Output files¶

QuanEstimation writes optimization and estimation results to files in the current working directory. Julia users receive the .dat (JLD2) result files together with f.csv. The Python interface converts the .dat files into .npy files after an optimization finishes; the .dat files are removed once they have been read. The objective-function history is written directly to f.csv by the Julia backend and is not converted.

Objective function values¶

The objective function is evaluated once per optimization episode. The values are written sequentially, one value per line, to f.csv.

Python

import numpy as np
f = np.loadtxt("f.csv")

Control optimization¶

The Python wrapper saves the control optimization results in controls.npy. The shape depends on savefile:

savefile=False (default): the final control coefficients are saved with shape (n_ctrl, n_ctrl_segments), where n_ctrl is the number of control Hamiltonians and n_ctrl_segments is the number of control segments.
savefile=True: the control coefficients after every optimization round are saved with shape (n_rounds, n_ctrl, n_ctrl_segments).

Python

import numpy as np
controls = np.load("controls.npy")

If the controls were exported as a flat CSV array, use csv2npy_controls to reshape them and save controls.npy:

from quanestimation import csv2npy_controls
csv2npy_controls(controls_csv_data, num=n_ctrl_segments)

State optimization¶

The Python wrapper saves the state optimization results in states.npy.

savefile=False (default): the final optimized state vector is saved.
savefile=True: the state vectors after every optimization round are saved.

Python

import numpy as np
states = np.load("states.npy")

If the states were exported as a flat CSV array, use csv2npy_states to regroup them and save states.npy:

from quanestimation import csv2npy_states
csv2npy_states(states_csv_data, num=1)

Measurement optimization¶

The Python wrapper saves the measurement optimization results in measurements.npy.

savefile=False (default): the final optimized POVM is saved as a list of measurement operators.
savefile=True: the POVM lists after every optimization round are saved.

Python

import numpy as np
M = np.load("measurements.npy", allow_pickle=True)

If the measurements were exported as a flat CSV array, use csv2npy_measurements to reshape them and save measurements.npy:

from quanestimation import csv2npy_measurements
csv2npy_measurements(M_csv_data, num=n_operators)

Comprehensive optimization¶

Comprehensive optimization combines several variable types. Each result file follows the same convention as the corresponding single-variable optimization:

controls are saved in controls.npy;
states are saved in states.npy;
measurements are saved in measurements.npy.

A file is present only if that variable type was optimized. For example, SC optimization produces controls.npy and states.npy, but not measurements.npy; SCM optimization produces all three.

Bayesian estimation¶

Bayes() and MLE() save their results with numpy.save, so the files are read with numpy.load.

Maximum a posteriori / mean estimation¶

Bayes() produces:

File	Contents	Shape
`pout.npy`	Posterior distribution. With `savefile=False` it is the final posterior; with `savefile=True` it contains the posterior after every iteration.	grid shape of `p`; `(n_iter, *grid_shape)` when `savefile=True`
`xout.npy`	Estimated parameter values for every iteration.	`(n_iter,)` for one parameter, `(n_iter, n_params)` for several parameters

The return value of Bayes() is (p, x_last): the final posterior distribution and the estimate from the final iteration. The complete history is in xout.npy.

Maximum likelihood estimation¶

MLE() produces:

File	Contents	Shape
`Lout.npy`	Log-likelihood values. With `savefile=False` it is the final log-likelihood; with `savefile=True` it contains the log-likelihood after every iteration.	grid shape of the parameter grid; `(n_iter, *grid_shape)` when `savefile=True`
`xout.npy`	Estimated parameter values for every iteration.	`(n_iter,)` for one parameter, `(n_iter, n_params)` for several parameters

Python

import numpy as np
pout = np.load("pout.npy")
xout = np.load("xout.npy")

Adaptive measurement schemes¶

The Adapt module writes CSV files. With savefile=False, pout.csv contains only the final posterior distribution while xout.csv and y.csv contain the results of all iterations. With savefile=True, a new row is appended for every iteration.

File	Contents
`pout.csv`	Posterior distribution; the final one, or one row per iteration when `savefile=True`
`xout.csv`	Estimated parameter value(s) at each iteration
`y.csv`	Experimental outcome at each iteration

Python

import numpy as np
pout = np.loadtxt("pout.csv")
xout = np.loadtxt("xout.csv")
y = np.loadtxt("y.csv")

Offline optimization with Adapt_MZI additionally writes deltaphi.csv (the optimized phase differences) and f.csv (the objective-function history).