def size_test():
n = 50
m = 10
d = 5
K = 4
print("\n\nRunning Size Test")
train_X = generate_data(n, m, 'uniform')
L = generate_loss_matrix(d, m, 'uniform')
U = generate_utility_matrix(d, m, 'uniform')
U_X = np.matmul(train_X, U.T)
max_val = np.amax(U_X)
learned_betas, learned_predictions, _, alphas = train_erm(train_X, L)
final_loss = compute_final_loss(alphas, L, train_X, learned_predictions)
welfare = compute_welfare(alphas, U, train_X, learned_predictions)
opt_loss = get_optimal_loss(L, train_X)
print("Optimal loss is: ", opt_loss)
print("ERM loss is: ", final_loss)
print("Welfare is: ", welfare)
#conjugate_UX = max_val*np.ones((n, d)) - U_X
#learned_betas, learned_predictions, _, alphas = train_erm(train_X, conjugate_UX)
#best_welfare = compute_welfare(alphas, U, train_X, learned_predictions)
#print("Best possible welfare is: ", best_welfare)
learned_betas, learned_predictions, _, opt_alphas = train_erm_welfare(train_X, L, U, lamb=10)
final_loss = compute_final_loss(opt_alphas, L, train_X, learned_predictions)
welfare = compute_welfare(opt_alphas, U, train_X, learned_predictions)
print("ERM with welfre loss is: ", final_loss)
print("Welfare is: ", welfare)
def size_test():
n = 50
m = 14
d = 5
K = 4
print("\nRunning Size Test")
train_X = generate_data(n, m, 'uniform')
train_X = normalize(train_X, axis=1, norm='l1')
group_dist = [0.25, 0.25, 0.25, 0.25]
samples_by_group = define_groups(train_X, group_dist)
L = generate_loss_matrix(d, m, 'uniform')
U = generate_utility_matrix_var(d, m, 'uniform', 0)
erm_betas, learned_predictions, learned_pred_group, alphas = train_erm(train_X, L, U, \
samples_by_group)
final_loss = compute_final_loss(alphas, L, train_X, learned_predictions)
avg_envy, violations = total_average_envy(alphas, U, train_X,
learned_predictions)
opt_loss = get_optimal_loss(L, train_X)
print("Optimal loss is: ", opt_loss)
print("ERM loss is: ", final_loss)
print("ERM average envy: ", avg_envy, "ERM total violations: ", violations)
start_time = time.time()
cons_betas, learned_predictions, learned_pred_group, opt_alphas = \
train_erm_envy_free(train_X, L, U, samples_by_group, lamb=100)
final_loss = compute_final_loss(opt_alphas, L, train_X,
learned_predictions)
avg_envy, violations = total_average_envy(opt_alphas, U, train_X,
learned_predictions)
print("ERM-Envy Free loss is: ", final_loss)
print("ERM-Envy Free average envy: ", avg_envy, "ERM total violations: ",
violations)
end_time = time.time()
print("Time is: ", end_time - start_time)
print(learned_predictions)
test_X = generate_data(n, m, 'uniform')
group_dist = [0.25, 0.25, 0.25, 0.25]
test_s_by_group = define_groups(test_X, group_dist)
st_learned_predictions, st_learned_pred_group = \
get_all_predictions(erm_betas, test_X, test_s_by_group, K)
st_avg_envy, st_envy_violations = total_average_envy(opt_alphas, U, train_X, \
st_learned_predictions)
#print("ERM get this much envy on test: ", st_avg_envy, "Violations: ", st_envy_violations)
st_learned_predictions, st_learned_pred_group = \
get_all_predictions(cons_betas, test_X, test_s_by_group, K)
st_avg_envy, st_envy_violations = total_average_envy(opt_alphas, U, test_X, \
st_learned_predictions)
def size_test():
n = 60
m = 16
d = 5
K = 4
print("\nRunning Size Test")
train_X = generate_data(n, m, 'uniform')
group_dist = [0.25, 0.25, 0.25, 0.25]
samples_by_group = define_groups(train_X, group_dist)
L = generate_loss_matrix(d, m, 'uniform')
U = generate_utility_matrix(d, m, 'uniform')
erm_betas, learned_predictions, learned_pred_group, alphas = train_erm(train_X, L, U, \
samples_by_group)
final_loss = compute_final_loss(alphas, L, train_X, learned_predictions)
total_equi, violations = total_group_equi(alphas, U, samples_by_group,
learned_pred_group)
opt_loss = get_optimal_loss(L, train_X)
print("Optimal loss is: ", opt_loss)
print("ERM loss is: ", final_loss)
print("ERM total equi: ", total_equi, "ERM total violations: ", violations)
start_time = time.time()
cons_betas, learned_predictions, learned_pred_group, opt_alphas = \
train_erm_equi(train_X, L, U, samples_by_group, lamb=3)
final_loss = compute_final_loss(opt_alphas, L, train_X,
learned_predictions)
total_equi, violations = total_group_equi(alphas, U, samples_by_group,
learned_pred_group)
print("ERM-equi loss is: ", final_loss)
print("ERM-equi total equi: ", total_equi, "ERM total violations: ",
violations)
end_time = time.time()
print("Time is: ", end_time - start_time)
test_X = generate_data(100, m, 'uniform')
group_dist = [0.25, 0.25, 0.25, 0.25]
test_s_by_group = define_groups(test_X, group_dist)
st_learned_predictions, st_learned_pred_group = \
get_all_predictions(erm_betas, test_X, test_s_by_group, K)
st_total_equi, st_envy_violations = total_group_equi(opt_alphas, U, test_s_by_group, \
st_learned_pred_group)
print("ERM get this much equi diff on test: ", st_total_equi)
st_learned_predictions, st_learned_pred_group = \
get_all_predictions(cons_betas, test_X, test_s_by_group, K)
st_total_equi, st_envy_violations = total_group_equi(opt_alphas, U, test_s_by_group, \
st_learned_pred_group)
print("ERM-Equi get this much equi diff on test: ", st_total_equi)
def size_test():
n = 60
m = 16
d = 5
K = 4
Lambda = 1000
print("\nRunning Size Test")
train_X = generate_data(n, m, 'uniform')
group_dist = [0.25, 0.25, 0.25, 0.25]
samples_by_group = define_groups(train_X, group_dist)
L = generate_loss_matrix(d, m, 'uniform')
U = generate_utility_matrix(d, m, 'uniform')
erm_betas, learned_predictions, learned_pred_group, alphas = train_erm(train_X, L, U, \
samples_by_group)
final_loss = compute_final_loss(alphas, L, train_X, learned_predictions)
min_welfare = min_group_welfare(alphas, U, samples_by_group,
learned_pred_group)
opt_loss = get_optimal_loss(L, train_X)
print("Optimal loss is: ", opt_loss)
print("ERM loss is: ", final_loss)
print("ERM min_welfare: ", min_welfare)
start_time = time.time()
cons_betas, learned_predictions, learned_predictions_group, opt_alphas = \
train_erm_min_welfare(train_X, L, U, samples_by_group, lamb=Lambda)
final_loss = compute_final_loss(opt_alphas, L, train_X,
learned_predictions)
min_welfare = min_group_welfare(opt_alphas, U, samples_by_group,
learned_predictions_group)
print("ERM-min_welfare loss is: ", final_loss)
print("ERM-min_welfare min welfare is: ", min_welfare)
end_time = time.time()
print("Time is: ", end_time - start_time)
test_X = generate_data(100, m, 'uniform')
group_dist = [0.25, 0.25, 0.25, 0.25]
test_s_by_group = define_groups(test_X, group_dist)
st_learned_predictions, st_learned_pred_group = \
get_all_predictions(erm_betas, test_X, test_s_by_group, K)
min_welfare = min_group_welfare(opt_alphas, U, test_s_by_group, \
st_learned_pred_group)
print("ERM get this much min welfare on test: ", min_welfare)
st_learned_predictions, st_learned_pred_group = \
get_all_predictions(cons_betas, test_X, test_s_by_group, K)
min_welfare = min_group_welfare(opt_alphas, U, test_s_by_group, \
st_learned_pred_group)
print("ERM-min welfare get this much min welfare on test: ", min_welfare)
def test_erm_equi():
Lambda = 10
print("Group Envy Free Test: ")
print("First compute ERM solution: ")
train_X = np.array([[0.8, 0.3, 1.5, 0.1], \
[0.3, 1.1, 1.7, 0.9], \
[1.0, 1.4, 0.5, 1.2],
[0.3, 0.5, 1.2, 1.3],
[1.0, 0.2, 0.7, 0.9],
[0.7, 1.5, 1.9, 0.3],
[0.2, 0.9, 1.7, 0.3],
[0.1, 0.2, 1.9, 1.3],
[0.7, 0.277, 0.9, 1.1],
[1.0, 1.2, 0.7, 0.9],
[0.1, 0.8, 0.3, 0.5], \
]) # 11 x 5
group_dist = [0.25, 0.25, 0.25, 0.25]
samples_by_group = define_groups(train_X, group_dist)
L = np.array([[0.3, 0.1, 0.4, 0.2],\
[0.7, 0.4, 0.1, 0.7],\
[0.3, 0.55, 0.7, 0.3],\
[0.4, 0.1, 0.4, 0.2]])
U = np.array([[.1, 0.3, 0.3, 0.9],\
[0.5, 0.9, 0.1, 0.5],\
[0.3, 0.55, 0.7, 0.3],\
[0.1, 0.9, 0.9, 0.1]])
learned_betas, learned_pred_all, learned_pred_group, opt_alphas = \
train_erm(train_X, L, U, samples_by_group)
total_equi, violations = total_group_equi(opt_alphas, U, samples_by_group,
learned_pred_group)
loss = compute_final_loss(opt_alphas, L, train_X, learned_pred_all)
optimal_loss = get_optimal_loss(L, train_X)
print("Optimal Loss is:", optimal_loss)
print("ERM loss is: ", loss)
print("ERM total equi diff: ", total_equi, "ERM total violations: ",
violations)
print("")
print("Now do wth Equitability constraint with lambda: ", Lambda)
learned_betas, learned_pred_all, learned_pred_group, opt_alphas = \
train_erm_equi(train_X, L, U, samples_by_group, lamb=Lambda)
total_equi, violations = total_group_equi(opt_alphas, U, samples_by_group, \
learned_pred_group)
loss = compute_final_loss(opt_alphas, L, train_X, learned_pred_all)
print("ERM-equi Loss is : ", loss)
print("ERM-equi total equi diff: ", total_equi,
"ERM-equi total violations: ", violations)
def test_erm_envy_free():
Lambda = 10
print("Envy Free Test: ")
print("First compute ERM solution: ")
train_X = np.array([[0.8, 0.3, 1.5, 0.1], \
[0.3, 1.1, 1.7, 0.9], \
[1.1, 1.4, 0.5, 1.2],
[0.3, 0.5, 1.2, 1.3],
[1.0, 0.2, 0.7, 0.9],
[0.7, 1.5, 1.9, 0.3],
[0.2, 0.9, 1.7, 0.3],
[0.1, 0.2, 1.9, 1.3],
[0.7, 0.277, 0.9, 1.1],
[1.0, 1.2, 0.7, 0.9],
[0.1, 0.8, 0.3, 0.5], \
]) # 11 x 5
group_dist = [0.25, 0.25, 0.25, 0.25]
samples_by_group = define_groups(train_X, group_dist)
L = np.array([[0.3, 0.1, 0.4, 0.2],\
[0.7, 0.4, 0.1, 0.7],\
[0.3, 0.55, 0.7, 0.3],\
[0.4, 0.1, 0.4, 0.2]])
U = np.array([[.1, 0.3, 0.3, 0.9],\
[0.5, 0.9, 0.1, 0.5],\
[0.3, 0.55, 0.7, 0.3],\
[0.1, 0.9, 0.9, 0.1]])
learned_betas, learned_pred_all, learned_pred_group, opt_alphas = \
train_erm(train_X, L, U, samples_by_group)
avg_envy, violations = total_average_envy(opt_alphas, U, train_X,
learned_pred_all)
loss = compute_final_loss(opt_alphas, L, train_X, learned_pred_all)
optimal_loss = get_optimal_loss(L, train_X)
print("Optimal Loss is:", optimal_loss)
print("ERM loss is: ", loss)
print("ERM total envy: ", avg_envy, "violations: ", violations)
print("")
print("Now do wth Envy Free constraint with lambda: ", Lambda)
learned_betas, learned_pred_all, learned_pred_group, opt_alphas = \
train_erm_envy_free(train_X, L, U, samples_by_group, lamb=Lambda)
avg_envy, violations = total_average_envy(opt_alphas, U, train_X,
learned_pred_all)
loss = compute_final_loss(opt_alphas, L, train_X, learned_pred_all)
print("ERM-Envy Free Loss is : ", loss)
print("ERM-Envy Free avg envy: ", avg_envy, "violations: ", violations)
def test_erm_equi():
Lambda = 1000
print("Group Envy Free Test: ")
print("First compute ERM solution: ")
train_X = np.array([[0.8, 0.3, 1.5, 0.1], \
[0.3, 1.1, 1.7, 0.9], \
[1.0, 1.4, 0.5, 1.2],
[0.3, 0.5, 1.2, 1.3],
[1.0, 0.2, 0.7, 0.9],
[0.7, 1.5, 1.9, 0.3],
[0.2, 0.9, 1.7, 0.3],
[0.1, 0.2, 1.9, 1.3],
[0.7, 0.277, 0.9, 1.1],
[1.0, 1.2, 0.7, 0.9],
[0.1, 0.8, 0.3, 0.5], \
]) # 11 x 5
group_dist = [0.25, 0.25, 0.25, 0.25]
samples_by_group = define_groups(train_X, group_dist)
L = np.array([[0.3, 0.1, 0.4, 0.2],\
[0.7, 0.4, 0.1, 0.7],\
[0.3, 0.55, 0.7, 0.3],\
[0.4, 0.1, 0.4, 0.2]])
U = np.array([[.1, 0.3, 0.3, 0.9],\
[0.5, 0.9, 0.1, 0.5],\
[0.3, 0.55, 0.7, 0.3],\
[0.1, 0.9, 0.9, 0.1]])
learned_betas, learned_pred_all, learned_pred_group, opt_alphas = \
train_erm(train_X, L, U, samples_by_group)
min_welfare = min_group_welfare(opt_alphas, U, samples_by_group,
learned_pred_group)
loss = compute_final_loss(opt_alphas, L, train_X, learned_pred_all)
optimal_loss = get_optimal_loss(L, train_X)
print("Optimal Loss is:", optimal_loss)
print("ERM loss is: ", loss)
print("ERM min welfare: ", min_welfare)
print("")
print("Now do with min welfare constraint with lambda: ", Lambda)
learned_betas, learned_pred_all, learned_pred_group, opt_alphas = \
train_erm_min_welfare(train_X, L, U, samples_by_group, lamb=Lambda)
cons_min_welfare = min_group_welfare(opt_alphas, U, samples_by_group, \
learned_pred_group)
loss = compute_final_loss(opt_alphas, L, train_X, learned_pred_all)
print("ERM-Min Welfare Loss is : ", loss)
print("ERM-Min Welfare min welfare: ", cons_min_welfare)
def test_erm_welfare():
print("Test 1: n=8, d=4, m=4, Loss mat, Utility is identity")
train_X = np.array([[0.4, 0.3, 0.5, 0.1], \
[0.3, 0.1, 0.7, 0.9], \
[1.1, 0.4, 0.5, 1.2],
[0.3, 0.5, 1.2, 1.3],
[1.3, 0.2, 0.7, 0.9],
[1.7, 1.5, 1.9, 0.3],
[2.2, 0.9, 1.7, 0.3],
[0.1, 0.2, 1.9, 1.3],
]) # 8 x 4
L = np.array([[0.3, 0.1, 0.4, 0.2],\
[0.2, 0.4, 0.1, 0.7],\
[0.3, 0.55, 0.7, 0.3],\
[0.4, 0.1, 0.4, 0.2]])
U = np.array([[0.4, 0.2, 0.3, 0.4],\
[0.5, 0.1, 0.9, 0.5],\
[0.3, 0.4, 0.6, 0.5],\
[0.2, 0.3, 0.6, 0.1]])
U_X = np.matmul(train_X, U.T)
n, d = U_X.shape
max_val = np.amax(U_X)
learned_betas, learned_predictions, _, opt_alphas = train_erm(train_X, L, U, groups=None, lamb=0)
final_loss = compute_final_loss(opt_alphas, L, train_X, learned_predictions)
welfare = compute_welfare(opt_alphas, U, train_X, learned_predictions)
opt_loss = get_optimal_loss(L, train_X)
print("Optimal loss is: ", opt_loss)
print("ERM loss is: ", final_loss)
print("Welfare is: ", welfare)
#conjugate_UX = max_val*np.ones((n, d)) - U_X
#learned_betas, learned_predictions, _, opt_alphas = train_erm(train_X, conjugate_UX)
#best_welfare = compute_welfare(opt_alphas, U, train_X, learned_predictions)
#print("Best possible welfare is: ", best_welfare)
learned_betas, learned_predictions, _, opt_alphas = train_erm_welfare(train_X, L, U, lamb=10)
final_loss = compute_final_loss(opt_alphas, L, train_X, learned_predictions)
welfare = compute_welfare(opt_alphas, U, train_X, learned_predictions)
print("ERM with welfre loss is: ", final_loss)
print("Welfare is: ", welfare)
