(sig_derivs)
"""
scalar_term = _infer_scalar_term(sig, dimension, degree, time_aug=time_aug, lead_lag=lead_lag)
check_type(dimension, "dimension", int)
check_non_neg(dimension, "dimension")
check_type(degree, "degree", int)
check_non_neg(degree, "degree")
check_type(time_aug, "time_aug", bool)
check_type(lead_lag, "lead_lag", bool)
check_type(method, "method", int)
check_log_sig_method(method)
aug_dimension = aug_dim(dimension, time_aug, lead_lag)
sig_len = sig_length(dimension, degree, time_aug=time_aug, lead_lag=lead_lag, scalar_term=scalar_term)
log_sig_len = log_sig_length(dimension, degree, time_aug=time_aug, lead_lag=lead_lag) if method else sig_length(dimension, degree, time_aug=time_aug, lead_lag=lead_lag, scalar_term=scalar_term)
data = SigInputHandler(sig, sig_len, "sig")
derivs_data = SigInputHandler(log_sig_derivs, log_sig_len, "log_sig_derivs")
if data.dtype != derivs_data.dtype:
raise ValueError("sig and log_sig_derivs must have the same dtype")
result = SigOutputHandler(data, sig_len)
if data.batch_size == 0:
return result.data
check_n_jobs(n_jobs)
if data.device == "cpu":
err_code = CPSIG_SIG_TO_LOG_SIG_BACKPROP[data.dtype](
data.data_ptr, result.data_ptr, derivs_data.data_ptr,
data.batch_size, dimension, degree,
time_aug, lead_lag, method, scalar_term, n_jobs)
else:
err_code = CUSIG_SIG_TO_LOG_SIG_BACKPROP_CUDA[data.dtype](
data.data_ptr, result.data_ptr, derivs_data.data_ptr,
data.batch_size, aug_dimension, degree, method, scalar_term)
if err_code:
raise Exception("Error in pysiglib.sig_to_log_sig_backprop: " + err_msg(err_code))
return result.data
def _log_sig_from_path_backprop(
grad_output : Union[np.ndarray, torch.tensor],
path : Union[np.ndarray, torch.tensor],
degree : int,
*,
n_jobs : int = 1,
) -> Union[np.ndarray, torch.tensor]:
"""Backpropagates through the method=3 log-signature-from-path computation."""
check_type(degree, "degree", int)
check_non_neg(degree, "degree")
check_n_jobs(n_jobs)
data = PathInputHandler(path, False, False, 1.0, "path")
ls_len = log_sig_length(data.data_dimension, degree)
derivs_data = SigInputHandler(grad_output, ls_len, "grad_output")
if data.dtype != derivs_data.dtype:
raise ValueError("grad_output and path must have the same dtype")
result = PathOutputHandler(data.data_length, data.data_dimension, data)
if data.batch_size == 0:
return result.data
if data.device == "cpu":
err_code = CPSIG_LOG_SIG_FROM_PATH_BACKPROP[data.dtype](
derivs_data.data_ptr, result.data_ptr, data.data_ptr,
data.batch_size, data.data_length, data.data_dimension, degree, n_jobs)
else:
err_code = CUSIG_LOG_SIG_FROM_PATH_BACKPROP_CUDA[data.dtype](
derivs_data.data_ptr, result.data_ptr, data.data_ptr,
data.batch_size, data.data_length, data.data_dimension, degree)
if err_code:
raise Exception("Error in pysiglib.log_sig_from_path_backprop: " + err_msg(err_code))
return result.data
