Merge branch 'main' into main

Author: stefan-k

README.md

@@ -17,7 +17,7 @@
   |
   <a href="https://argmin-rs.github.io/argmin/argmin/">Docs (main branch)</a>
   |
-  <a href="https://github.com/argmin-rs/argmin/tree/v0.5.0/examples">Examples (latest release)</a>
+  <a href="https://github.com/argmin-rs/argmin/tree/argmin-v0.8.1/argmin/examples">Examples (latest release)</a>
   |
   <a href="https://github.com/argmin-rs/argmin/tree/main/argmin/examples">Examples (main branch)</a>
 </p>

argmin-math/src/ndarray_m/inv.rs

@@ -22,6 +22,14 @@ macro_rules! make_inv {
                 Ok(<Self as Inverse>::inv(&self)?)
             }
         }
+
+        // inverse for scalars (1d solvers)
+        impl ArgminInv<$t> for $t {
+            #[inline]
+            fn inv(&self) -> Result<$t, Error> {
+                Ok(1.0 / self)
+            }
+        }
     };
 }
 
@@ -60,6 +68,16 @@ mod tests {
                     }
                 }
             }
+
+            item! {
+                #[test]
+                fn [<test_inv_scalar_ $t>]() {
+                    let a = 2.0;
+                    let target = 0.5;
+                    let res = <$t as ArgminInv<$t>>::inv(&a).unwrap();
+                    assert!(((res - target) as f64).abs() < 0.000001);
+                }
+            }
         };
     }
 

argmin/Cargo.toml

@@ -54,7 +54,7 @@ serde1 = ["serde", "serde_json", "rand/serde1", "bincode", "slog-json", "rand_xo
 _ndarrayl = ["argmin-math/ndarray_latest-serde", "argmin-math/_dev_linalg_latest"]
 _nalgebral = ["argmin-math/nalgebra_latest-serde"]
 # When adding new features, please consider adding them to either `full` (for users)
-# or `_full_dev` (only for local development, tesing and computing test coverage).
+# or `_full_dev` (only for local development, testing and computing test coverage).
 full = ["default", "slog-logger", "serde1", "ctrlc"]
 _full_dev = ["full", "_ndarrayl", "_nalgebral"]
 
@@ -129,6 +129,10 @@ required-features = ["argmin-math/nalgebra_latest-serde", "slog-logger"]
 name = "morethuente"
 required-features = ["slog-logger"]
 
+[[example]]
+name = "neldermead-cubic"
+required-features = ["slog-logger"]
+
 [[example]]
 name = "neldermead"
 required-features = ["argmin-math/ndarray_latest-serde", "slog-logger"]
@@ -177,6 +181,10 @@ required-features = ["argmin-math/ndarray_latest-serde", "slog-logger"]
 name = "steepestdescent"
 required-features = ["slog-logger"]
 
+[[example]]
+name = "steepestdescent_manifold"
+required-features = ["slog-logger"]
+
 [[example]]
 name = "trustregion_nd"
 required-features = ["argmin-math/ndarray_latest-serde", "slog-logger"]

argmin/examples/neldermead-cubic.rs

@@ -0,0 +1,127 @@
+// Copyright 2018-2022 argmin developers
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://apache.org/licenses/LICENSE-2.0> or the MIT license <LICENSE-MIT or
+// http://opensource.org/licenses/MIT>, at your option. This file may not be
+// copied, modified, or distributed except according to those terms.
+
+//! A (hopefully) simple example of using Nelder-Mead to find the roots of a
+//! cubic polynomial.
+//!
+//! You can run this example with:
+//! `cargo run --example neldermead-cubic --features slog-logger`
+
+use argmin::core::observers::{ObserverMode, SlogLogger};
+use argmin::core::{CostFunction, Error, Executor, State};
+use argmin::solver::neldermead::NelderMead;
+
+/// Coefficients describing a cubic `f(x) = ax^3 + bx^2 + cx + d`
+#[derive(Clone, Copy)]
+struct Cubic {
+    /// Coefficient of the `x^3` term
+    a: f64,
+    /// Coefficient of the `x^2` term
+    b: f64,
+    /// Coefficient of the `x` term
+    c: f64,
+    /// Coefficient of the `x^0` term
+    d: f64,
+}
+
+impl Cubic {
+    /// Evaluate the cubic at `x`.
+    fn eval(self, x: f64) -> f64 {
+        self.a * x.powi(3) + self.b * x.powi(2) + self.c * x + self.d
+    }
+}
+
+impl CostFunction for Cubic {
+    type Param = f64;
+    type Output = f64;
+
+    fn cost(&self, p: &Self::Param) -> Result<Self::Output, Error> {
+        // The cost function is the evaluation of the polynomial with our
+        // parameter, squared. The parameter is a guess of `x`, and the
+        // objective is to minimize `x` (i.e. find a polynomial root). The
+        // square value can be considered an error. We want the error to (1)
+        // always be positive and (2) bigger the further it is from a polynomial
+        // root.
+        Ok(self.eval(*p).powi(2))
+    }
+}
+
+fn run() -> Result<(), Error> {
+    // Define the cost function. This needs to be something with an
+    // implementation of `CostFunction`; in this case, the impl is right
+    // above. Here, our cubic is `(x-2)(x+2)(x-5)`; see
+    // <https://www.wolframalpha.com/input?i=%28x-2%29%28x%2B2%29%28x-5%29> for
+    // more info.
+    let cost = Cubic {
+        a: 1.0,
+        b: -5.0,
+        c: -4.0,
+        d: 20.0,
+    };
+
+    // Let's find a root of the cubic (+5).
+    {
+        // Set up solver -- note that the proper choice of the vertices is very
+        // important! This example should find 5, because our vertices are 6 and 7.
+        let solver = NelderMead::new(vec![6.0, 7.0]).with_sd_tolerance(0.0001)?;
+
+        // Run solver
+        let res = Executor::new(cost, solver)
+            .configure(|state| state.max_iters(100))
+            .add_observer(SlogLogger::term(), ObserverMode::Always)
+            .run()?;
+
+        // Wait a second (lets the logger flush everything before printing again)
+        std::thread::sleep(std::time::Duration::from_secs(1));
+
+        // Print result
+        println!(
+            "Polynomial root: {}",
+            res.state.get_best_param().expect("Found a root")
+        );
+    }
+
+    // Now find -2.
+    {
+        let solver = NelderMead::new(vec![-3.0, -4.0]).with_sd_tolerance(0.0001)?;
+        let res = Executor::new(cost, solver)
+            .configure(|state| state.max_iters(100))
+            .add_observer(SlogLogger::term(), ObserverMode::Always)
+            .run()?;
+        std::thread::sleep(std::time::Duration::from_secs(1));
+        println!("{res}");
+        println!(
+            "Polynomial root: {}",
+            res.state.get_best_param().expect("Found a root")
+        );
+    }
+
+    // This example will find +2, even though it might look like we're trying to
+    // find +5.
+    {
+        let solver = NelderMead::new(vec![4.0, 6.0]).with_sd_tolerance(0.0001)?;
+        let res = Executor::new(cost, solver)
+            .configure(|state| state.max_iters(100))
+            .add_observer(SlogLogger::term(), ObserverMode::Always)
+            .run()?;
+        std::thread::sleep(std::time::Duration::from_secs(1));
+        println!("{res}");
+        println!(
+            "Polynomial root: {}",
+            res.state.get_best_param().expect("Found a root")
+        );
+    }
+
+    Ok(())
+}
+
+fn main() {
+    if let Err(ref e) = run() {
+        println!("{e}");
+        std::process::exit(1);
+    }
+}

argmin/examples/steepestdescent.rs

@@ -38,7 +38,7 @@ impl Gradient for Rosenbrock {
 }
 
 fn run() -> Result<(), Error> {
-    // Define cost function (must implement `ArgminOperator`)
+    // Define cost function (must implement `CostFunction` and `Gradient`)
     let cost = Rosenbrock { a: 1.0, b: 100.0 };
 
     // Define initial parameter vector

argmin/examples/steepestdescent_manifold.rs

@@ -0,0 +1,94 @@
+// Copyright 2018-2022 argmin developers
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://apache.org/licenses/LICENSE-2.0> or the MIT license <LICENSE-MIT or
+// http://opensource.org/licenses/MIT>, at your option. This file may not be
+// copied, modified, or distributed except according to those terms.
+
+#![allow(unused_imports)]
+
+use argmin::core::observers::{ObserverMode, SlogLogger};
+use argmin::core::{CostFunction, Error, Executor, Gradient};
+use argmin::solver::gradientdescent::SteepestDescent;
+use argmin::solver::linesearch::condition::{ArmijoCondition, LineSearchCondition};
+use argmin::solver::linesearch::BacktrackingLineSearch;
+use argmin_math::ArgminScaledAdd;
+
+use serde::{Deserialize, Serialize};
+
+#[derive(Clone, Copy, Debug)]
+struct ClosestPointOnCircle {
+    x: f64,
+    y: f64,
+}
+
+#[derive(Clone, Copy, Debug, Serialize, Deserialize)]
+struct CirclePoint {
+    angle: f64,
+}
+
+impl CostFunction for ClosestPointOnCircle {
+    type Param = CirclePoint;
+    type Output = f64;
+
+    fn cost(&self, p: &Self::Param) -> Result<Self::Output, Error> {
+        let x_circ = p.angle.cos();
+        let y_circ = p.angle.sin();
+        let x_diff = x_circ - self.x;
+        let y_diff = y_circ - self.y;
+        Ok(x_diff.powi(2) + y_diff.powi(2))
+    }
+}
+
+impl Gradient for ClosestPointOnCircle {
+    type Param = CirclePoint;
+    type Gradient = f64;
+
+    fn gradient(&self, p: &Self::Param) -> Result<Self::Gradient, Error> {
+        Ok(2.0 * (p.angle.cos() - self.x) * (-p.angle.sin())
+            + 2.0 * (p.angle.sin() - self.y) * p.angle.cos())
+    }
+}
+
+impl ArgminScaledAdd<f64, f64, CirclePoint> for CirclePoint {
+    fn scaled_add(&self, alpha: &f64, delta: &f64) -> Self {
+        CirclePoint {
+            angle: self.angle + alpha * delta,
+        }
+    }
+}
+
+fn run() -> Result<(), Error> {
+    // Define cost function (must implement `CostFunction` and `Gradient`)
+    let cost = ClosestPointOnCircle { x: 1.0, y: 1.0 };
+
+    // Define initial parameter vector
+    let init_param = CirclePoint { angle: 0.0 };
+
+    // Pick a line search.
+    let cond = ArmijoCondition::new(0.5)?;
+    let linesearch = BacktrackingLineSearch::new(cond);
+
+    // Set up solver
+    let solver = SteepestDescent::new(linesearch);
+
+    // Run solver
+    let res = Executor::new(cost, solver)
+        .configure(|state| state.param(init_param).max_iters(10))
+        .add_observer(SlogLogger::term(), ObserverMode::Always)
+        .run()?;
+
+    // Wait a second (lets the logger flush everything first)
+    std::thread::sleep(std::time::Duration::from_secs(1));
+
+    // print result
+    println!("{res}");
+    Ok(())
+}
+
+fn main() {
+    if let Err(ref e) = run() {
+        println!("{e}");
+        std::process::exit(1);
+    }
+}

argmin/src/solver/conjugategradient/cg.rs

@@ -247,10 +247,10 @@ mod tests {
 
         assert_relative_eq!(b[0], 1.0, epsilon = f64::EPSILON);
         assert_relative_eq!(b[1], 2.0, epsilon = f64::EPSILON);
-        let r0 = vec![2.0f64, 2.0];
+        let r0 = [2.0f64, 2.0];
         assert_relative_eq!(r0[0], r.as_ref().unwrap()[0], epsilon = f64::EPSILON);
         assert_relative_eq!(r0[1], r.as_ref().unwrap()[1], epsilon = f64::EPSILON);
-        let pp = vec![-2.0f64, -2.0];
+        let pp = [-2.0f64, -2.0];
         assert_relative_eq!(pp[0], p.as_ref().unwrap()[0], epsilon = f64::EPSILON);
         assert_relative_eq!(pp[1], p.as_ref().unwrap()[1], epsilon = f64::EPSILON);
         assert_relative_eq!(rtr, 8.0, epsilon = f64::EPSILON);

argmin/src/solver/gradientdescent/steepestdescent.rs

@@ -7,7 +7,7 @@
 
 use crate::core::{
     ArgminFloat, CostFunction, DeserializeOwnedAlias, Error, Executor, Gradient, IterState,
-    LineSearch, OptimizationResult, Problem, SerializeAlias, Solver, KV,
+    LineSearch, OptimizationResult, Problem, SerializeAlias, Solver, State, KV,
 };
 use argmin_math::ArgminMul;
 #[cfg(feature = "serde1")]
@@ -54,24 +54,27 @@ impl<O, L, P, G, F> Solver<O, IterState<P, G, (), (), (), F>> for SteepestDescen
 where
     O: CostFunction<Param = P, Output = F> + Gradient<Param = P, Gradient = G>,
     P: Clone + SerializeAlias + DeserializeOwnedAlias,
-    G: Clone + SerializeAlias + DeserializeOwnedAlias + ArgminMul<F, P>,
-    L: Clone + LineSearch<P, F> + Solver<O, IterState<P, G, (), (), (), F>>,
+    G: Clone + SerializeAlias + DeserializeOwnedAlias + ArgminMul<F, G>,
+    L: Clone + LineSearch<G, F> + Solver<O, IterState<P, G, (), (), (), F>>,
     F: ArgminFloat,
 {
     const NAME: &'static str = "Steepest Descent";
 
     fn next_iter(
         &mut self,
         problem: &mut Problem<O>,
-        mut state: IterState<P, G, (), (), (), F>,
+        state: IterState<P, G, (), (), (), F>,
     ) -> Result<(IterState<P, G, (), (), (), F>, Option<KV>), Error> {
-        let param_new = state.take_param().ok_or_else(argmin_error_closure!(
-            NotInitialized,
-            concat!(
-                "`SteepestDescent` requires an initial parameter vector. ",
-                "Please provide an initial guess via `Executor`s `configure` method."
-            )
-        ))?;
+        let param_new = state
+            .get_param()
+            .ok_or_else(argmin_error_closure!(
+                NotInitialized,
+                concat!(
+                    "`SteepestDescent` requires an initial parameter vector. ",
+                    "Please provide an initial guess via `Executor`s `configure` method."
+                )
+            ))?
+            .clone();
         let new_cost = problem.cost(&param_new)?;
         let new_grad = problem.gradient(&param_new)?;
 
@@ -153,6 +156,20 @@ mod tests {
         );
     }
 
+    #[test]
+    fn test_next_iter_prev_param_not_erased() {
+        let linesearch: BacktrackingLineSearch<Vec<f64>, Vec<f64>, ArmijoCondition<f64>, f64> =
+            BacktrackingLineSearch::new(ArmijoCondition::new(0.2).unwrap());
+        let mut sd = SteepestDescent::new(linesearch);
+        let (state, _kv) = sd
+            .next_iter(
+                &mut Problem::new(TestProblem::new()),
+                IterState::new().param(vec![1.0, 2.0]),
+            )
+            .unwrap();
+        state.prev_param.unwrap();
+    }
+
     #[test]
     fn test_next_iter_regression() {
         struct SDProblem {}