Adjoint improvement

.github/workflows/release.yml

@@ -1,6 +1,9 @@
 name: Publish Python 🐍 distribution 📦 to PyPI and TestPyPI
+on:
+  push:
+    branches:
+      - master  # Set this to your default branch
 
-on: push
 
 jobs:
   build:

.github/workflows/run_tests.yml

@@ -1,10 +1,7 @@
 name: Python Tests
 
 on:
-  push:
-    branches: [ master ]
   pull_request:
-    branches: [ master ]
 
 jobs:
   test:

docs/usage/faq.md

@@ -10,11 +10,11 @@ delays = torch.tensor([1.0, 2.0])
 history_function = lambda t : ...
 ts = ...
 
-def simple_dde(t, y, args, *, history):
+def simple_dde(t, y, func_args, *, history):
     # this correspond to y'(t) = -y(t-1) - y(t-2)
     return - history[0] - history[1]
 
-ys = torchdde.integrate(f, solver, ts[0], ts[-1], ts, history_function, args=None, dt0=ts[1]-ts[0], delays=delays)
+ys = torchdde.integrate(f, solver, ts[0], ts[-1], ts, history_function, func_args=None, dt0=ts[1]-ts[0], delays=delays)
 ```
 
 ## How about if I want a neural network to have also several delays ?

docs/usage/getting-started.md

@@ -12,7 +12,7 @@ import matplotlib.pyplot as plt
 from torchdde import RK4
 import torch
 
-def simple_dde(t, y, args, *, history):
+def simple_dde(t, y, func_args, *, history):
     # `history` corresponds to the list of
     # delayed states defined in your DDE
     # i.e. here history=[y(t-2)]
@@ -54,7 +54,7 @@ import matplotlib.pyplot as plt
 from torchdde import RK4
 import torch
 
-def simple_ode(t, y, args):
+def simple_ode(t, y, func_args):
     return -y**2
 
 device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

docs/usage/integration-de.md

@@ -13,10 +13,10 @@ What essentially differentiates DDEs with ODEs are :
 
     In practice, your function will be defined like this :  
     ```python
-    def f_ode(t,y,args):
+    def f_ode(t,y,func_args):
         return ...
 
-    def f_dde(t,y,args, history):
+    def f_dde(t,y,func_args, history):
         return ...
     ```
 

docs/usage/neural-dde.md

@@ -43,7 +43,7 @@ class NDDE(nn.Module):
             hidden_channels=depth * [width_size] + [out_size],
         )
 
-    def forward(self, t, z, args, *, history):
+    def forward(self, t, z, func_args, *, history):
         # `history` corresponds to the list of
         # delayed states defined in your DDE
         # i.e. here history=[y(t-tau1), ..., y(t-taun)]
@@ -54,11 +54,11 @@ We generate the toy dataset of the [delayed logistic equation](https://www.math.
 
 ```python
 def get_data(y0, ts, tau=torch.tensor([1.0])):
-    def f(t, y, args, history):
+    def f(t, y, func_args, history):
         return y * (1 - history[0])
 
     history_function = lambda t: torch.unsqueeze(y0, dim=1)
-    ys = integrate(f, Euler(), ts[0], ts[-1], ts, history_function, args=None, dt0=ts[1]-ts[0], delays=tau)
+    ys = integrate(f, Euler(), ts[0], ts[-1], ts, history_function, func_args=None, dt0=ts[1]-ts[0], delays=tau)
     return ys
 
 
@@ -136,7 +136,7 @@ def main(
         plt.plot(ts.cpu(), data[0].cpu(), c="dodgerblue", label="Real")
         history_values = data[0, 0][..., None]
         history_fn = lambda t: history_values
-        ys_pred = integrate(model, Euler(), ts[0], ts[-1], ts, history_fn, args=None, dt0=ts[1]-ts[0], delays=tau)
+        ys_pred = integrate(model, Euler(), ts[0], ts[-1], ts, history_fn, func_args=None, dt0=ts[1]-ts[0], delays=tau)
         plt.plot(
             ts.cpu(),
             ys_pred[0].cpu().detach(),

docs/usage/training-de.md

@@ -22,7 +22,7 @@ for step, data in enumerate(train_loader):
         t1=ts[-1],
         ts=ts,
         y0=...,
-        args=None,
+        func_args=...,
         dt0=ts[1] - ts[0],
         delays=...,
     )

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "torchdde"
-version = "0.1.2"
+version = "0.2.0"
 description = "DDE numerical solvers in Python."
 readme = "README.md"
 requires-python =">=3.9"

torchdde/__init__.py

@@ -11,6 +11,7 @@
     FourthOrderPolynomialInterpolation as FourthOrderPolynomialInterpolation,
     ThirdOrderPolynomialInterpolation as ThirdOrderPolynomialInterpolation,
 )
+from .misc import TupleTensorTransformer as TupleTensorTransformer
 from .solver import (
     AbstractOdeSolver as AbstractOdeSolver,
     Bosh3 as Bosh3,