def cb_update(val):
"""
Slider Event Callback Func.
"""
dist = ds.Beta(sAlpha.val, sBeta.val)
line2dObj.set_ydata(dist.pdf(x))
fig.canvas.draw_idle()
def run_bayesian_test(probs, config):
t_start = time.time()
beta = config.get("beta", 0.3)
prior = d.Beta(config["prior"]["params"][0],
config["prior"]["params"][1])
if beta < 0:
beta = (1-prior.cdf(config["s"])) * 0.9
config["adaptive"] = 1
else:
config["adaptive"] = 0
test = binomial_bayesian_test(
config["s"],
prior,
alpha=config["alpha"],
beta=beta,
N=config["N"],
steps=config.get("steps", 1))
output = run_experiment(test, probs, config)
output["beta"] = beta
output["time"] = time.time() - t_start
return output
def run_fixed_test_early_stopping_experiment(probs, config):
t_start = time.time()
prior = d.Beta(config["prior"]["params"][0],
config["prior"]["params"][1])
test = fixed_test_early_stopping(config["N"],
config["s"],
prior,
alpha=config["alpha"])
output = run_experiment(test, probs, config)
output["time"] = time.time() - t_start
return output
def run_msprt_experiment(probs, config):
t_start = time.time()
prior = d.Beta(config["prior"]["params"][0],
config["prior"]["params"][1])
test = lambda p: binomial_msprt_test(p,
config["s"],
prior,
alpha=config["alpha"],
N=config["N"],
steps=config.get("steps", 10))
output = run_experiment(test, probs, config)
output["time"] = time.time() - t_start
return output
def run_optimal_boundary_experiment(probs, config):
t_start = time.time()
prior = d.Beta(config["prior"]["params"][0],
config["prior"]["params"][1])
print("Finding boundary...")
boundary, ET = solve_binomial(config["s"],
prior, config["N"], ET=config.get("ET", 5000), alpha=config["alpha"], verbose=True)
test = lambda p: boundary_test(p, boundary, slack=0)
output = run_experiment(test, probs, config)
output["ET"] = ET
output["time"] = time.time() - t_start
return output
def run_heuristic_boundary_experiment(probs, config):
t_start = time.time()
prior = d.Beta(config["prior"]["params"][0],
config["prior"]["params"][1])
T = config.get("ET", 2000)
beta = config.get("beta", 0.3)
print("Finding boundary...")
boundary = binomial_heuristic(config["s"],
prior, config["N"], T=T, alpha=config["alpha"], beta=beta)
test = lambda p: boundary_test(p, boundary, slack=0)
output = run_experiment(test, probs, config)
output["ET"] = T
output["time"] = time.time() - t_start
output["beta"] = beta
return output
n_s = 1.5
n_prior = d.Gaussian(0, 1)
n_data = d.Gaussian(0, 3)
normal_boundaries, normal_ET = solve_normal(n_s,
n_prior,
n_data,
N,
ET=307,
lb=300,
ub=310,
verbose=True)
n_heuristic = normal_heuristic(n_s, n_prior, n_data, N, normal_ET, beta=0.4)
bin_s = 0.4
bin_prior = d.Beta(3, 10)
bin_boundaries, bin_ET = solve_binomial(bin_s,
bin_prior,
N,
ET=200,
verbose=True,
adaptive=False)
bin_heuristic = binomial_heuristic(bin_s, bin_prior, N, bin_ET, beta=0.3)
f, axes = plt.subplots(1, 2, figsize=(7, 3))
ax = axes[0]
gamma = (n_prior.tau / n_data.tau)**2
ax.plot([bdy.n for bdy in normal_boundaries],
beta_min, beta_max, beta0 = 0, max_param, 1.0
# Drawing range of Beta distribution
x_min, x_max = 0, 1
y_min, y_max = 0, 12
x_step = 0.01
x = np.arange(x_min, x_max, x_step)
fig, ax = plt.subplots(figsize=(6,6))
plt.subplots_adjust(left=0.1, bottom=0.25)
ax.grid()
ax.set_xlim(x_min, x_max)
ax.set_ylim(y_min, y_max)
dist = ds.Beta(alpha0, beta0)
line2dObj, = plt.plot(x, dist.pdf(x), lw=2, color='red')
# Slider, Button, RadioButton のfaceの色
axcolor = 'lightgoldenrodyellow'
#
# Slider
# Left Btm Wid Hight
axAlpha = plt.axes([0.1, 0.15, 0.8, 0.03], facecolor=axcolor)
axBeta = plt.axes([0.1, 0.10, 0.8, 0.03], facecolor=axcolor)
#sAlpha = Slider(axAlpha, 'Alpha', alpha_min, alpha_max, valinit=alpha0)
#sBeta = Slider(axBeta , 'Beta', beta_min, beta_max, valinit=beta0)
sAlpha = Slider(axAlpha, r'$\alpha$', alpha_min, alpha_max, valinit=alpha0)
sBeta = Slider(axBeta , r'$\beta$', beta_min, beta_max, valinit=beta0)
ax.fill_between(sdf[x],
sdf[y] - sdf[err],
sdf[y] + sdf[err],
alpha=0.2)
if legend:
ax.legend()
min_AB = 200
alpha = 0.05
batting = load_baseball_data(min_AB)
probs = batting["H/AB"]
ahat, bhat = util.beta_mom(probs)
betahat = d.Beta(ahat, bhat)
df = pd.read_csv("../data/experiment_summary.csv")
df >>= mutate(efficiency=X.n_discoveries / (X.avg_samples * X.n_tests))
df >>= mutate(efficacy=1 / X.efficiency)
df["fdp_err"] = 2 * np.sqrt(df.fdp * (1 - df.fdp) / df.n_discoveries)
df["power_err"] = 2 * np.sqrt(df.power * (1 - df.power) / df.true_effects)
plot_df = df >> mask(X.alpha == alpha, X.min_AB == min_AB, X.s < 0.33,
X.n_tests == 572100, X.n_discoveries > 0)
plot_df = plot_df.sort_values("s")
f, ax = plt.subplots(2, 2, figsize=(10, 6.25))
hist_ax = ax[0][0]