Vectorize and remove for loop in sparse expvals #1596

josh146 · 2021-08-27T06:51:19Z

Context:

In #1551, support for computing the expectation values of Hamiltonians using sparse methods was directly added to default.qubit. However, two improvements were noticed:

The for loop over the Hamiltonian observables and coefficients could be replaced with vectorization
When not in backprop mode, the full sparse matrix could be computed directly.

Description of the Change:

The for loop discussed above has been removed and replaced with vectorization.
When non-backprop mode is detected, we simply re-use the same logic that is currently used for qml.SparseHamiltonian.

Benefits:

Vectorizing the for loop allows us to remove the casting of Hamiltonian coefficients to NumPy arrays, that we were performing as a bugfix.
qml.SparseHamiltonian and qml.Hamiltonian now share the same code for expvals when using parameter-shift.

Faster execution as opposed to backprop in master

Backprop:
  Using ExpvalCost  : 0.9697349071502686     (1x)
  expval(H), master : 0.2827181816101074   (3.4x)
  expval(H), this PR: 0.08908772468566895 (10.9x)

Faster execution as opposed to parameter-shift in master

Parameter-shift:
  Using ExpvalCost  : 6.549182176589966    (1x)
  expval(H), master : 2.6744325160980225 (2.4x)
  expval(H), this PR: 1.9294066429138184 (3.4x)

Possible Drawbacks:

While improvements to the parameter-shift pipeline in default.qubit are nice, we plan to transition to lightning.qubit by default, and likely restrict default.qubit to purely backprop going forward.
Is the SparseHamiltonian class deprecated? Should we remove it?

Related GitHub Issues: n/a

github-actions · 2021-08-27T06:51:41Z

Hello. You may have forgotten to update the changelog!
Please edit .github/CHANGELOG.md with:

A one-to-two sentence description of the change. You may include a small working example for new features.
A link back to this PR.
Your name (or GitHub username) in the contributors section.

mariaschuld · 2021-08-27T07:12:16Z

pennylane/devices/default_qubit.py

+                    * coo.data
+                    * qml.math.gather(self.state, coo.col)
+                )
+                c = qml.math.cast(qml.math.convert_like(coeff, product), "complex128")


I am such an idiot, when I tried to add this yesterday I must have converted to the wrong object - and the rewrite is indeed that simple.

Strange that the qchem tests fail!

Seems to be an unrelated issue, something to do with h5py 🤔

Yes I just saw...Feel free to tag me for a quick merge once things pass!

Sounds like big fun!

josh146 · 2021-08-27T07:14:15Z

pennylane/devices/default_qubit.py

-                    # todo: remove this hack that avoids errors when attempting to multiply
-                    # a nontrainable qml.tensor to a trainable Arraybox
-                    if isinstance(coeff, qml.numpy.tensor) and not coeff.requires_grad:


@mariaschuld this is really weird, but the rewrite allows this to be removed 🤔

mariaschuld · 2021-08-27T13:21:14Z

pennylane/devices/default_qubit.py

+                for op, coeff in zip(observable.ops, observable.data):
+
+                    # extract a scipy.sparse.coo_matrix representation of this Pauli word
+                    coo = qml.operation.Tensor(op).sparse_matrix(wires=self.wires)


@soranjh sorry, I just realised I never addressed your suggestion to rename. Do you have a better idea than coo?

I thought that coo might be confused with the coo sparse format. Maybe consider using mat or something similar, but it is not important at all.

mariaschuld · 2021-08-27T13:21:49Z

pennylane/devices/default_qubit.py

+                if observable.name == "Hamiltonian":
+                    Hmat = qml.utils.sparse_hamiltonian(observable, wires=self.wires)
+                elif observable.name == "SparseHamiltonian":
+                    Hmat = observable.matrix


mariaschuld · 2021-08-27T13:22:05Z

pennylane/utils.py

@@ -150,7 +150,7 @@ def sparse_hamiltonian(H, wires=None):
    n = len(wires)
    matrix = scipy.sparse.coo_matrix((2 ** n, 2 ** n), dtype="complex128")

-    coeffs = qml.math.toarray(H.coeffs)
+    coeffs = qml.math.toarray(H.data)


Good point!

mariaschuld

Very nice, thanks so much @josh146 . So we are back to about 15x speedup now? :)

(Approved, assuming the qchem issue gets fixed - but that is on master, right?)

josh146 · 2021-08-27T13:25:43Z

(Approved, assuming the qchem issue gets fixed - but that is on master, right?)

Yep, finally fixed this in #1597 🎉

soranjh · 2021-08-27T13:38:47Z

pennylane/devices/default_qubit.py

+
+            backprop_mode = not isinstance(self.state, np.ndarray)
+
+            if backprop_mode:


Is backprop the only method we can support with the new addition?

We still support parameter-shift with qml.expval(H), however we use the else: statement, which is more performant

Yes, I meant if backprop is the only method that works with the new procedure that Maria added which computes the sparse matrix for each term in the Hamiltonian.

Yep 👍 Parameter-shift defaults to computing the full sparse matrix first, and then only a single expval

codecov · 2021-08-27T13:40:05Z

Codecov Report

Merging #1596 (5b08e68) into master (65320c9) will increase coverage by 2.16%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master    #1596      +/-   ##
==========================================
+ Coverage   96.97%   99.13%   +2.16%     
==========================================
  Files         195      195              
  Lines       14103    14104       +1     
==========================================
+ Hits        13676    13982     +306     
+ Misses        427      122     -305

Impacted Files	Coverage Δ
pennylane/devices/default_qubit.py	`100.00% <100.00%> (ø)`
pennylane/utils.py	`98.25% <100.00%> (ø)`
pennylane/beta/devices/default_tensor.py	`96.93% <0.00%> (+1.70%)`	⬆️
pennylane/interfaces/batch/tensorflow.py	`100.00% <0.00%> (+2.17%)`	⬆️
pennylane/devices/default_qubit_tf.py	`92.00% <0.00%> (+2.66%)`	⬆️
pennylane/beta/devices/default_tensor_tf.py	`90.62% <0.00%> (+3.12%)`	⬆️
pennylane/interfaces/batch/torch.py	`100.00% <0.00%> (+3.27%)`	⬆️
pennylane/interfaces/torch.py	`100.00% <0.00%> (+3.29%)`	⬆️
pennylane/qnode.py	`98.47% <0.00%> (+3.43%)`	⬆️
pennylane/collections/qnode_collection.py	`100.00% <0.00%> (+3.50%)`	⬆️
... and 12 more

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 65320c9...5b08e68. Read the comment docs.

soranjh

Thanks Josh, looks good to me. I am not sure that the SparseHamiltonian class is needed anymore.

ixfoduap

Love it!

A beautiful example of great teamwork! I think the four of us contributed in different ways to enable this great UI and high performance 🚀

Vectorize and remove for loop in sparse expvals

80cab4f

mariaschuld reviewed Aug 27, 2021

View reviewed changes

more

e3ba19e

josh146 commented Aug 27, 2021

View reviewed changes

josh146 added 5 commits August 27, 2021 15:23

more

8ca8720

more

e01551b

black

1cfc169

bugfix

c03af62

fix

6425bf2

josh146 requested review from mariaschuld, soranjh and ixfoduap August 27, 2021 08:28

josh146 added the review-ready 👌 PRs which are ready for review by someone from the core team. label Aug 27, 2021

josh146 added 7 commits August 27, 2021 16:32

remove print

6947f91

try increasing h5py

cab5551

try again

b533ae1

fix

4bb180c

fix

dca05c4

pin numpy

8c63c89

more

c86797f

mariaschuld reviewed Aug 27, 2021

View reviewed changes

mariaschuld approved these changes Aug 27, 2021

View reviewed changes

revert qchem changes

f286163

Merge branch 'master' into vectorize-sparse

09a1ee8

soranjh reviewed Aug 27, 2021

View reviewed changes

soranjh approved these changes Aug 27, 2021

View reviewed changes

josh146 added 2 commits August 27, 2021 22:22

merge master

6ba4016

update

96959f2

ixfoduap approved these changes Aug 27, 2021

View reviewed changes

Merge branch 'master' into vectorize-sparse

5b08e68

josh146 merged commit 802c45c into master Aug 29, 2021

josh146 deleted the vectorize-sparse branch August 29, 2021 13:37

josh146 mentioned this pull request Aug 30, 2021

Add support for the new VQE workflows to the differentiable batch execution pipeline #1608

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Vectorize and remove for loop in sparse expvals #1596

Vectorize and remove for loop in sparse expvals #1596

josh146 commented Aug 27, 2021 •

edited

Loading

github-actions bot commented Aug 27, 2021

mariaschuld Aug 27, 2021

josh146 Aug 27, 2021

mariaschuld Aug 27, 2021

josh146 Aug 27, 2021

mariaschuld Aug 27, 2021

josh146 Aug 27, 2021

mariaschuld Aug 27, 2021

soranjh Aug 27, 2021

mariaschuld Aug 27, 2021

mariaschuld Aug 27, 2021

mariaschuld left a comment •

edited

Loading

josh146 commented Aug 27, 2021

soranjh Aug 27, 2021

josh146 Aug 27, 2021

soranjh Aug 27, 2021

josh146 Aug 27, 2021 •

edited

Loading

codecov bot commented Aug 27, 2021 •

edited

Loading

soranjh left a comment

ixfoduap left a comment


		backprop_mode = not isinstance(self.state, np.ndarray)

		if backprop_mode:

Vectorize and remove for loop in sparse expvals #1596

Vectorize and remove for loop in sparse expvals #1596

Conversation

josh146 commented Aug 27, 2021 • edited Loading

github-actions bot commented Aug 27, 2021

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

mariaschuld left a comment • edited Loading

Choose a reason for hiding this comment

josh146 commented Aug 27, 2021

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

josh146 Aug 27, 2021 • edited Loading

Choose a reason for hiding this comment

codecov bot commented Aug 27, 2021 • edited Loading

Codecov Report

soranjh left a comment

Choose a reason for hiding this comment

ixfoduap left a comment

Choose a reason for hiding this comment

josh146 commented Aug 27, 2021 •

edited

Loading

mariaschuld left a comment •

edited

Loading

josh146 Aug 27, 2021 •

edited

Loading

codecov bot commented Aug 27, 2021 •

edited

Loading