def run_perceptron(N, alpha, p_pos):
model = glm_generative(N=N,
alpha=alpha,
ensemble_type="gaussian",
prior_type="binary",
output_type="sgn",
prior_p_pos=p_pos)
scenario = BayesOptimalScenario(model, x_ids=["x"])
early = EarlyStopping()
records = scenario.run_all(max_iter=200, callback=early)
return records
def run_cs(N, alpha, ensemble_type, prior_rho):
model = glm_generative(
N=N, alpha=alpha, ensemble_type=ensemble_type,
prior_type="gauss_bernoulli", output_type="gaussian",
prior_rho=prior_rho, output_var=1e-11
)
scenario = BayesOptimalScenario(model, x_ids=["x"])
early = EarlyStopping()
records = scenario.run_all(
metrics=["mse"], max_iter=200, callback=early
)
return records
def run_EP(alpha, rho, seed):
model = glm_generative(N=2000,
alpha=alpha,
ensemble_type="gaussian",
prior_type="gauss_bernoulli",
output_type="abs",
prior_rho=rho,
prior_mean=0)
scenario = BayesOptimalScenario(model, x_ids=["x"])
scenario.setup(seed)
x_data = scenario.run_ep(max_iter=200, damping=0.2)
x_pred = x_data["x"]["r"]
mse = sign_symmetric_mse(x_pred, scenario.x_true["x"])
return dict(source="EP", v=mse)
def run_EP(alpha, rho, seed):
model = glm_generative(N=2000,
alpha=alpha,
ensemble_type="gaussian",
prior_type="gauss_bernoulli",
output_type="gaussian",
prior_rho=rho,
output_var=1e-10)
scenario = BayesOptimalScenario(model, x_ids=["x"])
scenario.setup(seed)
x_data = scenario.run_ep(max_iter=200)
x_pred = x_data["x"]["r"]
mse = mean_squared_error(x_pred, scenario.x_true["x"])
return dict(source="EP", v=mse)
def run_phase_retrieval(N, alpha, prior_mean):
model = glm_generative(N=N,
alpha=alpha,
ensemble_type="complex_gaussian",
prior_type="gauss_bernouilli",
output_type="modulus",
prior_mean=prior_mean,
prior_rho=0.5)
scenario = BayesOptimalScenario(model, x_ids=["x"])
early = EarlyStopping(wait_increase=10)
records = scenario.run_all(metrics=["mse", "phase_mse"],
max_iter=200,
damping=0.3,
callback=early)
return records
def mse_lasso(alpha, param_scaled, seed):
# create scenario
N, rho, noise_var = 1000, 0.05, 1e-2
M = int(alpha * N)
A = GaussianEnsemble(M=M, N=N).generate()
model = (GaussBernoulliPrior(size=N, rho=rho) @ V("x") @ LinearChannel(A)
@ V("z") @ GaussianChannel(var=noise_var) @ O("y")).to_model()
scenario = BayesOptimalScenario(model, x_ids=["x"])
scenario.setup(seed)
y = scenario.observations["y"]
# run lasso
param_scikit = noise_var * param_scaled / (M * rho)
lasso = Lasso(alpha=param_scikit)
lasso.fit(A, y)
x_pred = lasso.coef_
mse = mean_squared_error(x_pred, scenario.x_true["x"])
return mse
def run_benchmark(alpha, algo, seed):
# create scenario
N, rho, noise_var = 1000, 0.05, 1e-2
M = int(alpha * N)
A = GaussianEnsemble(M=M, N=N).generate()
t0 = time()
model = (GaussBernoulliPrior(size=N, rho=rho) @ V("x") @ LinearChannel(A)
@ V("z") @ GaussianChannel(var=noise_var) @ O("y")).to_model()
t1 = time()
record = {"svd_time": t1 - t0} # svd precomputation time
scenario = BayesOptimalScenario(model, x_ids=["x"])
scenario.setup(seed)
y = scenario.observations["y"]
# run algo
t0 = time()
if algo == "SE":
x_data = scenario.run_se(max_iter=1000, damping=0.1)
record["mse"] = x_data["x"]["v"]
record["n_iter"] = x_data["n_iter"]
if algo == "EP":
x_data = scenario.run_ep(max_iter=1000, damping=0.1)
x_pred = x_data["x"]["r"]
record["n_iter"] = x_data["n_iter"]
if algo == "LassoCV":
lasso = LassoCV(cv=5)
lasso.fit(A, y)
x_pred = lasso.coef_
record["param_scikit"] = lasso.alpha_
record["n_iter"] = lasso.n_iter_
if algo == "Lasso":
optim = pd.read_csv("optimal_param_lasso.csv")
param_scaled = np.interp(alpha, optim["alpha"], optim["param_scaled"])
param_scikit = noise_var * param_scaled / (M * rho)
lasso = Lasso(alpha=param_scikit)
lasso.fit(A, y)
x_pred = lasso.coef_
record["param_scikit"] = param_scikit
record["n_iter"] = lasso.n_iter_
if algo == "pymc3":
with pm.Model():
ber = pm.Bernoulli("ber", p=rho, shape=N)
nor = pm.Normal("nor", mu=0, sd=1, shape=N)
x = pm.Deterministic("x", ber * nor)
likelihood = pm.Normal("y",
mu=pm.math.dot(A, x),
sigma=np.sqrt(noise_var),
observed=y)
trace = pm.sample(draws=1000, chains=1, return_inferencedata=False)
x_pred = trace.get_values('x').mean(axis=0)
t1 = time()
record["time"] = t1 - t0
if algo != "SE":
record["mse"] = mean_squared_error(x_pred, scenario.x_true["x"])
return record
from tramp.experiments import BayesOptimalScenario, qplot, plot_compare
# %%
# Model
# -----
# We wish to infer the binary signal
# $x \sim \mathrm{Bin}( . | p_+) \in \pm^N$ from
# $y = \mathrm{sgn}(Fx) \in \pm^M$, where
# $F \in \mathbb{R}^{M \times N}$ is a Gaussian random matrix.
# You can build the perceptron directly, or use the `glm_generative` model builder.
teacher = glm_generative(N=1000,
alpha=1.7,
ensemble_type="gaussian",
prior_type="binary",
output_type="sgn")
scenario = BayesOptimalScenario(teacher, x_ids=["x"])
scenario.setup(seed=42)
scenario.student.plot()
# %%
# EP dynamics
ep_evo = scenario.ep_convergence(metrics=["mse"],
max_iter=30,
callback=EarlyStoppingEP())
qplot(ep_evo, x="iter", y=["mse", "v"], y_markers=[".", "-"], y_legend=True)
# %%
# Recovered signal
plot_compare(scenario.x_true["x"], scenario.x_pred["x"])
# %%
# %%
# Model
alpha = 1.6
N = 1000
teacher = glm_generative(N=N,
alpha=alpha,
ensemble_type="gaussian",
prior_type="binary",
output_type="door",
output_width=1.,
prior_p_pos=0.51)
for factor in teacher.factors:
print(factor)
scenario = BayesOptimalScenario(teacher, x_ids=["x", "z"])
scenario.setup(seed=42)
scenario.student.plot()
# %%
# EP dyanmics
ep_evo = scenario.ep_convergence(metrics=["mse", "sign_mse"],
damping=0.1,
max_iter=40)
qplot(ep_evo,
x="iter",
y=["mse", "sign_mse", "v"],
y_markers=["x", ".", "-"],
column="id",
y_legend=True)
plot_compare(scenario.x_true["x"], scenario.x_pred["x"])
import matplotlib.pyplot as plt
import pandas as pd
import numpy as np
# %%
# Model
np.random.seed(42)
model = glm_generative(N=1000,
alpha=2,
ensemble_type="complex_gaussian",
prior_type="gauss_bernouilli",
output_type="modulus",
prior_mean=0.01,
prior_rho=0.5)
scenario = BayesOptimalScenario(model, x_ids=["x"])
scenario.setup()
scenario.student.plot()
for factor in scenario.student.factors:
print(factor.id, factor)
# %%
# EP dyanmics
# Damping is needed
# really bad without damping
ep_evo = scenario.ep_convergence(metrics=["mse", "phase_mse"], max_iter=20)
qplot(ep_evo,
x="iter",
y=["phase_mse", "v"],
y_markers=["x", "-"],
import logging
logging.basicConfig(level=logging.INFO)
# %%
# Model
alpha = 2.
N = 1000
teacher = glm_generative(
N=N, alpha=alpha, ensemble_type="gaussian", prior_type="gauss_bernoulli",
output_type="relu", prior_rho=0.5
)
for factor in teacher.factors:
print(factor)
scenario = BayesOptimalScenario(teacher, x_ids=["x", "z"])
scenario.setup(seed=42)
scenario.student.plot()
# %%
# EP dyanmics
ep_evo = scenario.ep_convergence(metrics=["mse"], max_iter=10)
qplot(
ep_evo, x="iter", y=["mse", "v"],
y_markers=[".", "-"], column="id", y_legend=True
)
plot_compare(scenario.x_true["x"], scenario.x_pred["x"])
# %%
# MSE curve