Confusion regarding the output of circuit.best_method_str(dev, circuit.interface)

r_taeja · August 20, 2024, 5:19pm

Hello everyone!

I’m trying out the following model, and specified diff_method = "parameter-shift". I was going through some of the discussions, and found one that was talking about how you can print out the differentiation method being used under the hood when you don’t specify one. Following that, I tried out circuit.best_method_str(dev, circuit.interface) just to see what it gives (which I had assumed would give param-shift since I had specified the diff_method already). But, I got backprop.

So, does that mean that circuit.best_method_str(dev, circuit.interface) just returns the ‘best’ differentiation method, which just happens to be backprop in this case, but not the differentiation method that is being used under the hood? Or, does it mean that backprop was used instead of parameter shift under the hood?

If it is the former, is there any other way to figure out what differentiation method is being used under the hood? Or in case of the latter, is there a way to fix the diff_method to only take parameter_shift?

Here’s the model I was running for context :

import pennylane as qml
from pennylane import numpy as np

#Setting up qubit
dev = qml.device('default.qubit', wires = 2)

#Setting up quantum circuit
@qml.qnode(dev)
def circuit(params, diff_method = "parameter-shift"):
    qml.RX(params[0], wires = 0)
    qml.RY(params[1], wires = 1)
    qml.CNOT(wires = [0, 1])
    return qml.probs(wires = (0, 1))

#Cost function
def cost_fn(params):
  probs = circuit(params)
  return np.mean(np.abs(probs - target_probs))

#Setting up optimizer to Gradient Descent
opt = qml.GradientDescentOptimizer(0.3)

#Target Probability distribution
target_probs = [0.2, 0.4, 0.2, 0.2]

#Parameters for the quantum circuit
params = np.array([np.pi/4, np.pi/4], requires_grad=True)

#No. of steps
steps = 500

#Running the model
for _ in range(steps):
  params, l = opt.step_and_cost(cost_fn, params)

print(circuit.best_method_str(dev, circuit.interface))

The output of print(circuit.best_method_str(dev, circuit.interface)) is :

backprop

Here’s the output of qml.about() :

Name: PennyLane
Version: 0.37.0
Summary: PennyLane is a cross-platform Python library for quantum computing, quantum machine learning, and quantum chemistry. Train a quantum computer the same way as a neural network.
Home-page: https://github.com/PennyLaneAI/pennylane
Author: 
Author-email: 
License: Apache License 2.0
Location: /usr/local/lib/python3.10/dist-packages
Requires: appdirs, autograd, autoray, cachetools, networkx, numpy, packaging, pennylane-lightning, requests, rustworkx, scipy, semantic-version, toml, typing-extensions
Required-by: PennyLane_Lightning

Platform info:           Linux-6.1.85+-x86_64-with-glibc2.35
Python version:          3.10.12
Numpy version:           1.26.4
Scipy version:           1.13.1
Installed devices:
- default.clifford (PennyLane-0.37.0)
- default.gaussian (PennyLane-0.37.0)
- default.mixed (PennyLane-0.37.0)
- default.qubit (PennyLane-0.37.0)
- default.qubit.autograd (PennyLane-0.37.0)
- default.qubit.jax (PennyLane-0.37.0)
- default.qubit.legacy (PennyLane-0.37.0)
- default.qubit.tf (PennyLane-0.37.0)
- default.qubit.torch (PennyLane-0.37.0)
- default.qutrit (PennyLane-0.37.0)
- default.qutrit.mixed (PennyLane-0.37.0)
- default.tensor (PennyLane-0.37.0)
- null.qubit (PennyLane-0.37.0)
- lightning.qubit (PennyLane_Lightning-0.37.0)

I’d appreciate any insight! Thank you so much for reading and the help!

CatalinaAlbornoz · August 21, 2024, 10:42pm

Hi @r_taeja , thank you for your question!

Yes, best_method_str just returns the best method, not the one used.

You might want to use qml.specs instead to get the actual diff method being used.

r_taeja · August 22, 2024, 5:48am

Hey @CatalinaAlbornoz ! Thank you so much for the reply!

I tried using qml.specs() instead of circuit.best_method_str() as follows :

import pennylane as qml
from pennylane import numpy as np

#Setting up qubit
dev = qml.device('default.qubit', wires = 2)

#Setting up quantum circuit
@qml.qnode(dev)
def circuit(params, diff_method = "parameter-shift"):
    qml.RX(params[0], wires = 0)
    qml.RY(params[1], wires = 1)
    qml.CNOT(wires = [0, 1])
    return qml.probs(wires = (0, 1))

#Cost function
def cost_fn(params):
  probs = circuit(params)
  return np.mean(np.abs(probs - target_probs))

#Setting up optimizer to Gradient Descent
opt = qml.GradientDescentOptimizer(0.3)

#Target Probability distribution
target_probs = [0.2, 0.4, 0.2, 0.2]

#Parameters for the quantum circuit
params = np.array([np.pi/4, np.pi/4], requires_grad=True)

#No. of steps
steps = 500

#Running the model
for _ in range(steps):
  params, l = opt.step_and_cost(cost_fn, params)

qml.specs(circuit)(params, diff_method = 'parameter-shift')

and here’s what I got :

{'resources': Resources(num_wires=2, num_gates=3, gate_types=defaultdict(<class 'int'>, {'RX': 1, 'RY': 1, 'CNOT': 1}), gate_sizes=defaultdict(<class 'int'>, {1: 2, 2: 1}), depth=2, shots=Shots(total_shots=None, shot_vector=())),
 'errors': {},
 'num_observables': 1,
 'num_diagonalizing_gates': 0,
 'num_trainable_params': 2,
 'num_device_wires': 2,
 'num_tape_wires': 2,
 'device_name': 'default.qubit',
 'level': 'gradient',
 'gradient_options': {},
 'interface': 'auto',
 'diff_method': 'best',
 'gradient_fn': 'backprop'}

It should have given out 'diff_method': 'parameter-shift' given that I had specified it while defining the circuit. But, it shows that 'diff_method': 'best'. This leads back to the same discussion of what the best method is and using circuit.best_method_str() to get it. On running circuit.best_method_str(dev, circuit.interface), it gave me 'backprop' yet again.

So, does that mean parameter-shift just isn’t running despite specifying it? Or, could I specify the method of differentiation differently somehow?

Another thing I wanted to mention is that I’m running the code on Google Colab. Could that have something to do with anything? Perhaps some device/performance issue that isn’t letting parameter-shift to function? Maybe parameter-shift requires certain hardware specs to run? I’m just thinking out loud. Would love to hear any insights.

Here’s the output of qml.about():

Name: PennyLane
Version: 0.37.0
Summary: PennyLane is a cross-platform Python library for quantum computing, quantum machine learning, and quantum chemistry. Train a quantum computer the same way as a neural network.
Home-page: https://github.com/PennyLaneAI/pennylane
Author: 
Author-email: 
License: Apache License 2.0
Location: /usr/local/lib/python3.10/dist-packages
Requires: appdirs, autograd, autoray, cachetools, networkx, numpy, packaging, pennylane-lightning, requests, rustworkx, scipy, semantic-version, toml, typing-extensions
Required-by: PennyLane_Lightning

Platform info:           Linux-6.1.85+-x86_64-with-glibc2.35
Python version:          3.10.12
Numpy version:           1.26.4
Scipy version:           1.13.1
Installed devices:
- default.clifford (PennyLane-0.37.0)
- default.gaussian (PennyLane-0.37.0)
- default.mixed (PennyLane-0.37.0)
- default.qubit (PennyLane-0.37.0)
- default.qubit.autograd (PennyLane-0.37.0)
- default.qubit.jax (PennyLane-0.37.0)
- default.qubit.legacy (PennyLane-0.37.0)
- default.qubit.tf (PennyLane-0.37.0)
- default.qubit.torch (PennyLane-0.37.0)
- default.qutrit (PennyLane-0.37.0)
- default.qutrit.mixed (PennyLane-0.37.0)
- default.tensor (PennyLane-0.37.0)
- null.qubit (PennyLane-0.37.0)
- lightning.qubit (PennyLane_Lightning-0.37.0)

CatalinaAlbornoz · August 22, 2024, 4:02pm

Hi @r_taeja,

I’ve found the issue! You actually need to specify the diff_method in the decorator, not as an argument of the circuit. So …

@qml.qnode(dev, diff_method="parameter-shift")
def circuit(params):

If you then run print(qml.specs(circuit)(params)) you get the right result.

I tested on Google Colab and it works!

r_taeja · August 22, 2024, 5:25pm

OH MY GOD!

I can’t believe I made such a basic error!! It works now. Please forgive the… totally could’ve corrected it myself - kinda question.

But, I’m very grateful for the help! Thank you so much @CatalinaAlbornoz

CatalinaAlbornoz · August 23, 2024, 10:00pm

No problem! Sometimes it’s hard to see these mistakes. I also missed it the first time myself!

In any case I’m glad you asked the question and now your code works.

Enjoy using PennyLane @r_taeja !

Topic		Replies	Views
Printing out diff_method PennyLane Help	5	310	May 29, 2023
Way to tell how PennyLane will differentiate a QNode PennyLane Help	1	325	February 18, 2022
Question about Generalized parameter shift PennyLane Help	3	291	April 25, 2023
Optimize circuits returning counts PennyLane Help	3	407	March 2, 2023
Possible discrepancy in the "Quantum gradients with backpropagation" tutorial PennyLane Help	3	197	March 8, 2024

Confusion regarding the output of circuit.best_method_str(dev, circuit.interface)

Related topics