# Release notes¶

## Release 0.20.0 (development release)¶

### Contributors

This release contains contributions from (in alphabetical order):

## Release 0.19.0 (current release)¶

### New features since last release

• Compact decompositions as described in https://arxiv.org/abs/2104.07561, (rectangular_compact and triangular_compact) are now available in the sf.decompositions module, and as options in the Interferometer operation. (#584)

This decomposition allows for lower depth photonic circuits in physical devices by applying two independent phase shifts in parallel inside each Mach-Zehnder interferometer. rectangular_compact reduces the layers of phase shifters from 2N+1 to N+2 for an N mode interferometer when compared to e.g. rectangular_MZ.

Example:

import numpy as np
from strawberryfields import Program
from strawberryfields.ops import Interferometer
from scipy.stats import unitary_group

M = 10

# generate a 10x10 Haar random unitary
U = unitary_group.rvs(M)

prog = Program(M)

with prog.context as q:
Interferometer(U, mesh='rectangular_compact') | q

# check that applied unitary is correct
compiled_circuit = prog.compile(compiler="gaussian_unitary")
commands = compiled_circuit.circuit
S = commands[0].op.p[0] # symplectic transformation
Uout = S[:M,:M] + 1j * S[M:,:M] # unitary transformation

print(np.allclose(U, Uout))

• A new compiler, GaussianMerge, has been added. It is aimed at reducing calculation overhead for non-Gaussian circuits by minimizing the amount of Gaussian operations in a circuit, while retaining the same functionality. (#591)

GaussianMerge merges Gaussian operations, where allowed, into GaussianTransform and Dgate operations. It utilizes the existing GaussianUnitary compiler to merge operations and Directed Acyclic Graphs to determine which operations can be merged.

modes = 4
cutoff_dim = 6

# prepare an intial state with 4 photons in as many modes
initial_state = np.zeros([cutoff_dim] * modes, dtype=complex)
initial_state[1, 1, 1, 1] = 1

prog = sf.Program(4)

with prog.context as q:
ops.Ket(initial_state) | q  # Initial state preparation
# Gaussian Layer
ops.S2gate(0.01, 0.01) | (q[0], q[1])
ops.BSgate(1.9, 1.7) | (q[1], q[2])
ops.BSgate(0.9, 0.2) | (q[0], q[1])
# Non-Gaussian Layer
ops.Kgate(0.5) | q[3]
ops.CKgate(0.7) | (q[2], q[3])
# Gaussian Layer
ops.BSgate(1.0, 0.4) | (q[0], q[1])
ops.BSgate(2.0, 1.5) | (q[1], q[2])
ops.Dgate(0.01) | q[0]
ops.Dgate(0.01) | q[0]
ops.Sgate(0.01, 0.01) | q[1]
# Non-Gaussian Layer
ops.Vgate(0.5) | q[2]

prog_merged = prog.compile(compiler="gaussian_merge")

• A new operation, PassiveChannel has been added. It allows for arbitrary linear/passive transformations (i.e., any operation which is linear in creation operators). Currently only supported by the gaussian backend. (#600)

from strawberryfields.ops import PassiveChannel, Sgate
import strawberryfields as sf
from scipy.stats import unitary_group
import numpy as np

M = 4

circuit = sf.Program(M)
U1 = unitary_group.rvs(M)
U2 = unitary_group.rvs(M)
losses = np.random.random(M)

T = U2 @ np.diag(losses) @ U1

eng = sf.Engine(backend='gaussian')
circuit = sf.Program(M)
with circuit.context as q:
for i in range(M):
ops.Sgate(1) | q[i]
ops.PassiveChannel(T) | q

cov = eng.run(circuit).state.cov()

• A new compiler, passive, allows for a circuit which only consists of passive elements to be compiled into a single PassiveChannel. (#600)

from strawberryfields.ops import BSgate, LossChannel, Rgate
import strawberryfields as sf

circuit = sf.Program(2)
with circuit.context as q:
Rgate(np.pi) | q[0]
BSgate(0.25 * np.pi, 0) | (q[0], q[1])
LossChannel(0.9) | q[1]

compiled_circuit = circuit.compile(compiler="passive")

>>> print(compiled_circuit)
PassiveChannel([[-0.7071+8.6596e-17j -0.7071+0.0000e+00j]
[-0.6708+8.2152e-17j  0.6708+0.0000e+00j]]) | (q[0], q[1])


### Improvements

• backends/tfbackend/ops.py is cleaned up to reduce line count, clarify function similarity across backend ops, and replace tensorflow.tensordot with broadcasting. (#567)

• Support is added for using a TDMProgram to construct time-domain circuits with Fock measurements and multiple loops. (#601)

• measure_threshold in the gaussian backend now supports displaced Gaussian states. (#615)

• Speed improvements are addded to gaussian_unitary compiler. (#603)

• Adds native support in the Fock backend for the MZgate. (#610)

• measure_threshold is now supported in the bosonic backend. (#618)

### Bug fixes

• Fixes an unexpected behaviour that can result in increasing memory usage due to sympy.lambdify caching too much data using linecache. (#579)

• Keeps symbolic expressions when converting a Strawberry Fields circuit to a Blackbird program by storing them as blackbird.RegRefTransforms in the resulting Blackbird program. (#596)

• Fixes a bug in the validation step of strawberryfields.tdm.TdmProgram.compile which almost always used the wrong set of allowed gate parameter ranges to validate the parameters in a program. (#605)

• The correct samples are now returned when running a TDMProgram with several shots, where timebins % concurrent_modes != 0. (#611)

• Fixes the formula used for sampling generaldyne outcomes in the gaussian backend. (#614)

• Measurement arguments are now stored as non-keyword arguments, instead of keyword arguments, in the resulting Blackbird program when using the io.to_blackbird() converter function. (#622)

• Factorials of numbers larger than 170 are now calculated using long integer arithmetic, using the flag exact=True in scipy.special.factorial, when calling sf.apps.similarity.orbit_cardinality. (#628)

### Documentation

• References to the simulon simulator target have been rewritten to simulon_gaussian to reflect changes made on the Xanadu Quantum Cloud. The language has been modified to imply that multiple simulators could be available on XQC. (#576)

### Contributors

This release contains contributions from (in alphabetical order):

J. Eli Bourassa, Jake Bulmer, Sebastian Duque, Theodor Isacsson, Aaron Robertson, Jeremy Swinarton, Antal Száva, Federico Rueda, Yuan Yao.

## Release 0.18.0¶

### New features since last release

• Adds the Bosonic backend, which can simulate states represented as linear combinations of Gaussian functions in phase space. (#533) (#538) (#539) (#541) (#546) (#549)

It can be regarded as a generalization of the Gaussian backend, since transformations on states correspond to modifications of the means and covariances of each Gaussian in the linear combination, along with changes to the coefficients of the linear combination. Example states that can be expressed using the new backend include all Gaussian, Gottesman-Kitaev-Preskill, cat and Fock states.

prog = sf.Program(1)
eng = sf.Engine('bosonic')

with prog.context as q:
sf.ops.GKP(epsilon=0.1) | q
sf.ops.MeasureX | q

results = eng.run(prog, shots=200)
samples = results.samples[:, 0]

plt.hist(samples, bins=100)
plt.show()

• Adds the sf.ops.GKP operation, which allows the Gottesman-Kitaev-Preskill state to be initialized on both the Bosonic and Fock backends. (#553) (#546)

GKP states are qubits, with the qubit state defined by:

$\ket{\psi}_{gkp} = \cos\frac{\theta}{2}\ket{0}_{gkp} + e^{-i\phi}\sin\frac{\theta}{2}\ket{1}_{gkp},$

where the computational basis states are $$\ket{\mu}_{gkp} = \sum_{n} \ket{(2n+\mu)\sqrt{\pi\hbar}}_{q}$$.

• Adds the measurement-based squeezing gate MSgate; a new front-end operation for the Bosonic backend. (#538) (#539) (#541)

MSgate is an implementation of inline squeezing that can be performed by interacting the target state with an ancillary squeezed vacuum state at a beamsplitter, measuring the ancillary mode with homodyne, and then applying a feed-forward displacement. The channel is implemented either on average (as a Gaussian CPTP map) or in the single-shot implementation. If the single-shot implementation is used, the measurement outcome of the ancillary mode is stored in the results object.

prog = sf.Program(1)
eng = sf.Engine('bosonic')

with prog.context as q:
sf.ops.Catstate(alpha=2) | q
r = 0.3
# Average map
sf.ops.MSgate(r, phi=0, r_anc=1.2, eta_anc=1, avg=True) | q
# Single-shot map
sf.ops.MSgate(r, phi=0, r_anc=1.2, eta_anc=1, avg=False) | q

results = eng.run(prog)
ancilla_samples = results.ancilla_samples

xvec = np.arange(-5, 5, 0.01)
pvec = np.arange(-5, 5, 0.01)
wigner = results.state.wigner(0, xvec, pvec)

plt.contourf(xvec, pvec, wigner)
plt.show()

• The tf backend now accepts the Tensor DType as argument. (#562)

Allows high cutoff dimension to give numerically correct calculations:

prog = sf.Program(2)
eng  = sf.Engine("tf", backend_options={"cutoff_dim": 50, "dtype": tf.complex128})
with prog.context as q:
Sgate(0.8) | q[0]
Sgate(0.8) | q[1]
BSgate(0.5,0.5) | (q[0], q[1])
BSgate(0.5,0.5) | (q[0], q[1])
state = eng.run(prog).state
N0, N0var = state.mean_photon(0)
N1, N1var = state.mean_photon(1)
print(N0)
print(N1)
print("analytical:", np.sinh(0.8)**2)


### Improvements

• Program compilation has been modified to support the XQC simulation service, Simulon. (#545)

• The sympmat, rotation_matrix, and haar_measure functions have been removed from backends/shared_ops.py. These functions are now imported from The Walrus. In addition, various outdated functionality from the shared_ops.py file has been removed, including the caching of beamsplitter and squeezing pre-factors. (#560) (#558)

• Sample processing in the TDMProgram is now more efficient, by replacing calls to pop with fancy indexing. (#548)

• No VisibleDeprecationWarning is raised when using the state wigner method. (#564)

• The backend utility module shared_ops.py has been removed, with all of its functionality now provided by The Walrus. (#573)

### Breaking changes

• Removes support for Python 3.6. (#573)

### Bug fixes

• Connection objects now send requests to the platform API at version 0.2.0 instead of the incorrect version number 1.0.0. (#540)

• TDM programs now expect a flat (not nested) dictionary of modes in device specifications obtained from the XQC platform API. (#566)

• Fixes a bug in the CatState operation, whereby the operation would return incorrect results for a high cutoff value. (#557) (#556)

### Documentation

• The “Hardware” quickstart page has been renamed to “Xanadu Quantum Cloud” to encompass both hardware and cloud simulators. A new “Cloud simulator” entry has been added, describing how to submit programs to be executed via the XQC simulator. (#547)

• Cleanup docs to make contribution easier. (#561)

• Add development requirements and format script to make contribution easier. (#563)

### Contributors

This release contains contributions from (in alphabetical order):

J. Eli Bourassa, Guillaume Dauphinais, Ish Dhand, Theodor Isacsson, Josh Izaac, Leonhard Neuhaus, Nicolás Quesada, Aaron Robertson, Krishna Kumar Sabapathy, Jeremy Swinarton, Antal Száva, Ilan Tzitrin.

## Release 0.17.0¶

### New features since last release

• TDMProgram objects can now be compiled and submitted via the API. (#476)

• Wigner functions can be plotted directly via Strawberry Fields using Plot.ly. (#495)

prog = sf.Program(1)
eng = sf.Engine('fock', backend_options={"cutoff_dim": 10})

with prog.context as q:
gamma = 2
Vgate(gamma) | q[0]

state = eng.run(prog).state

xvec = np.arange(-4, 4, 0.01)
pvec = np.arange(-4, 4, 0.01)
mode = 0

sf.plot_wigner(state, mode, xvec, pvec, renderer="browser")

• Fock state marginal probabilities can be plotted directly via Strawberry Fields using Plot.ly. (#510)

prog = sf.Program(1)
eng = sf.Engine('fock', backend_options={"cutoff_dim":5})

with prog.context as q:
Sgate(0.5) | q[0]

state = eng.run(prog).state
state.all_fock_probs()

modes = [0]

sf.plot_fock(state, modes, cutoff=5, renderer="browser")

• Position and momentum quadrature probabilities can be plotted directly via Strawberry Fields using Plot.ly. (#510)

prog = sf.Program(1)
eng = sf.Engine('fock', backend_options={"cutoff_dim":5})

with prog.context as q:
Sgate(0.5) | q[0]

state = eng.run(prog).state

modes = [0]
xvec = np.arange(-4, 4, 0.1)
pvec = np.arange(-4, 4, 0.1)


• Strawberry Fields code can be generated from a program (and an engine) by calling sf.io.generate_code(program, eng=engine). (#496)

### Improvements

• Connection objects now send versioned requests to the platform API. (#512)

• TDMProgram allows application of gates with more than one symbolic parameter. #492

• The copies option, when constructing a TDMProgram, has been removed. Instead, the number of copies of a TDM algorithm can now be set by passing the shots keyword argument to the eng.run() method. (#489)

>>> with prog.context([1, 2], [3, 4]) as (p, q):
...     ops.Sgate(0.7, 0) | q[1]
...     ops.BSgate(p[0]) | (q[0], q[1])
...     ops.MeasureHomodyne(p[1]) | q[0]
>>> eng = sf.Engine("gaussian")
>>> results = eng.run(prog, shots=3)


Furthermore, the TDMProgram.unrolled_circuit attribute now only contains the single-shot unrolled circuit. Unrolling with multiple shots can still be specified via the unroll method: TDMProgram.unroll(shots=60).

• The Result.samples returned by TDM programs has been updated to return samples of shape (shots, spatial modes, timebins) instead of (shots, spatial modes * timebins). (#489)

• A sample post-processing function is added that allows users to move vacuum mode measurements from the first shots to the last shots, and potentially crop out the final shots containing these measurements. (#489)

• pytest-randomly is added to the SF tests. (#480)

• TDMProgram objects can now be serialized into Blackbird scripts, and vice versa. (#476)

### Breaking Changes

• Jobs are submitted to the Xanadu Quantum Cloud through a new OAuth based authentication flow using offline refresh tokens and access tokens. (#520)

### Bug fixes

• Fixes a bug where Dgate, Coherent, and DisplacedSqueezed do not support TensorFlow tensors if the tensor has an added dimension due to the existence of batching. (#507)

• Fixes an issue with reshape_samples where the samples were sometimes reshaped in the wrong way. (#489)

• The list of modes is now correctly added to the Blackbird program when using the io.to_blackbird function. (#476)

• Fixes a bug where printing the Result object containing samples from a time-domain job would result in an error. Printing the result object now correctly displays information about the results. (#493)

• Removes the antlr4 requirement due to version conflicts. (#494)

• TDMProgram.run_options is now correctly used when running a TDM program. (#500)

• Fixes a bug where a single parameter list passed to the TDMProgram context results in an error. (#503)

### Documentation

• TDMProgram docstring is updated to make it clear that only Gaussian programs are allowed. (#519)

• Clarifies special cases for the MZgate in the docstring. (#479)

### Contributors

This release contains contributions from (in alphabetical order):

Tom Bromley, Jack Brown, Theodor Isacsson, Josh Izaac, Fabian Laudenbach, Tim Leisti, Nicolas Quesada, Antal Száva.

## Release 0.16.0¶

### New features since last release

• Moves the chemistry utility functions prob and marginals to the apps.qchem.utils module of Applications layer of Strawberry Fields. These functions were initially created as utility functions to help simulating vibrational dynamics. However, they can also be used in other applications and therefore should be moved to the apps.qchem.utils module which hosts general-purpose utility functions for chemistry applications. (#487)

• Adds the ability to construct time domain multiplexing algorithms via the new sf.TDMProgram class, for highly scalable simulation of Gaussian states. (#440)

For example, creating and simulating a time domain program with 2 concurrent modes:

>>> import strawberryfields as sf
>>> from strawberryfields import ops
>>> prog = sf.TDMProgram(N=2)
>>> with prog.context([1, 2], [3, 4], copies=3) as (p, q):
...     ops.Sgate(0.7, 0) | q[1]
...     ops.BSgate(p[0]) | (q[0], q[1])
...     ops.MeasureHomodyne(p[1]) | q[0]
>>> eng = sf.Engine("gaussian")
>>> results = eng.run(prog)
>>> print(results.all_samples)
{0: [array([1.26208025]), array([1.53910032]), array([-1.29648336]),
array([0.75743215]), array([-0.17850101]), array([-1.44751996])]}


For more details, see the code documentation.

• Adds the function VibronicTransition to the apps.qchem.vibronic module. This function generates a custom Strawberry Fields operation for applying the Doktorov operator on a given state. (#451)

>>> from strawberryfields.apps.qchem.vibronic import VibronicTransition
>>> modes = 2
>>> p = sf.Program(modes)
>>> with p.context as q:
...     VibronicTransition(U1, r, U2, alpha) | q

• Adds the TimeEvolution function to the apps.qchem.dynamics module. This function generates a custom Strawberry Fields operation for applying a time evolution operator on a given state. (#455)

>>> modes = 2
>>> p = sf.Program(modes)
>>> with p.context as q:
...     sf.ops.Fock(1) | q[0]
...     sf.ops.Interferometer(Ul.T) | q
...     TimeEvolution(w, t) | q
...     sf.ops.Interferometer(Ul) | q


where w is the normal mode frequencies, and t the time in femtoseconds.

• Molecular data and pre-generated samples for water and pyrrole have been added to the data module of the Applications layer of Strawberry Fields. For more details, please see the data module documentation (#463)

• Adds the function read_gamess to the qchem module to extract the atomic coordinates, atomic masses, vibrational frequencies, and normal modes of a molecule from the output file of a vibrational frequency calculation performed with the GAMESS quantum chemistry package. (#460)

>>> r, m, w, l = read_gamess('../BH_data.out')
>>> r # atomic coordinates
array([[0.0000000, 0.0000000, 0.0000000],
[1.2536039, 0.0000000, 0.0000000]])
>>> m # atomic masses
array([11.00931,  1.00782])
>>> w # vibrational frequencies
array([19.74, 19.73, 0.00, 0.00, 0.00, 2320.32])
>>> l # normal modes
array([[-0.0000000e+00, -7.5322000e-04, -8.7276210e-02,  0.0000000e+00,
8.2280900e-03,  9.5339055e-01],
[-0.0000000e+00, -8.7276210e-02,  7.5322000e-04,  0.0000000e+00,
9.5339055e-01, -8.2280900e-03],
[ 2.8846925e-01, -2.0000000e-08,  2.0000000e-08,  2.8846925e-01,
-2.0000000e-08,  2.0000000e-08],
[ 2.0000000e-08,  2.8846925e-01, -2.0000000e-08,  2.0000000e-08,
2.8846925e-01, -2.0000000e-08],
[-2.0000000e-08,  2.0000000e-08,  2.8846925e-01, -2.0000000e-08,
2.0000000e-08,  2.8846925e-01],
[-8.7279460e-02,  0.0000000e+00,  0.0000000e+00,  9.5342606e-01,
-0.0000000e+00, -0.0000000e+00]])


### Improvements

• When jobs submitted to the Xanadu Quantum Cloud are canceled, they will now display a cancel_pending JobStatus until the cancellation is confirmed. (#456)

### Bug fixes

• Fixed a bug where the function reduced_dm in backends/tfbackend/states.py gives the wrong output when passing it several modes. (#471)

• Fixed a bug in the function reduced_density_matrix in backends/tfbackend/ops.py which caused the wrong subsystems to be traced out. (#467) (#470)

• Fixed a bug where decompositions to Mach-Zehnder interferometers would return incorrect results on NumPy 1.19. (#473)

• The Walrus version 0.14 introduced modified function names. Affected functions have been updated in Strawberry Fields to avoid deprecation warnings. (#472)

### Documentation

• Adds further testing and coverage descriptions to the developer documentation. This includes details regarding the Strawberry Fields test structure and test decorators. (#461)

• Updates the minimum required version of TensorFlow in the development guide. (#468)

### Contributors

This release contains contributions from (in alphabetical order):

Juan Miguel Arrazola, Tom Bromley, Theodor Isacsson, Josh Izaac, Soran Jahangiri, Nathan Killoran, Fabian Laudenbach, Nicolás Quesada, Antal Száva, ‪Ilan Tzitrin.

## Release 0.15.1¶

### Improvements

• Adds the ability to bypass recompilation of programs if they have been compiled already to the target device. (#447)

### Breaking Changes

• Changes the default compiler for devices that don’t specify a default from "Xcov" to "Xunitary". This compiler is slightly more strict and only compiles the unitary, not the initial squeezers, however avoids any unintentional permutations. (#445)

### Bug fixes

• Fixes a bug where a program that amounts to the identity operation would cause an error when compiled using the xcov compiler. (#444)

### Documentation

• Updates the README.rst file and hardware access links. (#448)

### Contributors

This release contains contributions from (in alphabetical order):

Theodor Isacsson, Josh Izaac, Nathan Killoran, Nicolás Quesada, Antal Száva

## Release 0.15.0¶

### New features since last release

• Adds the ability to train variational GBS circuits in the applications layer. (#387) (#388) (#391) (#393) (#414) (#415)

Trainable parameters can be embedded into a VGBS class:

from strawberryfields.apps import data, train

d = data.Mutag0()
embedding = train.Exp(d.modes)
n_mean = 5

vgbs = train.VGBS(d.adj, 5, embedding, threshold=False, samples=np.array(d[:1000]))


Properties of the variational GBS distribution for different choices of trainable parameters can then be inspected:

>>> params = 0.1 * np.ones(d.modes)
>>> vgbs.n_mean(params)
3.6776094165797364


A cost function can then be created and its value and gradient accessed:

>>> h = lambda x: np.sum(x)
>>> cost = train.Stochastic(h, vgbs)
>>> cost(params, n_samples=1000)
3.940396998165503
array([-0.54988876, -0.49270263, -0.6628071 , -1.13057762, -1.13568456,
-0.70180571, -0.6266806 , -0.68803539, -1.11032533, -1.12853718,
-0.59172261, -0.47830748, -0.96901676, -0.66938217, -0.85162006,
-0.27188134, -0.26955011])


For more details, see the VGBS training demo.

• Feature vectors of graphs can now be calculated exactly in the apps.similarity module of the applications layer. Datasets of pre-calculated feature vectors are available in apps.data. (#390) (#401)

>>> from strawberryfields.apps import data
>>> from strawberryfields.apps.similarity import feature_vector_sampling
>>> samples = data.Mutag0()
>>> feature_vector_sampling(samples, [2, 4, 6])
[0.19035, 0.2047, 0.1539]


For more details, see the graph similarlity demo.

• A new strawberryfields.apps.qchem module has been introduced, centralizing all quantum chemistry applications. This includes various new features and improvements:

• Adds the apps.qchem.duschinsky() function for generation of the Duschinsky rotation matrix and displacement vector which are needed to simulate a vibronic process with Strawberry Fields. (#434)

• Adds the apps.qchem.dynamics module for simulating vibrational quantum dynamics in molecules. (#402) (#411) (#419) (#421) (#423) (#430)

This includes:

• dynamics.evolution() constructs a custom operation that encodes the input chemical information. This custom operation can then be used within a Strawberry Fields Program.

• dynamics.sample_coherent(), dynamics.sample_fock() and dynamics.sample_tmsv() functions allow for generation of samples from a variety of input states.

• The probability of an excited state can then be estimated with the dynamics.prob() function, which calculates the relative frequency of the excited state among the generated samples.

• Finally, the dynamics.marginals() function generates marginal distributions.

• The sf.apps.vibronic module has been relocated to within the qchem module. As a result, the apps.sample.vibronic() function is now accessible under apps.qchem.vibronic.sample(), providing a single location for quantum chemistry functionality. (#416)

For more details, please see the qchem documentation.

• The GaussianState returned from simulations using the Gaussian backend now has feature parity with the FockState object returned from the Fock backends. (#407)

In particular, it now supports the following methods:

• GaussianState.dm()

• GaussianState.ket()

• GaussianState.all_fock_probs()

In addition, the existing GaussianState.reduced_dm() method now supports multi-mode reduced density matrices.

• Adds the sf.utils.samples_expectation, sf.utils.samples_variance and sf.utils.all_fock_probs_pnr functions for obtaining counting statistics from samples. (#399)

• Compilation of Strawberry Fields programs has been overhauled.

• Strawberry Fields can now access the Xanadu Cloud device specifications API. The Connection class has a new method Connection.get_device, which returns a DeviceSpec class. (#429) (#432)

• New Xstrict, Xcov, and Xunitary compilers for compiling programs into the X architecture have been added. (#358) (#438)

• Finally, the strawberryfields.circuitspecs module has been renamed to strawberryfields.compilers.

• Adds diagonal_expectation method for the BaseFockState class, which returns the expectation value of any operator that is diagonal in the number basis. (#389)

• Adds parity_expectation method as an instance of diagonal_expectation for the BaseFockState class, and its own function for BaseGaussianState. This returns the expectation value of the parity operator, defined as (-1)^N. (#389)

### Improvements

• Modifies the rectangular interferometer decomposition to make it more efficient for hardware devices. Rather than decomposing the interferometer using Clements $$T$$ matrices, the decomposition now directly produces Mach-Zehnder interferometers corresponding to on-chip phases. (#363)

• Changes the number_expectation method for the BaseFockState class to be an instance of diagonal_expectation. (#389)

• Increases the speed at which the following gates are generated: Dgate, Sgate, BSgate and S2gate by relying on a recursive implementation recently introduced in thewalrus. This has substantial effects on the speed of the Fockbackend and the TFbackend, especially for high cutoff values. (#378) (#381)

• All measurement samples can now be accessed via the results.all_samples attribute, which returns a dictionary mapping the mod index to a list of measurement values. This is useful for cases where a single mode may be measured multiple times. (#433)

### Breaking Changes

• Removes support for Python 3.5. (#385)

• Complex parameters now are expected in polar form as two separate real parameters. (#378)

### Contributors

This release contains contributions from (in alphabetical order):

Juan Miguel Arrazola, Tom Bromley, Jack Ceroni, Aroosa Ijaz, Theodor Isacsson, Josh Izaac, Nathan Killoran, Soran Jahangiri, Shreya P. Kumar, Filippo Miatto, Nicolás Quesada, Antal Száva

## Release 0.14.0¶

### New features since last release

• Dark counts can now be added to the samples received from photon measurements in the Fock basis (sf.ops.MeasureFock) during a simulation.

• The "tf" backend now supports TensorFlow 2.0 and above. (#283) (#320) (#323) (#361) (#372) (#373) (#374) (#375) (#377)

For more details and demonstrations of the new TensorFlow 2.0-compatible backend, see our optimization and machine learning tutorials.

For example, using TensorFlow 2.0 to train a variational photonic circuit:

eng = sf.Engine(backend="tf", backend_options={"cutoff_dim": 7})
prog = sf.Program(1)

with prog.context as q:
# Apply a single mode displacement with free parameters
Dgate(prog.params("a"), prog.params("p")) | q[0]

alpha = tf.Variable(0.1)
phi = tf.Variable(0.1)

for step in range(50):
# reset the engine if it has already been executed
if eng.run_progs:
eng.reset()

# execute the engine
results = eng.run(prog, args={'a': alpha, 'p': phi})
# get the probability of fock state |1>
prob = results.state.fock_prob([1])
# negative sign to maximize prob
loss = -prob

print("Value at step {}: {}".format(step, prob))

• Adds the method number_expectation that calculates the expectation value of the product of the number operators of a given set of modes. (#348)

prog = sf.Program(3)
with prog.context as q:
ops.Sgate(0.5) | q[0]
ops.Sgate(0.5) | q[1]
ops.Sgate(0.5) | q[2]
ops.BSgate(np.pi/3, 0.1) |  (q[0], q[1])
ops.BSgate(np.pi/3, 0.1) |  (q[1], q[2])


Executing this on the Fock backend,

>>> eng = sf.Engine("fock", backend_options={"cutoff_dim": 10})
>>> state = eng.run(prog).state


we can compute the expectation value $$\langle \hat{n}_0\hat{n}_2\rangle$$:

>>> state.number_expectation([0, 2])


### Improvements

• Add details to the error message for failed remote jobs. (#370)

• The required version of The Walrus was increased to version 0.12, for tensor number expectation support. (#380)

### Contributors

This release contains contributions from (in alphabetical order):

Tom Bromley, Theodor Isacsson, Josh Izaac, Nathan Killoran, Filippo Miatto, Nicolás Quesada, Antal Száva, Paul Tan.

## Release 0.13.0¶

### New features since last release

• Adds initial support for the Xanadu’s photonic quantum hardware. (#101) (#148) (#294) (#327) (#328) (#329) (#330) (#334) (#336) (#337) (#339)

Jobs can now be submitted to the Xanadu cloud platform to be run on supported hardware using the new RemoteEngine:

import strawberryfields as sf
from strawberryfields import ops
from strawberryfields.utils import random_interferometer

con = sf.api.Connection(token="AUTH_TOKEN")
eng = sf.RemoteEngine("X8", connection=con)
prog = sf.Program(8)

U = random_interferometer(4)

with prog.context as q:
ops.S2gate(1.0) | (q[0], q[4])
ops.S2gate(1.0) | (q[1], q[5])
ops.S2gate(1.0) | (q[2], q[6])
ops.S2gate(1.0) | (q[3], q[7])

ops.Interferometer(U) | q[:4]
ops.Interferometer(U) | q[4:]
ops.MeasureFock() | q

result = eng.run(prog, shots=1000)


For more details, see the photonic hardware quickstart and tutorial.

• Significantly speeds up the Fock backend of Strawberry Fields, through a variety of changes:

• The Fock backend now uses The Walrus high performance implementations of the displacement, squeezing, two-mode squeezing, and beamsplitter operations. (#287) (#289)

• Custom tensor contractions which make use of symmetry relations for the beamsplitter and the two-mode squeeze gate have been added, as well as more efficient contractions for diagonal operations in the Fock basis. (#292)

• New sf command line program for configuring Strawberry Fields for access to the Xanadu cloud platform, as well as submitting and executing jobs from the command line. (#146) (#312)

The new Strawberry Fields command line program sf provides several utilities including:

• sf configure [--token] [--local]: configure the connection to the cloud platform

• sf run input [--output FILE]: submit and execute quantum programs from the command line

• sf --ping: verify your connection to the Xanadu cloud platform

For more details, see the documentation.

• New configuration functions to load configuration from keyword arguments, environment variables, and configuration files. (#298) (#306)

This includes the ability to automatically store Xanadu cloud platform credentials in a configuration file using the new function

sf.store_account("AUTHENTICATION_TOKEN")


as well as from the command line,

\$ sf configure --token AUTHENTICATION_TOKEN


Configuration files can be saved globally, or locally on a per-project basis. For more details, see the configuration documentation

• Adds configuration functions for resetting, deleting configurations, as well as displaying available configuration files. (#359)

• Adds the x_quad_values and p_quad_values methods to the state class. This allows calculation of x and p quadrature probability distributions by integrating across the Wigner function. (#270)

• Adds support in the applications layer for node-weighted graphs.

Sample from graphs with node weights using a special-purpose encoding (#295):

from strawberryfields.apps import sample

# generate a random graph
g = nx.erdos_renyi_graph(20, 0.6)
a = nx.to_numpy_array(g)

# define node weights
# and encode into the adjacency matrix
w = [i for i in range(20)]
a = sample.waw_matrix(a, w)

s = sample.sample(a, n_mean=10, n_samples=10)
s = sample.postselect(s, min_count=4, max_count=20)
s = sample.to_subgraphs(s, g)


Node weights can be input to search algorithms in the clique and subgraph modules (#296) (#297):

from strawberryfields.apps import clique
c = [clique.shrink(s_, g, node_select=w) for s_ in s]
[clique.search(c_, g, iterations=10, node_select=w) for c_ in c]

from strawberryfields.apps import subgraph
subgraph.search(s, g, min_size=5, max_size=8, node_select=w)


### Improvements

• Moved Fock backend apply-gate functions to Circuit class, and removed apply_gate_einsum and Circuits._apply_gate, since they were no longer used. (#293)

• Results returned from all backends now have a unified type and shape. In addition, attempting to use batching, post-selection and feed-foward together with multiple shots now raises an error. (#300)

• Modified the rectangular decomposition to ensure that identity-like unitaries are implemented with no swaps. (#311)

### Bug fixes

• Symbolic Operation parameters are now compatible with TensorFlow 2.0 objects. (#282)

• Added sympy>=1.5 to the list of dependencies. Removed the sympy.functions.atan2 workaround now that SymPy has been fixed. (#280)

• Removed two unnecessary else statements that pylint complained about. (#290)

• Fixed a bug in the MZgate, where the internal and external phases were in the wrong order in both the docstring and the argument list. The new signature is MZgate(phase_in, phase_ex), matching the existing rectangular_symmetric decomposition. (#301)

• Updated the relevant methods in RemoteEngine and Connection to derive shots from the Blackbird script or Program if not explicitly specified. (#327)

• Fixed a bug in homodyne measurements in the Fock backend, where computed probability values could occasionally include small negative values due to floating point precision error. (#364)

• Fixed a bug that caused an exception when printing results with no state. (#367)

• Improves the Takagi decomposition, by making explicit use of the eigendecomposition of real symmetric matrices. (#352)

### Contributors

This release contains contributions from (in alphabetical order):

Ville Bergholm, Tom Bromley, Jack Ceroni, Theodor Isacsson, Josh Izaac, Nathan Killoran, Shreya P Kumar, Leonhard Neuhaus, Nicolás Quesada, Jeremy Swinarton, Antal Száva, Paul Tan, Zeid Zabaneh.

## Release 0.12.1¶

### New features

• A new gaussian_unitary circuitspec that can be used to compile any sequency of Gaussian transformations into a single GaussianTransform gate and a sequence of single mode Dgates. (#238)

### Improvements

• Add new Strawberry Fields applications paper to documentation (#274)

• Update figure for GBS device in documentation (#275)

### Bug fixes

• Fix installation issue with incorrect minimum version number for thewalrus (#272) (#277)

• Correct URL for image in README (#273)

• Add applications data to MANIFEST.in (#278)

### Contributors

This release contains contributions from (in alphabetical order):

Ville Bergholm, Tom Bromley, Nicolás Quesada, Paul Tan

## Release 0.12.0¶

### New features

• A new applications layer, allowing users to interface samples generated from near-term photonic devices with problems of practical interest. The apps package consists of the following modules:

• The apps.sample module, for encoding graphs and molecules into Gaussian boson sampling (GBS) and generating corresponding samples.

• The apps.subgraph module, providing a heuristic algorithm for finding dense subgraphs from GBS samples.

• The apps.clique module, providing tools to convert subgraphs sampled from GBS into cliques and a heuristic to search for larger cliques.

• The apps.similarity module, allowing users to embed graphs into high-dimensional feature spaces using GBS. Resulting feature vectors provide measures of graph similarity for machine learning tasks.

• The apps.points module, allowing users to sample subsets of points according to new point processes that can be generated from a GBS device.

• The apps.vibronic module, providing functionality to construct the vibronic absorption spectrum of a molecule from GBS samples.

### Improvements

• The documentation was improved and refactored. Changes include:

• A brand new theme, now matching PennyLane (#262)

• The documentation has been restructured to make it easier to navigate (#266)

### Contributors

This release contains contributions from (in alphabetical order):

Juan Miguel Arrazola, Tom Bromley, Josh Izaac, Soran Jahangiri, Nicolás Quesada

## Release 0.11.2¶

### New features

• Adds the MZgate to ops.py, representing a Mach-Zehnder interferometer. This is not a primitive of the existing simulator backends; rather, _decompose() is defined, decomposing it into an external phase shift, two 50-50 beamsplitters, and an internal phase shift. (#127)

• The Chip0Spec circuit class now defines a compile method, allowing arbitrary unitaries comprised of {Interferometer, BSgate, Rgate, MZgate} operations to be validated and compiled to match the topology of chip0. (#127)

• strawberryfields.ops.BipartiteGraphEmbed quantum decomposition now added, allowing a bipartite graph to be embedded on a device that allows for initial two-mode squeezed states, and block diagonal unitaries.

• Added threshold measurements, via the new operation MeasureThreshold, and provided implementation of this operation in the Gaussian backend. (#152)

• Programs can now have free parameters/arguments which are only bound to numerical values when the Program is executed, by supplying the actual argument values to the Engine.run method. (#163)

### API Changes

• The strawberryfields.ops.Measure shorthand has been deprecated in favour of strawberryfields.ops.MeasureFock(). (#145)

• Several changes to the strawberryfields.decompositions module: (#127)

• The name clements has been replaced with rectangular to correspond with the shape of the resulting decomposition.

• All interferometer decompositions (rectangular, rectangular_phase_end, rectangular_symmetric, and triangular) now have standardized outputs (tlist, diag, tilist), so they can easily be swapped.

• Several changes to ops.Interferometer: (#127)

• The calculation of the ops.Interferometer decomposition has been moved from __init__ to _decompose(), allowing the interferometer decomposition type to be set by a CircuitSpec during compilation.

• **kwargs is now passed through from Operation.decompose -> Gate.decompose -> SpecificOp._decompose, allowing decomposition options to be passed during compilation.

• ops.Interferometer now accepts the keyword argument mesh to be set during initialization, allowing the user to specify the decomposition they want.

• Moves the Program.compile_seq method to CircuitSpecs.decompose. This allows it to be accessed from the CircuitSpec.compile method. Furthermore, it now must also be passed the program registers, as compilation may sometimes require this. (#127)

• Parameter class is replaced by MeasuredParameter and FreeParameter, both inheriting from sympy.Symbol. Fixed numeric parameters are handled by the built-in Python numeric classes and numpy arrays. (#163)

• Parameter, RegRefTransform and convert are removed. (#163)

### Improvements

• Photon-counting measurements can now be done in the Gaussian backend for states with nonzero displacement. (#154)

• Added a new test for the cubic phase gate (#160)

• Added new integration tests for the Gaussian gates that are not primitive, i.e., P, CX, CZ, and S2. (#173)

### Bug fixes

• Fixed bug in strawberryfields.decompositions.rectangular_symmetric so its returned phases are all in the interval [0, 2*pi), and corrects the function docstring. (#196)

• When using the 'gbs' compilation target, the measured registers are now sorted in ascending order in the resulting compiled program. (#144)

• Fixed typo in the Gaussian Boson Sampling example notebook. (#133)

• Fixed a bug in the function smeanxp of the Gaussian Backend simulator. (#154)

• Clarified description of matrices that are accepted by graph embed operation. (#147)

• Fixed typos in the documentation of the CX gate and BSgate (#166) (#167) (#169)

## Release 0.11.1¶

### Improvements

• Added the circuit_spec attribute to BaseBackend to denote which CircuitSpecs class should be used to validate programs for each backend (#125).

• Removed the return_state keyword argument from LocalEngine.run(). Now no state object is returned if modes==[]. (#126)

• Fixed a typo in the boson sampling tutorial. (#133)

### Bug fixes

• Allows imported Blackbird programs to store target options as default run options. During eng.run, if no run options are provided as a keyword argument, the engine will fall back on the run options stored within the program. This fixes a bug where shots specified in Blackbird scripts were not being passed to eng.run. (#130)

• Removes ModuleNotFoundError from the codebase, replacing all occurrences with ImportError. Since ModuleNotFoundError was only introduced in Python 3.6+, this fixes a bug where Strawberry Fields was not importable on Python 3.5 (#124).

• Updates the Chip0 template to use MeasureFock() | [0, 1, 2, 3], which will allow correct fock measurement behaviour when simulated on the Gaussian backend (#124).

• Fixed a bug in the GraphEmbed op, which was not correctly determining when a unitary was the identity (#128).

## Release 0.11.0¶

This is a significant release, with breaking changes to how quantum programs are constructed and executed. For example, the following Strawberry Fields program, <= version 0.10:

eng, q = sf.Engine(2, hbar=0.5)

with eng:
Sgate(0.5) | q[0]
MeasureFock() | q[0]

state = eng.run("fock", cutoff_dim=5)
ket = state.ket()
print(q[0].val)


would now be written, in v0.11, as follows:

sf.hbar = 0.5
prog = sf.Program(2)
eng = sf.Engine("fock", backend_options={"cutoff_dim": 5})

with prog.context as q:
Sgate(0.5) | q[0]
MeasureFock() | q[0]

results = eng.run(prog)
ket = results.state.ket()
print(results.samples[0])


### New features

• The functionality of the Engine class has been divided into two new classes: Program, which represents a quantum circuit or a fragment thereof, and Engine, which executes Program instances.

• Introduced the BaseEngine abstract base class and the LocalEngine child class. Engine is kept as an alias for LocalEngine.

• The Engine API has been changed slightly:

The engine is initialized with the required backend, as well as a backend_options dictionary, which is passed to the backend:

eng = sf.Engine("fock", backend_options={"cutoff_dim": 5}


LocalEngine.run() now accepts a program to execute, and returns a Result object that contains both a state object (Result.state) and measurement samples (Result.samples):

results = eng.run(prog)
state = results.state
samples = results.samples

• compile_options can be provided when calling LocalEngine.run(). These are passed to the compile() method of the program before execution.

• run_options can be provided when calling LocalEngine.run(). These are used to determine the characteristics of the measurements and state contained in the Results object returned after the program is finished executing.

• shots keyword argument can be passed to run_options, enabling multi-shot sampling. Supported only in the Gaussian backend, and only for Fock measurements.

• The Gaussian backend now officially supports Fock-basis measurements (MeasureFock), but does not update the quantum state after a Fock measurement.

• Added the io module, which is used to save/load standalone Blackbird scripts from/into Strawberry Fields. Note that the Blackbird DSL has been spun off as an independent package and is now a dependency of Strawberry Fields.

• Added a new interferometer decomposition mach_zehnder to the decompositions module.

• Added a Configuration class, which is used to load, store, save, and modify configuration options for Strawberry Fields.

• hbar is now set globally for the entire session, by setting the value of sf.hbar (default is 2).

• Added the ability to generate random real (orthogonal) interferometers and random block diagonal symplectic and covariance matrices.

• about(), which prints human-readable system info including installed versions of various Python packages.

• cite(), which prints a bibtex citation for SF.

• Added a glossary to the documentation.

### API Changes

• Added the circuitspecs subpackage, containing the CircuitSpecs class and a quantum circuit database.

The database can be used to

• Validate that a Program belongs in a specific circuit class.

• Compile a Program for a desired circuit target, e.g., so that it can be executed on a given backend. The database includes a number of compilation targets, including Gaussian Boson Sampling circuits.

• The way hbar is handled has been simplified:

• The backend API is now entirely hbar-independent, i.e., every backend API method is defined in terms of a and a^dagger only, not x and p.

• The backends always explicitly use hbar=2 internally.

• hbar is now a global, frontend-only variable that the user can set at the beginning of the session. It is used at the Operation.apply() level to scale the inputs and outputs of the backend API calls as needed, and inside the State objects.

• The only backend API calls that need to do hbar scaling for the input parameters are the X, Z, and V gates, the Gaussian state decomposition, and homodyne measurements (both the returned value and postselection argument are scaled).

### Improvements

• Removed TensorFlow as an explicit dependency of Strawberry Fields. Advanced users can still install TensorFlow manually using pip install tensorflow==1.3 and use as before.

• The behaviour and function signature of the GraphEmbed operation has been updated.

• Remove the unused Command.decomp instance attribute.

• Better error messages for the New operation when used outside of a circuit.

• Docstrings updated in the decompositions module.

• Docstrings for Fock backend reformatted and cleaned up.

• Cleaning up of citations and references.bib file.

• Typos in documentation fixed.

### Bug fixes

• Fixed a bug with installation on Windows for certain locales.

• Fixed a bug in the New operation.

• Bugfix in Gate.merge()

• Fixed bugs in measure_fock in the TensorFlow backend which caused samples to be evaluated independently and for conditional states to be potentially decoupled from the measurement results.

• Fixed a latent bug in graph_embed.

• Bugfix for Bloch-Messiah returning non-symplectic matrices when input is passive.

### Contributors

This release contains contributions from (in alphabetical order):

Ville Bergholm, Tom Bromley, Ish Dhand, Karel Dumon, Xueshi Guo, Josh Izaac, Nathan Killoran, Leonhard Neuhaus, Nicolás Quesada.

## Release 0.10¶

### New features

• Added two new utility functions to extract a numerical representation of a circuit from an Engine object: extract_unitary and extract_channel.

• Added a LaTeX quantum circuit drawer, that outputs the engine queue or the applied operations as a qcircuit compatible circuit diagram.

• Added support for an alternative form of Clements decomposition, where the local phases occur at the end rather than in the middle of the beamsplitter array. This decomposition is more symmetric than the intermediate one, which could make it more robust. This form also makes it easier to implement a tensor-network simulation of linear optics.

• Adds the GraphEmbed quantum operation/decomposition to the Strawberry Fields frontend. This allows the embedding of an arbitrary (complex-valued) weighted adjacency matrix into a Gaussian boson sampler.

• Adds support for the Reck decomposition

• Added documentation to the Quantum Algorithms section on CV quantum neural networks

### Improvements

• Test suite has been ported to pytest

• Linting improvements

• Made corrections to the Clements decomposition documentation and docstring, and fixed the Clements unit tests to ensure they are deterministic.

### Bug fixes

• Fixed Bloch-Messiah bug arising when singular values were degenerate. Previously, the Bloch-Messiah decomposition did not return matrices in the canonical symplectic form if one or more of the Bloch-Messiah singular values were degenerate.

### Contributors

This release contains contributions from (in alphabetical order):

Shahnawaz Ahmed, Thomas R. Bromley, Ish Dhand, Marcus Edwards, Christian Gogolin, Josh Izaac, Nathan Killoran, Filippo Miatto, Nicolás Quesada.

## Release 0.9¶

### New features

• Updated the Strawberry Fields gallery, featuring community-submitted content (tutorials, notebooks, repositories, blog posts, research papers, etc.) using Strawberry Fields

• Added the @operation decorator, which allows commonly-used algorithms and subroutines to be declared in blackbird code as one-liner operations

• Added a ThermalLossChannel to the Strawberry Fields API (currently supported by the Gaussian backend)

• Added a poly_quad_expectation method to the state objects for Gaussian and Fock backends

### Improvements

• New and improved tests

• Fixed typos in code/documentation

### Contributors

This release contains contributions from:

Juan Leni, Arthur Pesah, Brianna Gopaul, Nicolás Quesada, Josh Izaac, and Nathan Killoran.

## Release 0.8¶

### New features

• You can now prepare multimode states in all backends, via the following new quantum operations in strawberryfields.ops:

• Ket

• DensityMatrix

• Gaussian

Both Ket and DensityMatrix work with the Fock backends, while Gaussian works with all three, applying the Williamson decomposition or, optionally, directly preparing the Gaussian backend with the provided Gaussian state.

• Added Gaussian decompositions to the front-end; these can be accessed via the new quantum operations Interferometer, GaussianTransform, Gaussian. These allow you to apply interferometers, Gaussian symplectic transformations, and prepare a state based on a covariance matrix respectively. You can also query the engine to determine the CV gate decompositions applied.

• Added the cross-Kerr interaction, accessible via the quantum operation CKgate().

• Added utilities for creating random covariance, symplectic, and Gaussian unitary matrices in strawberryfields.utils.

• States can now be compared directly for equality - this is defined separately for Gaussian states and Fock basis states.

### Improvements

• The engine logic and behaviour has been overhauled, making it simpler to use and understand.

• eng.run() and eng.reset() now allow the user to alter parameters such as cutoff_dim between runs.

• eng.reset_backend() has been renamed to eng.reset(), and now also implicitly resets the queue.

• The engine can now be reset even in the case of modes having being added/deleted, with no side effects. This is due to the presence of register checkpoints, allowing the engine to keep track of register changes.

• eng.print_applied() keeps track of multiple simulation runs, by using nested lists.

• A new parameter class is introduced - this is a developmental change, and does not affect the user-facing parts of Strawberry Fields. All parameters passed to quantum operations are ‘wrapped’ in this parameter class, which also contains several high level mathematical and array/tensor manipulation functions and methods.

### Contributors

This release contains contributions from:

Ville Bergholm, Christian Gogolin, Nicolás Quesada, Josh Izaac, and Nathan Killoran.

## Release 0.7.3¶

### New features

• Added Gaussian decompositions to the front-end; these can be accessed via the new quantum operations Interferometer, GaussianTransform, CovarianceState. These allow you to apply interferometers, Gaussian symplectic transformations, and prepare a state based on a covariance matrix respectively. You can also query the engine to determine the CV gate decompositions applied.

• Added utilities for creating random covariance, symplectic, and gaussian unitary matrices in strawberryfields.utils.

### Improvements

• Created a separate package strawberryfields-gpu that requires tensorflow-gpu.

• Modified TFBackend to cache non-variable parts of the beamsplitter, to speed up computation.

• Minor performance improvement in fock_prob() by avoiding inverting a matrix twice.

### Bug fixes

• Fixed bug #10 by adding the ability to reset the Fock modeMap and GaussianCircuit class

• Fixed bug #11 by reshaping the Fock probabilities if the state happens to be pure states

• Fixed Clements decomposition bug where some phase angles weren’t applied

• Fixed typo in displaced squeezed formula in documentation

• Fix to prevent beamsplitter prefactor cache from breaking things if using two graphs

• Fix bug #13, GaussianBackend.state() raises an IndexError if all modes in the state have been deleted.

## Release 0.7.2¶

### Bug fixes

• Fixed Tensorflow requirements in setup.py, so that installation will now work for versions of tensorflow>=1.3,<1.7

### Known issues

• Tensorflow version 1.7 introduces some breaking API changes, so is currently not supported by Strawberry Fields.

## Release 0.7.1¶

Initial public release.

### Contributors

This release contains contributions from:

Nathan Killoran, Josh Izaac, Nicolás Quesada, Matthew Amy, and Ville Bergholm.