update function tests

This commit is contained in:
Simon Gardling 2022-05-16 11:17:06 -04:00
parent c0188ccaee
commit d620f41b28

View File

@ -559,8 +559,6 @@ impl FunctionEntry {
self.calculate(true, true, &settings);
assert!(!self.back_data.is_empty());
assert_eq!(self.back_data.len(), settings.plot_width + 1);
let back_vec_tuple = self.back_data.to_tuple();
assert_eq!(back_vec_tuple, back_target);
assert!(self.integral);
assert!(self.derivative);
@ -570,9 +568,36 @@ impl FunctionEntry {
assert!(!self.derivative_data.is_empty());
assert!(self.integral_data.is_some());
assert_eq!(self.derivative_data.to_tuple(), derivative_target);
assert_eq!(self.integral_data.clone().unwrap().1, area_target);
let a = self.derivative_data.to_tuple();
assert_eq!(a.len(), derivative_target.len());
for i in 0..a.len() {
if !emath::almost_equal(a[i].0 as f32, derivative_target[i].0 as f32, f32::EPSILON)
| !emath::almost_equal(
a[i].1 as f32,
derivative_target[i].1 as f32,
f32::EPSILON,
) {
panic!("Expected: {:?}\nGot: {:?}", a, derivative_target);
}
}
let a_1 = self.back_data.to_tuple();
assert_eq!(a_1.len(), back_target.len());
assert_eq!(a.len(), back_target.len());
for i in 0..a.len() {
if !emath::almost_equal(a_1[i].0 as f32, back_target[i].0 as f32, f32::EPSILON)
| !emath::almost_equal(a_1[i].1 as f32, back_target[i].1 as f32, f32::EPSILON)
{
panic!("Expected: {:?}\nGot: {:?}", a_1, back_target);
}
}
}
{
@ -580,41 +605,59 @@ impl FunctionEntry {
settings.max_x += 1.0;
self.calculate(true, true, &settings);
assert_eq!(
self.derivative_data
.to_tuple()
.iter()
.take(6)
.cloned()
.map(|(a, b)| (a.round(), b.round())) // round to account for floating point issues
.collect::<Vec<(f64, f64)>>(),
derivative_target
.iter()
.rev()
.take(6)
.rev()
.cloned()
.map(|(a, b)| (a.round(), b.round())) // round to account for floating point issues
.collect::<Vec<(f64, f64)>>()
);
let a = self
.derivative_data
.to_tuple()
.iter()
.take(6)
.cloned()
.collect::<Vec<(f64, f64)>>();
assert_eq!(
self.back_data
.to_tuple()
.iter()
.take(6)
.cloned()
.map(|(a, b)| (a.round(), b.round())) // round to account for floating point issues
.collect::<Vec<(f64, f64)>>(),
back_target
.iter()
.rev()
.take(6)
.rev()
.cloned()
.map(|(a, b)| (a.round(), b.round())) // round to account for floating point issues
.collect::<Vec<(f64, f64)>>()
);
let b = derivative_target
.iter()
.rev()
.take(6)
.rev()
.cloned()
.collect::<Vec<(f64, f64)>>();
assert_eq!(a.len(), b.len());
for i in 0..a.len() {
if !emath::almost_equal(a[i].0 as f32, b[i].0 as f32, f32::EPSILON)
| !emath::almost_equal(a[i].1 as f32, b[i].1 as f32, f32::EPSILON)
{
panic!("Expected: {:?}\nGot: {:?}", a, b);
}
}
let a_1 = self
.back_data
.to_tuple()
.iter()
.take(6)
.cloned()
.collect::<Vec<(f64, f64)>>();
let b_1 = back_target
.iter()
.rev()
.take(6)
.rev()
.cloned()
.collect::<Vec<(f64, f64)>>();
assert_eq!(a_1.len(), b_1.len());
assert_eq!(a.len(), b_1.len());
for i in 0..a.len() {
if !emath::almost_equal(a_1[i].0 as f32, b_1[i].0 as f32, f32::EPSILON)
| !emath::almost_equal(a_1[i].1 as f32, b_1[i].1 as f32, f32::EPSILON)
{
panic!("Expected: {:?}\nGot: {:?}", a_1, b_1);
}
}
}
{
@ -622,41 +665,59 @@ impl FunctionEntry {
settings.max_x -= 2.0;
self.calculate(true, true, &settings);
assert_eq!(
self.derivative_data
.to_tuple()
.iter()
.rev()
.take(6)
.rev()
.cloned()
.map(|(a, b)| (a.round(), b.round())) // round to account for floating point issues
.collect::<Vec<(f64, f64)>>(),
derivative_target
.iter()
.take(6)
.cloned()
.map(|(a, b)| (a.round(), b.round())) // round to account for floating point issues
.collect::<Vec<(f64, f64)>>()
);
let a = self
.derivative_data
.to_tuple()
.iter()
.rev()
.take(6)
.rev()
.cloned()
.collect::<Vec<(f64, f64)>>();
assert_eq!(
self.back_data
.to_tuple()
.iter()
.rev()
.take(6)
.rev()
.cloned()
.map(|(a, b)| (a.round(), b.round())) // round to account for floating point issues
.collect::<Vec<(f64, f64)>>(),
back_target
.iter()
.take(6)
.cloned()
.map(|(a, b)| (a.round(), b.round())) // round to account for floating point issues
.collect::<Vec<(f64, f64)>>()
);
let b = derivative_target
.iter()
.take(6)
.cloned()
.collect::<Vec<(f64, f64)>>();
assert_eq!(a.len(), b.len());
for i in 0..a.len() {
if !emath::almost_equal(a[i].0 as f32, b[i].0 as f32, f32::EPSILON)
| !emath::almost_equal(a[i].1 as f32, b[i].1 as f32, f32::EPSILON)
{
panic!("Expected: {:?}\nGot: {:?}", a, b);
}
}
let a_1 = self
.back_data
.to_tuple()
.iter()
.rev()
.take(6)
.rev()
.cloned()
.collect::<Vec<(f64, f64)>>();
let b_1 = back_target
.iter()
.take(6)
.cloned()
.collect::<Vec<(f64, f64)>>();
assert_eq!(a_1.len(), b_1.len());
assert_eq!(a.len(), b_1.len());
for i in 0..a.len() {
if !emath::almost_equal(a_1[i].0 as f32, b_1[i].0 as f32, f32::EPSILON)
| !emath::almost_equal(a_1[i].1 as f32, b_1[i].1 as f32, f32::EPSILON)
{
panic!("Expected: {:?}\nGot: {:?}", a_1, b_1);
}
}
}
{