Skip to content

model_classes module

NestedLogitPrimitives dataclass

Source code in cupid_matching/model_classes.py
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
@dataclass
class NestedLogitPrimitives:
    Phi: np.ndarray
    n: np.ndarray
    m: np.ndarray
    nests_for_each_x: (
        NestsList  # given by user, e.g. [[1, 3], [2,4]] has y=1 and y=3 in first nest
    )
    nests_for_each_y: NestsList
    nests_over_Y: (
        NestsList  # rebased to zero: the above example becomes [[0, 2], [1,3]]
    )
    nests_over_X: NestsList
    i_nest_of_x: Nest  # mapping x -> n'
    i_nest_of_y: Nest  # mapping y -> n
    n_alphas: int
    mus: Matching | None = None
    true_alphas: np.ndarray | None = None

    def __init__(
        self,
        Phi: np.ndarray,
        n: np.ndarray,
        m: np.ndarray,
        nests_for_each_x: NestsList,
        nests_for_each_y: NestsList,
        true_alphas: np.ndarray | None = None,
    ):
        """
        We only model two-level nested logit, with {0} as the first nest,
        and nests and nests parameters that do not depend on the type.

        Args:
            Phi: the (X,Y) joint surplus matrix
            n: the X-vector of men margins
            m: the X-vector of women margins
            nests_for_each_x: the composition of the nests over 1...Y, a list of r lists
            nests_for_each_y: the composition of the nests over 1...X, a list of d lists
            true_alphas: the true nest parameters, if any; should be an (r+d)-vector
        """
        X, Y = check_matrix(Phi)
        Xn = check_vector(n)
        Ym = check_vector(m)

        # we need to rebase the indices to zero
        self.nests_over_X = change_indices(nests_for_each_y)
        self.nests_over_Y = change_indices(nests_for_each_x)

        self.n_alphas = len(nests_for_each_y) + len(nests_for_each_x)

        if Xn != X:
            bs_error_abort(f"Phi is a ({X}, {Y}) matrix but n has {Xn} elements.")
        if Ym != Y:
            bs_error_abort(f"Phi is a ({X}, {Y}) matrix but m has {Ym} elements.")

        if true_alphas is not None:
            alpha_size = check_vector(true_alphas)
            if alpha_size != self.n_alphas:
                bs_error_abort(
                    f"true_alphas shoud have {self.n_alphas} elements, not {alpha_size}"
                )

        self.Phi = Phi
        self.n = n
        self.m = m
        self.true_alphas = true_alphas
        self.nests_for_each_x = nests_for_each_x
        self.nests_for_each_y = nests_for_each_y

        # check that every x is in a nest, and just once
        nests_check = []
        i_nest_of_x = np.zeros(X, int)
        for x in range(X):
            i_nest_of_x[x] = find_nest_of(self.nests_over_X, x)
            nests_check.append(i_nest_of_x[x])
        if -1 in nests_check or len(set(nests_check)) != len(nests_for_each_y):
            bs_error_abort("Check your nests_for_each_y")
        # check that every y is in a nest, and just once
        nests_check = []
        i_nest_of_y = np.zeros(Y, int)
        for y in range(Y):
            i_nest_of_y[y] = find_nest_of(self.nests_over_Y, y)
            nests_check.append(i_nest_of_y[y])
        if -1 in nests_check or len(set(nests_check)) != len(nests_for_each_x):
            bs_error_abort("Check your nests_for_each_x")

        self.i_nest_of_x = i_nest_of_x.tolist()
        self.i_nest_of_y = i_nest_of_y.tolist()

    def __str__(self):
        X, Y = self.Phi.shape
        nmen, nwomen = np.sum(self.n), np.sum(self.m)
        repr_str = (
            f"This is a 2-level nested logit with {nmen} men of {X} types"
            + f" and {nwomen} women of {Y} types.\n"
        )
        repr_str += (
            f" We have {self.n_nests_over_Y} nests over 1...Y "
            + f" and {self.n_nests_over_X} nests over 1...X,\n"
        )
        if self.true_alphas is None:
            repr_str += "     with unspecified nests parameters."
        else:
            alpha_vals = self.true_alphas
            repr_str += "     with respective nests parameters:\n"
            repr_str += f"    {alpha_vals[:self.n_nests_over_Y]}\n"
            repr_str += f" and {alpha_vals[self.n_nests_over_Y:]}\n"
        print_stars(repr_str)

    def ipfp_nested_logit_solver(
        self, tol: float = 1e-9, verbose: bool = False, maxiter: int = 1000
    ) -> tuple[Matching, np.ndarray, np.ndarray]:
        """Solves for equilibrium in a two-level nested logit market
        given systematic surplus and margins and nests parameters;
        does not compute_ the gradient of the matching patterns

        Args:
            tol: tolerance on change in solution
            verbose: if `True`, prints information
            maxiter: maximum number of iterations

        Returns:
             the matching patterns
             marg_err_x, marg_err_y: the errors on the margins
        """
        alphas = self.true_alphas
        if alphas is None:
            bs_error_abort("cannot solve without nest parameters")
        else:
            alphas = cast(np.ndarray, alphas)
            n_rhos = len(self.nests_over_Y)
            n_deltas = len(self.nests_over_X)
            rhos = alphas[:n_rhos]
            deltas = alphas[n_rhos:]

        #############################################################################
        # we solve the equilibrium equations
        #   starting with a reasonable initial point  muxy, mux0, mu0y = bigc
        #   it is important that it fit the number of individuals
        #############################################################################

        n, m = self.n, self.m
        X, Y = n.size, m.size

        nests_over_X, nests_over_Y = self.nests_over_X, self.nests_over_Y
        i_nest_of_x, i_nest_of_y = self.i_nest_of_x, self.i_nest_of_y

        rho_vals = rhos[i_nest_of_y]  # rho(n) for y in n in the paper
        delta_vals = deltas[i_nest_of_x]  # delta(n') for x in n' in the paper

        ephi = npexp(self.Phi / np.add.outer(delta_vals, rho_vals))

        # initial values
        nindivs = np.sum(n) + np.sum(m)
        bigc = nindivs / (X + Y + 2.0 * np.sum(ephi))

        mux0, mu0y, muxy = (
            np.full(X, bigc),
            np.full(Y, bigc),
            np.full((X, Y), bigc),
        )
        muxn = np.zeros((X, n_rhos))
        for i_nest_y, nest_y in enumerate(nests_over_Y):
            muxn[:, i_nest_y] = np.sum(muxy[:, nest_y], 1)
        muny = np.zeros((n_deltas, Y))
        for i_nest_x, nest_x in enumerate(nests_over_X):
            muny[i_nest_x, :] = np.sum(muxy[nest_x, :], 0)

        err_diff = bigc
        tol_diff = tol * bigc
        tol_newton = tol
        max_newton = 2000
        MIN_REST = 1e-4 * bigc  # used to bound mus below in the Newton iterations

        niter = 0
        while (err_diff > tol_diff) and (niter < maxiter):  # IPFP main loop
            # Newton iterates for men
            err_newton = bigc
            i_newton = 0
            while err_newton > tol_newton:
                gbar = np.zeros(
                    (X, n_rhos)
                )  # this will be the $\bar{G}^x_n$ of the note
                gbar_pow = np.zeros((X, n_rhos))
                biga = np.zeros(X)  # this will be the $A_x$ of the note
                for i_nest_x, nest_x in enumerate(nests_over_X):
                    # i_nest_x is n' in the paper
                    delta_x = deltas[i_nest_x]
                    muny_x = muny[i_nest_x, :]  # mu(n', :)
                    for x in nest_x:
                        ephi_x = ephi[x, :]
                        for i_nest_y, nest_y in enumerate(nests_over_Y):
                            # i_nest_y is n in the paper
                            mu_n = muny_x[nest_y]
                            mu0_n = mu0y[nest_y]
                            evec_n = ephi_x[nest_y]
                            rho_n = rhos[i_nest_y]
                            sum_rd = rho_n + delta_x
                            mun_term = nppow(mu_n, (delta_x - 1.0) / sum_rd)
                            mu0_term = nppow(mu0_n, 1.0 / sum_rd)
                            gbar[x, i_nest_y] = np.sum(mun_term * mu0_term * evec_n)
                            gbar_pow[x, i_nest_y] = nppow(
                                gbar[x, i_nest_y], sum_rd / (delta_x + 1.0)
                            )
                            biga[x] += gbar_pow[x, i_nest_y]

                # now we take one Newton step for all types of men
                delta_vals1 = 1.0 + delta_vals
                mux0_term = nppow(mux0, 1.0 / delta_vals1)
                bigb = mux0_term * biga  # this is the $B_x$ of the note
                numer = n * delta_vals1 - delta_vals * bigb
                lower_bound = np.full(X, MIN_REST)
                mux0_new = mux0 * np.maximum(
                    numer / (delta_vals1 * mux0 + bigb), lower_bound
                )
                muxn_new = gbar_pow * mux0_term.reshape((-1, 1))

                mux0 = mux0_new
                muxn = muxn_new
                errxi = mux0 + np.sum(muxn, 1) - n
                err_newton = npmaxabs(errxi)
                i_newton += 1
                if i_newton > max_newton:
                    bs_error_abort(
                        f"Newton solver failed for men after {max_newton} iterations"
                    )

            if verbose:
                print(
                    f"Newton error on men is {err_newton} after {i_newton} iterations"
                )

            # Newton iterates for women
            err_newton = bigc
            i_newton = 0
            while err_newton > tol_newton:
                gbar = np.zeros((Y, n_deltas))
                gbar_pow = np.zeros((Y, n_deltas))
                biga = np.zeros(Y)
                for i_nest_y, nest_y in enumerate(nests_over_Y):
                    # i_nest_y is n in the paper
                    rho_y = rhos[i_nest_y]
                    muxn_y = muxn[:, i_nest_y]  # mu(:, n)
                    for y in nest_y:
                        ephi_y = ephi[:, y]
                        for i_nest_x, nest_x in enumerate(nests_over_X):
                            mu_n = muxn_y[nest_x]
                            mu0_n = mux0[nest_x]
                            evec_n = ephi_y[nest_x]
                            delta_n = deltas[i_nest_x]
                            sum_rd = rho_y + delta_n
                            mun_term = nppow(mu_n, (rho_n - 1.0) / sum_rd)
                            mu0_term = nppow(mu0_n, 1.0 / sum_rd)
                            gbar[y, i_nest_x] = np.sum(mun_term * mu0_term * evec_n)
                            gbar_pow[y, i_nest_x] = nppow(
                                gbar[y, i_nest_x], sum_rd / (1.0 + rho_y)
                            )
                            biga[y] += gbar_pow[y, i_nest_x]

                # now we take one Newton step for all types of women
                rho_vals1 = 1.0 + rho_vals
                mu0y_term = nppow(mu0y, 1.0 / rho_vals1)
                bigb = mu0y_term * biga
                numer = m * rho_vals1 - rho_vals * bigb
                lower_bound = np.full(Y, MIN_REST)
                mu0y_new = mu0y * np.maximum(
                    numer / (rho_vals1 * mu0y + bigb), lower_bound
                )
                muny_new = gbar_pow.T * mu0y_term

                mu0y = mu0y_new
                muny = muny_new
                erryi = mu0y + np.sum(muny, 0) - m
                err_newton = npmaxabs(erryi)
                i_newton += 1
                if i_newton > max_newton:
                    bs_error_abort(
                        f"Newton solver failed for women after {max_newton} iterations"
                    )

            if verbose:
                print(
                    f"Newton error on women is {err_newton} after {i_newton} iterations"
                )

            muxy = np.zeros((X, Y))
            for x in range(X):
                i_nest_x = i_nest_of_x[x]  # n'
                ephi_x = ephi[x, :]
                mux0_x = mux0[x]
                muxn_x = muxn[x, :]
                delta_x = delta_vals[x]
                muny_x = muny[i_nest_x, :]
                for y in range(Y):
                    i_nest_y = i_nest_of_y[y]  # n
                    mu0y_y = mu0y[y]
                    rho_y = rho_vals[y]
                    muxn_xy = muxn_x[i_nest_y]
                    muny_xy = muny_x[y]
                    mu_term = (
                        mux0_x
                        * mu0y_y
                        * (muxn_xy ** (rho_y - 1.0))
                        * (muny_xy ** (delta_x - 1.0))
                    )
                    muxy[x, y] = ephi_x[y] * (mu_term ** (1.0 / (delta_x + rho_y)))

            n_sim, m_sim = compute_margins(muxy, mux0, mu0y)
            marg_err_x, marg_err_y = n_sim - n, m_sim - m

            if verbose:
                print(
                    f"Margin error on men is {marg_err_x} "
                    f" after {niter} IPFP iterations"
                )
                print(
                    f"Margin error on women is {marg_err_y} "
                    f" after {niter} IPFP iterations"
                )
            err_diff = npmaxabs(marg_err_x) + npmaxabs(marg_err_y)
            niter += 1

        n_sim, m_sim = compute_margins(muxy, mux0, mu0y)
        marg_err_x = n_sim - n
        marg_err_y = m_sim - m

        if verbose:
            print(
                f"Margin error on men is {npmaxabs(marg_err_x)} after {niter} IPFP"
                " iterations"
            )
            print(
                f"Margin error on women is {npmaxabs(marg_err_y)} after {niter} IPFP"
                " iterations"
            )

        return Matching(muxy, n, m), marg_err_x, marg_err_y

    def ipfp_solve(self) -> Matching:
        if self.true_alphas is None:
            bs_error_abort(
                "true_alphas must be specified to solve the nested logit by IPFP."
            )
        self.mus, err_x, err_y = self.ipfp_nested_logit_solver(verbose=False)
        return self.mus

    def simulate(self, n_households: int, seed: int | None = None) -> Matching:
        self.mus = self.ipfp_solve()
        mus_sim = simulate_sample_from_mus(self.mus, n_households, seed=seed)
        return mus_sim

__init__(Phi, n, m, nests_for_each_x, nests_for_each_y, true_alphas=None)

We only model two-level nested logit, with {0} as the first nest, and nests and nests parameters that do not depend on the type.

Parameters:

Name Type Description Default
Phi np.ndarray

the (X,Y) joint surplus matrix

required
n np.ndarray

the X-vector of men margins

required
m np.ndarray

the X-vector of women margins

required
nests_for_each_x NestsList

the composition of the nests over 1...Y, a list of r lists

required
nests_for_each_y NestsList

the composition of the nests over 1...X, a list of d lists

required
true_alphas np.ndarray | None

the true nest parameters, if any; should be an (r+d)-vector

None
Source code in cupid_matching/model_classes.py
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
def __init__(
    self,
    Phi: np.ndarray,
    n: np.ndarray,
    m: np.ndarray,
    nests_for_each_x: NestsList,
    nests_for_each_y: NestsList,
    true_alphas: np.ndarray | None = None,
):
    """
    We only model two-level nested logit, with {0} as the first nest,
    and nests and nests parameters that do not depend on the type.

    Args:
        Phi: the (X,Y) joint surplus matrix
        n: the X-vector of men margins
        m: the X-vector of women margins
        nests_for_each_x: the composition of the nests over 1...Y, a list of r lists
        nests_for_each_y: the composition of the nests over 1...X, a list of d lists
        true_alphas: the true nest parameters, if any; should be an (r+d)-vector
    """
    X, Y = check_matrix(Phi)
    Xn = check_vector(n)
    Ym = check_vector(m)

    # we need to rebase the indices to zero
    self.nests_over_X = change_indices(nests_for_each_y)
    self.nests_over_Y = change_indices(nests_for_each_x)

    self.n_alphas = len(nests_for_each_y) + len(nests_for_each_x)

    if Xn != X:
        bs_error_abort(f"Phi is a ({X}, {Y}) matrix but n has {Xn} elements.")
    if Ym != Y:
        bs_error_abort(f"Phi is a ({X}, {Y}) matrix but m has {Ym} elements.")

    if true_alphas is not None:
        alpha_size = check_vector(true_alphas)
        if alpha_size != self.n_alphas:
            bs_error_abort(
                f"true_alphas shoud have {self.n_alphas} elements, not {alpha_size}"
            )

    self.Phi = Phi
    self.n = n
    self.m = m
    self.true_alphas = true_alphas
    self.nests_for_each_x = nests_for_each_x
    self.nests_for_each_y = nests_for_each_y

    # check that every x is in a nest, and just once
    nests_check = []
    i_nest_of_x = np.zeros(X, int)
    for x in range(X):
        i_nest_of_x[x] = find_nest_of(self.nests_over_X, x)
        nests_check.append(i_nest_of_x[x])
    if -1 in nests_check or len(set(nests_check)) != len(nests_for_each_y):
        bs_error_abort("Check your nests_for_each_y")
    # check that every y is in a nest, and just once
    nests_check = []
    i_nest_of_y = np.zeros(Y, int)
    for y in range(Y):
        i_nest_of_y[y] = find_nest_of(self.nests_over_Y, y)
        nests_check.append(i_nest_of_y[y])
    if -1 in nests_check or len(set(nests_check)) != len(nests_for_each_x):
        bs_error_abort("Check your nests_for_each_x")

    self.i_nest_of_x = i_nest_of_x.tolist()
    self.i_nest_of_y = i_nest_of_y.tolist()

ipfp_nested_logit_solver(tol=1e-09, verbose=False, maxiter=1000)

Solves for equilibrium in a two-level nested logit market given systematic surplus and margins and nests parameters; does not compute_ the gradient of the matching patterns

Parameters:

Name Type Description Default
tol float

tolerance on change in solution

1e-09
verbose bool

if True, prints information

False
maxiter int

maximum number of iterations

1000

Returns:

Type Description
Matching

the matching patterns

np.ndarray

marg_err_x, marg_err_y: the errors on the margins

Source code in cupid_matching/model_classes.py
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
def ipfp_nested_logit_solver(
    self, tol: float = 1e-9, verbose: bool = False, maxiter: int = 1000
) -> tuple[Matching, np.ndarray, np.ndarray]:
    """Solves for equilibrium in a two-level nested logit market
    given systematic surplus and margins and nests parameters;
    does not compute_ the gradient of the matching patterns

    Args:
        tol: tolerance on change in solution
        verbose: if `True`, prints information
        maxiter: maximum number of iterations

    Returns:
         the matching patterns
         marg_err_x, marg_err_y: the errors on the margins
    """
    alphas = self.true_alphas
    if alphas is None:
        bs_error_abort("cannot solve without nest parameters")
    else:
        alphas = cast(np.ndarray, alphas)
        n_rhos = len(self.nests_over_Y)
        n_deltas = len(self.nests_over_X)
        rhos = alphas[:n_rhos]
        deltas = alphas[n_rhos:]

    #############################################################################
    # we solve the equilibrium equations
    #   starting with a reasonable initial point  muxy, mux0, mu0y = bigc
    #   it is important that it fit the number of individuals
    #############################################################################

    n, m = self.n, self.m
    X, Y = n.size, m.size

    nests_over_X, nests_over_Y = self.nests_over_X, self.nests_over_Y
    i_nest_of_x, i_nest_of_y = self.i_nest_of_x, self.i_nest_of_y

    rho_vals = rhos[i_nest_of_y]  # rho(n) for y in n in the paper
    delta_vals = deltas[i_nest_of_x]  # delta(n') for x in n' in the paper

    ephi = npexp(self.Phi / np.add.outer(delta_vals, rho_vals))

    # initial values
    nindivs = np.sum(n) + np.sum(m)
    bigc = nindivs / (X + Y + 2.0 * np.sum(ephi))

    mux0, mu0y, muxy = (
        np.full(X, bigc),
        np.full(Y, bigc),
        np.full((X, Y), bigc),
    )
    muxn = np.zeros((X, n_rhos))
    for i_nest_y, nest_y in enumerate(nests_over_Y):
        muxn[:, i_nest_y] = np.sum(muxy[:, nest_y], 1)
    muny = np.zeros((n_deltas, Y))
    for i_nest_x, nest_x in enumerate(nests_over_X):
        muny[i_nest_x, :] = np.sum(muxy[nest_x, :], 0)

    err_diff = bigc
    tol_diff = tol * bigc
    tol_newton = tol
    max_newton = 2000
    MIN_REST = 1e-4 * bigc  # used to bound mus below in the Newton iterations

    niter = 0
    while (err_diff > tol_diff) and (niter < maxiter):  # IPFP main loop
        # Newton iterates for men
        err_newton = bigc
        i_newton = 0
        while err_newton > tol_newton:
            gbar = np.zeros(
                (X, n_rhos)
            )  # this will be the $\bar{G}^x_n$ of the note
            gbar_pow = np.zeros((X, n_rhos))
            biga = np.zeros(X)  # this will be the $A_x$ of the note
            for i_nest_x, nest_x in enumerate(nests_over_X):
                # i_nest_x is n' in the paper
                delta_x = deltas[i_nest_x]
                muny_x = muny[i_nest_x, :]  # mu(n', :)
                for x in nest_x:
                    ephi_x = ephi[x, :]
                    for i_nest_y, nest_y in enumerate(nests_over_Y):
                        # i_nest_y is n in the paper
                        mu_n = muny_x[nest_y]
                        mu0_n = mu0y[nest_y]
                        evec_n = ephi_x[nest_y]
                        rho_n = rhos[i_nest_y]
                        sum_rd = rho_n + delta_x
                        mun_term = nppow(mu_n, (delta_x - 1.0) / sum_rd)
                        mu0_term = nppow(mu0_n, 1.0 / sum_rd)
                        gbar[x, i_nest_y] = np.sum(mun_term * mu0_term * evec_n)
                        gbar_pow[x, i_nest_y] = nppow(
                            gbar[x, i_nest_y], sum_rd / (delta_x + 1.0)
                        )
                        biga[x] += gbar_pow[x, i_nest_y]

            # now we take one Newton step for all types of men
            delta_vals1 = 1.0 + delta_vals
            mux0_term = nppow(mux0, 1.0 / delta_vals1)
            bigb = mux0_term * biga  # this is the $B_x$ of the note
            numer = n * delta_vals1 - delta_vals * bigb
            lower_bound = np.full(X, MIN_REST)
            mux0_new = mux0 * np.maximum(
                numer / (delta_vals1 * mux0 + bigb), lower_bound
            )
            muxn_new = gbar_pow * mux0_term.reshape((-1, 1))

            mux0 = mux0_new
            muxn = muxn_new
            errxi = mux0 + np.sum(muxn, 1) - n
            err_newton = npmaxabs(errxi)
            i_newton += 1
            if i_newton > max_newton:
                bs_error_abort(
                    f"Newton solver failed for men after {max_newton} iterations"
                )

        if verbose:
            print(
                f"Newton error on men is {err_newton} after {i_newton} iterations"
            )

        # Newton iterates for women
        err_newton = bigc
        i_newton = 0
        while err_newton > tol_newton:
            gbar = np.zeros((Y, n_deltas))
            gbar_pow = np.zeros((Y, n_deltas))
            biga = np.zeros(Y)
            for i_nest_y, nest_y in enumerate(nests_over_Y):
                # i_nest_y is n in the paper
                rho_y = rhos[i_nest_y]
                muxn_y = muxn[:, i_nest_y]  # mu(:, n)
                for y in nest_y:
                    ephi_y = ephi[:, y]
                    for i_nest_x, nest_x in enumerate(nests_over_X):
                        mu_n = muxn_y[nest_x]
                        mu0_n = mux0[nest_x]
                        evec_n = ephi_y[nest_x]
                        delta_n = deltas[i_nest_x]
                        sum_rd = rho_y + delta_n
                        mun_term = nppow(mu_n, (rho_n - 1.0) / sum_rd)
                        mu0_term = nppow(mu0_n, 1.0 / sum_rd)
                        gbar[y, i_nest_x] = np.sum(mun_term * mu0_term * evec_n)
                        gbar_pow[y, i_nest_x] = nppow(
                            gbar[y, i_nest_x], sum_rd / (1.0 + rho_y)
                        )
                        biga[y] += gbar_pow[y, i_nest_x]

            # now we take one Newton step for all types of women
            rho_vals1 = 1.0 + rho_vals
            mu0y_term = nppow(mu0y, 1.0 / rho_vals1)
            bigb = mu0y_term * biga
            numer = m * rho_vals1 - rho_vals * bigb
            lower_bound = np.full(Y, MIN_REST)
            mu0y_new = mu0y * np.maximum(
                numer / (rho_vals1 * mu0y + bigb), lower_bound
            )
            muny_new = gbar_pow.T * mu0y_term

            mu0y = mu0y_new
            muny = muny_new
            erryi = mu0y + np.sum(muny, 0) - m
            err_newton = npmaxabs(erryi)
            i_newton += 1
            if i_newton > max_newton:
                bs_error_abort(
                    f"Newton solver failed for women after {max_newton} iterations"
                )

        if verbose:
            print(
                f"Newton error on women is {err_newton} after {i_newton} iterations"
            )

        muxy = np.zeros((X, Y))
        for x in range(X):
            i_nest_x = i_nest_of_x[x]  # n'
            ephi_x = ephi[x, :]
            mux0_x = mux0[x]
            muxn_x = muxn[x, :]
            delta_x = delta_vals[x]
            muny_x = muny[i_nest_x, :]
            for y in range(Y):
                i_nest_y = i_nest_of_y[y]  # n
                mu0y_y = mu0y[y]
                rho_y = rho_vals[y]
                muxn_xy = muxn_x[i_nest_y]
                muny_xy = muny_x[y]
                mu_term = (
                    mux0_x
                    * mu0y_y
                    * (muxn_xy ** (rho_y - 1.0))
                    * (muny_xy ** (delta_x - 1.0))
                )
                muxy[x, y] = ephi_x[y] * (mu_term ** (1.0 / (delta_x + rho_y)))

        n_sim, m_sim = compute_margins(muxy, mux0, mu0y)
        marg_err_x, marg_err_y = n_sim - n, m_sim - m

        if verbose:
            print(
                f"Margin error on men is {marg_err_x} "
                f" after {niter} IPFP iterations"
            )
            print(
                f"Margin error on women is {marg_err_y} "
                f" after {niter} IPFP iterations"
            )
        err_diff = npmaxabs(marg_err_x) + npmaxabs(marg_err_y)
        niter += 1

    n_sim, m_sim = compute_margins(muxy, mux0, mu0y)
    marg_err_x = n_sim - n
    marg_err_y = m_sim - m

    if verbose:
        print(
            f"Margin error on men is {npmaxabs(marg_err_x)} after {niter} IPFP"
            " iterations"
        )
        print(
            f"Margin error on women is {npmaxabs(marg_err_y)} after {niter} IPFP"
            " iterations"
        )

    return Matching(muxy, n, m), marg_err_x, marg_err_y