method to generate an initial guess. Fixes #354 #365

Peter230655 · 2025-02-18T16:31:03Z

Is it o.k. to be a method, or should I make it a property?

opty/direct_collocation.py

moorepants · 2025-02-18T19:46:11Z

Is there a good choice of one of the existing examples to add this to? You can show how it is used to the user.

Peter230655 · 2025-02-18T19:49:33Z

Is there a good choice of one of the existing examples to add this to? You can show how it is used to the user.

You mean in examples-gallery?
If yes, surely I will find an example.
Would you have to merge this first, or would I include the example I find in this PR #365?

moorepants · 2025-02-18T20:11:25Z

You can include the change to the example here.

moorepants · 2025-02-19T03:25:20Z

opty/direct_collocation.py

+
+        Returns
+        -------
+        initial_guess : (n*N + q*N + r + s)-ndarray


Suggested change

initial_guess : (n*N + q*N + r + s)-ndarray

initial_guess : ndarray, shape(n*N + q*N + r + s,)

opty/tests/test_direct_collocation.py

moorepants · 2025-02-19T03:40:19Z

opty/tests/test_direct_collocation.py

+        ux: (-1.0, 1.0),
+        y: (-4.0, 4.0),
+        uy: (-1.0, 1.0),
+    }


Easier to compare if placed here:

moorepants · 2025-02-19T03:43:17Z

opty/tests/test_direct_collocation.py

+        x.func(t1) - 1.5*b1,
+        y.func(t1) - 4.0,
+        ux.func(t1) - 1.0+b2,
+        uy.func(t1) - 4.5,


In this case I would expect uy to go from 0.0 to 4.5 to 0.0., but you set it to 4.5 to 4.5 to 4.5.

i.e. if not specified, the guess should be at zero

but maybe this more a design question. if there is only one instance constraint, what assumption to we make about any other values of time?

but maybe this more a design question. if there is only one instance constraint, what assumption to we make about any other values of time?

My assumption was that if there is only one instance constraint for a state variable it will have this value at all times. I thought this was the most reasonable one.

moorepants · 2025-02-19T04:12:22Z

opty/tests/test_direct_collocation.py

+
+    # A2: np.inf, -np.inf in bounds
+    bounds[a1] = (-np.inf, 10.0)
+    bounds[a2] = (-10.0, np.inf)


a1 and a2 are parameters, so it wouldn't affect the initial guess, right? i get same guess as first test:

a1, a2 are unknown parameters. I set their initial value like this: if there is a bound {a1: (s, t)} I set it at (s + t) /2.
If one of s or t is np.inf I set it to the other (finite) value.
I did not consider bounds which have np.inf on both ends, as they do not make sense in my opinion.

Parameters do not show in the plots.

moorepants · 2025-02-19T04:13:24Z

opty/tests/test_direct_collocation.py

+
+        x.func(tf) + 1.0,
+        ux.func(tf) + 5.0,
+    )


moorepants · 2025-02-19T04:14:04Z

opty/tests/test_direct_collocation.py

+        y: (-4.0, 4.0),
+        uy: (-1.0, 1.0),
+        h: (1.0, 2.0),
+    }


moorepants · 2025-02-19T04:14:36Z

opty/tests/test_direct_collocation.py

+
+    # B2: np.inf, -np.inf in bounds
+    bounds[a1] = (-np.inf, 10.0)
+    bounds[a2] = (-10.0, np.inf)


moorepants · 2025-02-19T04:15:13Z

opty/tests/test_direct_collocation.py

+    initial_guess = prob.create_linear_initial_guess()
+    np.testing.assert_allclose(initial_guess, expected_guess)
+
+    # B3: no bounds


These 'peak' plots happen like this If there is no bound for h, h is assumed to be zero.
So, all times ti = integer * h = 0.
The plots for u1, u2 should have a dot at (0 | 0).

I guess h should never be zero, right? So maybe we should always assume a value for it...but what value?

I guess h should never be zero, right? So maybe we should always assume a value for it...but what value?

I have found, that one should always bound h to be non-negative, to avoid senseless 'solutions'.
If there is a bound, say, {h: (0, 0.5)} I set h = 0.25.
So, maybe this would not be a bad default value?

Without the user giving some idea of what the total duration of the simulation might be, it's hard (or impossible) to pick a starting number. If someone is simulating molecular dynamics, maybe their duration is a tenth of a second. I guess we could default to some very small value of h. Or this function could take an extract kwarg: expected_duration=10.0 where the user can give us a bit more info.

The expected duration could default to 1.0 second (just as a simple number).

The expected duration could default to 1.0 second (just as a simple number).

I will do so.

So you like the idea of the kwarg to give a clue on duration?

So you like the idea of the kwarg to give a clue on duration?

I like it! I did not think of it, as you normally do not like them.
Default would be false?

I don't dislike kwargs.

Default would be 1.0.

moorepants · 2025-02-19T04:15:39Z

opty/tests/test_direct_collocation.py

+
+        x.func(tf0) + 1.0,
+        ux.func(tf0) + 5.0,
+    )


moorepants · 2025-02-19T04:17:34Z

I added a commit, so pull this branch before you edit more. You can now do:

In [1]: from opty.tests.test_direct_collocation import test_linear_initial_guess

In [2]: test_linear_initial_guess(plot=True)

In [3]: import matplotlib.pyplot as plt

In [4]: plt.show()

to quickly plot your guesses.

Peter230655 · 2025-02-19T06:10:20Z

I added a commit, so pull this branch before you edit more. You can now do:
In [1]: from opty.tests.test_direct_collocation import test_linear_initial_guess

In [2]: test_linear_initial_guess(plot=True)

In [3]: import matplotlib.pyplot as plt

In [4]: plt.show()
to quickly plot your guesses.

Not sure I totally understand (you know my computer skills...or lack thereof...)
So, I first update my remote branch
Then I update my local master
Then I go to the branch where this is, I called it linear-initial-guess
Then I do your steps?

moorepants · 2025-02-19T06:13:44Z

All your do is git pull origin linear-initial-guess. Then you are fine to continue.

Peter230655 · 2025-02-19T07:06:50Z

All your do is git pull origin linear-initial-guess. Then you are fine to continue.

Thanks!
I would do this when I am in the correct local branch, that is in create-linear-guess ?

moorepants · 2025-02-19T07:10:28Z

The branch associated with this PR is named linear-initial-guess:

Peter230655 · 2025-02-19T08:57:41Z

Just found a MAJOR plunder! Have to go back to the drawing board.

Peter230655 · 2025-02-19T13:44:55Z

corrected (hopefully) a problem with instance constraints of the form x(ti) - 0.0
set variable time intervall = 1 / (num_nodes - 1) if there are no bounds for it, so duration is 1.
set a kwarg: plot with default = False. If set to True, the initial guess will be plotted
corrected the test function accordingly
added two examples with this method: plot_pendulum_swing_up_fixed_duration and plot_sliding_block. Interestingly, with plot_pendulum_swing_up_variable_duration the linear initial guess did not lead to convergence, while the all zero initial guess works fine.

NB:
This shows once again, how important a good initial guess is.
Maybe a random initial guess could be like this:

start with a linear interpolation.
at each node take a random normally distributed value, with mean = value of the linear interpolation, sigma and a 'strength' to be user-settable.

Peter230655 added 2 commits February 18, 2025 17:29

method to generate an initial guess. Fixes csu-hmc#354

3b87de1

corrected length of underline

eaa52ed

moorepants reviewed Feb 18, 2025

View reviewed changes

opty/direct_collocation.py Outdated Show resolved Hide resolved

moorepants reviewed Feb 18, 2025

View reviewed changes

opty/direct_collocation.py Outdated Show resolved Hide resolved

moorepants reviewed Feb 18, 2025

View reviewed changes

opty/direct_collocation.py Outdated Show resolved Hide resolved

moorepants reviewed Feb 18, 2025

View reviewed changes

opty/direct_collocation.py Outdated Show resolved Hide resolved

moorepants reviewed Feb 18, 2025

View reviewed changes

opty/direct_collocation.py Show resolved Hide resolved

corrected as per suggestions

9d56ba3

moorepants reviewed Feb 19, 2025

View reviewed changes

opty/tests/test_direct_collocation.py Show resolved Hide resolved

moorepants reviewed Feb 19, 2025

View reviewed changes

opty/tests/test_direct_collocation.py

ux: (-1.0, 1.0),

y: (-4.0, 4.0),

uy: (-1.0, 1.0),

}

Copy link

Member

moorepants Feb 19, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Easier to compare if placed here:

moorepants reviewed Feb 19, 2025

View reviewed changes

opty/tests/test_direct_collocation.py

x.func(tf) + 1.0,

ux.func(tf) + 5.0,

)

Copy link

Member

moorepants Feb 19, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

moorepants reviewed Feb 19, 2025

View reviewed changes

opty/tests/test_direct_collocation.py

y: (-4.0, 4.0),

uy: (-1.0, 1.0),

h: (1.0, 2.0),

}

Copy link

Member

moorepants Feb 19, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

moorepants reviewed Feb 19, 2025

View reviewed changes

opty/tests/test_direct_collocation.py

# B2: np.inf, -np.inf in bounds

bounds[a1] = (-np.inf, 10.0)

bounds[a2] = (-10.0, np.inf)

Copy link

Member

moorepants Feb 19, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

moorepants reviewed Feb 19, 2025

View reviewed changes

opty/tests/test_direct_collocation.py

x.func(tf0) + 1.0,

ux.func(tf0) + 5.0,

)

Copy link

Member

moorepants Feb 19, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Add plotting to initial guess test function.

1a17afb

corrected error with instance = 0, added two examples

f2235d6

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

method to generate an initial guess. Fixes #354 #365

method to generate an initial guess. Fixes #354 #365

Peter230655 commented Feb 18, 2025

moorepants commented Feb 18, 2025

Peter230655 commented Feb 18, 2025

moorepants commented Feb 18, 2025

moorepants Feb 19, 2025

moorepants Feb 19, 2025

moorepants Feb 19, 2025

moorepants Feb 19, 2025

moorepants Feb 19, 2025

Peter230655 Feb 19, 2025

moorepants Feb 19, 2025

Peter230655 Feb 19, 2025

moorepants Feb 19, 2025

moorepants Feb 19, 2025

moorepants Feb 19, 2025

moorepants Feb 19, 2025

Peter230655 Feb 19, 2025

moorepants Feb 19, 2025

Peter230655 Feb 19, 2025

moorepants Feb 19, 2025

moorepants Feb 19, 2025

Peter230655 Feb 19, 2025

moorepants Feb 19, 2025

Peter230655 Feb 19, 2025

moorepants Feb 19, 2025

moorepants Feb 19, 2025

moorepants commented Feb 19, 2025

Peter230655 commented Feb 19, 2025

moorepants commented Feb 19, 2025

Peter230655 commented Feb 19, 2025

moorepants commented Feb 19, 2025

Peter230655 commented Feb 19, 2025

Peter230655 commented Feb 19, 2025 •

edited

Loading

	initial_guess : (nN + qN + r + s)-ndarray
	initial_guess : ndarray, shape(nN + qN + r + s,)

method to generate an initial guess. Fixes #354 #365

Are you sure you want to change the base?

method to generate an initial guess. Fixes #354 #365

Conversation

Peter230655 commented Feb 18, 2025

moorepants commented Feb 18, 2025

Peter230655 commented Feb 18, 2025

moorepants commented Feb 18, 2025

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

moorepants commented Feb 19, 2025

Peter230655 commented Feb 19, 2025

moorepants commented Feb 19, 2025

Peter230655 commented Feb 19, 2025

moorepants commented Feb 19, 2025

Peter230655 commented Feb 19, 2025

Peter230655 commented Feb 19, 2025 • edited Loading

Peter230655 commented Feb 19, 2025 •

edited

Loading