Rename solver arg to ode_solver for clarity

thebuckleylab · Jun 29, 2024 · 13fd3b9 · 13fd3b9
1 parent 2b34e0d
commit 13fd3b9
Show file tree

Hide file tree

Showing 4 changed files with 29 additions and 15 deletions.
diff --git a/README.md b/README.md
@@ -92,3 +92,17 @@ descent) to update the network parameters.
 
 ## Advanced usage
 
+## 📄 Citation
+
+If you found this library useful in your work, please cite (arXiv link):
+
+```bibtex
+@article{innocenti2024jpc,
+    title={JPC: Predictive Coding Networks in JAX},
+    author={Innocenti, Francesco and Kinghorn, Paul and Singh, Ryan and 
+    De Llanza Varona, Miguel and Buckley, Christopher},
+    journal={arXiv preprint},
+    year={2024}
+}
+```
+Also consider starring the repo! ⭐️ 
diff --git a/docs/index.md b/docs/index.md
@@ -4,9 +4,9 @@ JPC is a [JAX](https://github.com/google/jax) library to train neural networks
 with predictive coding. It is built on top of three main libraries:
 
 * [Equinox](https://github.com/patrick-kidger/equinox), to define neural 
-networks with PyTorch-like syntax, and
+networks with PyTorch-like syntax,
 * [Diffrax](https://github.com/patrick-kidger/diffrax), to solve the PC 
-activity (inference) dynamics.
+activity (inference) dynamics, and
 * [Optax](https://github.com/google-deepmind/optax), for parameter optimisation.
 
 JPC provides a simple but flexible API for research of PCNs compatible with

diff --git a/jpc/_test.py b/jpc/_test.py
@@ -50,7 +50,7 @@ def test_generative_pc(
         layer_sizes: PyTree[int],
         batch_size: int,
         sigma: Scalar = 0.05,
-        solver: AbstractSolver = Euler(),
+        ode_solver: AbstractSolver = Euler(),
         dt: float | int = 1,
         n_iters: int = 20,
         stepsize_controller: AbstractStepSizeController = ConstantStepSize()
@@ -74,7 +74,7 @@ def test_generative_pc(
 
     - `sigma`: Standard deviation for Gaussian to sample activities from.
         Defaults to 5e-2.
-    - `solver`: Diffrax (ODE) solver to be used. Default is Euler.
+    - `ode_solver`: Diffrax ODE solver to be used. Default is Euler.
     - `dt`: Integration step size. Defaults to 1.
     - `n_iters`: Number of integration steps (20 as default).
     - `stepsize_controller`: diffrax controller for step size integration.
@@ -96,7 +96,7 @@ def test_generative_pc(
         model=model,
         activities=activities,
         y=y,
-        solver=solver,
+        solver=ode_solver,
         n_iters=n_iters,
         stepsize_controller=stepsize_controller,
         dt=dt
@@ -116,7 +116,7 @@ def test_hpc(
       layer_sizes: PyTree[int],
       batch_size: int,
       sigma: Scalar = 0.05,
-      solver: AbstractSolver = Euler(),
+      ode_solver: AbstractSolver = Euler(),
       dt: float | int = 1,
       n_iters: int = 20,
       stepsize_controller: AbstractStepSizeController = ConstantStepSize()
@@ -142,7 +142,7 @@ def test_hpc(
 
     - `sigma`: Standard deviation for Gaussian to sample activities from.
         Defaults to 5e-2.
-    - `solver`: Diffrax (ODE) solver to be used. Default is Euler.
+    - `ode_solver`: Diffrax ODE solver to be used. Default is Euler.
     - `dt`: Integration step size. Defaults to 1.
     - `n_iters`: Number of integration steps (20 as default).
     - `stepsize_controller`: diffrax controller for step size integration.
@@ -163,7 +163,7 @@ def test_hpc(
         model=generator,
         activities=amort_activities,
         y=y,
-        solver=solver,
+        solver=ode_solver,
         n_iters=n_iters,
         stepsize_controller=stepsize_controller,
         dt=dt
@@ -179,7 +179,7 @@ def test_hpc(
         model=generator,
         activities=activities,
         y=y,
-        solver=solver,
+        solver=ode_solver,
         n_iters=n_iters,
         stepsize_controller=stepsize_controller,
         dt=dt

diff --git a/jpc/_train.py b/jpc/_train.py
@@ -32,7 +32,7 @@ def make_pc_step(
       opt_state: OptState,
       y: ArrayLike,
       x: Optional[ArrayLike] = None,
-      solver: AbstractSolver = Euler(),
+      ode_solver: AbstractSolver = Euler(),
       dt: float | int = 1,
       n_iters: Optional[int] = 20,
       stepsize_controller: AbstractStepSizeController = ConstantStepSize(),
@@ -61,7 +61,7 @@ def make_pc_step(
 
     **Other arguments:**
 
-    - `solver`: Diffrax (ODE) solver to be used. Default is Euler.
+    - `ode_solver`: Diffrax ODE solver to be used. Default is Euler.
     - `dt`: Integration step size. Defaults to 1.
     - `n_iters`: Number of integration steps (20 as default).
     - `stepsize_controller`: diffrax controller for step size integration.
@@ -112,7 +112,7 @@ def make_pc_step(
         activities=activities,
         y=y,
         x=x,
-        solver=solver,
+        solver=ode_solver,
         n_iters=n_iters,
         stepsize_controller=stepsize_controller,
         dt=dt,
@@ -161,7 +161,7 @@ def make_hpc_step(
       opt_states: Tuple[OptState],
       y: ArrayLike,
       x: ArrayLike,
-      solver: AbstractSolver = Euler(),
+      ode_solver: AbstractSolver = Euler(),
       dt: float | int = 1,
       n_iters: Optional[int] = 20,
       stepsize_controller: AbstractStepSizeController = ConstantStepSize(),
@@ -197,7 +197,7 @@ def make_hpc_step(
 
     **Other arguments:**
 
-    - `solver`: Diffrax (ODE) solver to be used. Default is Euler.
+    - `ode_solver`: Diffrax ODE solver to be used. Default is Euler.
     - `dt`: Integration step size. Defaults to 1.
     - `n_iters`: Number of integration steps (20 as default).
     - `stepsize_controller`: diffrax controller for step size integration.
@@ -240,7 +240,7 @@ def make_hpc_step(
         activities=amort_activities[1:],
         y=y,
         x=x,
-        solver=solver,
+        solver=ode_solver,
         n_iters=n_iters,
         stepsize_controller=stepsize_controller,
         dt=dt,