def plot_lb_w_time(num_sims: int = 20) -> None:
"""Plot lower bounds with sampling times.
Args:
num_sims: Number of simulations.
Returns:
Save plot.
"""
sim_net = sampcomp.NetworkHypothesisTesting()
sim_net.sigma_in_sq = 0
sim_net.sigma_te_sq = 0
num_genes = 20
prob_conn = 0.5
spec_rad = 0.8
time_list = [2, 4, 5, 8, 10, 20, 25, 40]
lower_bounds = []
for samp_time in time_list:
num_cond = 200 / samp_time
sim_net.samp_times = samp_time
lower_bounds.append(
sim_net.sim_er_genie_bhatta_lb(num_genes, prob_conn, spec_rad,
num_sims, num_cond)[1])
lower_bounds = np.asarray(lower_bounds)
plt.figure()
plt.errorbar(time_list, lower_bounds[:, 0], yerr=lower_bounds[:, 1])
plt.xlabel("number of sampling times")
plt.ylabel("lower bound on average error probability")
plt.savefig(
"/Users/veggente/Documents/workspace/python/sampcomp/pe_v_time_n{}.eps"
.format(num_sims))
def plot_lb_w_network_size(num_sims: int = 20, sigma_te_sq: float = 0) -> None:
"""Plot lower bounds with network size.
Args:
num_sims: Number of simulations.
sigma_te_sq: Technical variation.
Returns:
Save plot.
"""
sim_net = sampcomp.NetworkHypothesisTesting()
sim_net.sigma_in_sq = 0
sim_net.sigma_te_sq = sigma_te_sq
num_genes_list = [20, 40, 60, 80, 100, 120, 140, 160, 180, 200]
prob_conn = [10 / num_genes for num_genes in num_genes_list]
spec_rad = 0.8
samp_time = 10
num_cond = 20
lower_bounds = []
for counter, num_genes in enumerate(num_genes_list):
sim_net.samp_times = samp_time
lower_bounds.append(
sim_net.sim_er_genie_bhatta_lb(num_genes, prob_conn[counter],
spec_rad, num_sims, num_cond)[1])
lower_bounds = np.asarray(lower_bounds)
plt.figure()
plt.errorbar(num_genes_list, lower_bounds[:, 0], yerr=lower_bounds[:, 1])
plt.xlabel("network size")
plt.ylabel("lower bound on average error probability")
plt.savefig(
"/Users/veggente/Documents/workspace/python/sampcomp/pe_v_network_size{}_n{}_t{}.eps"
.format( # pylint: disable=line-too-long
num_genes_list[-1], num_sims, sim_net.sigma_te_sq))
def plot_bhatta_w_samp_rate(num_sims: int = 20) -> None:
"""Plot Bhattacharyya coefficient with sampling rate.
Args:
num_sims: Number of simulations.
Returns:
Save plot.
"""
sim_net = sampcomp.NetworkHypothesisTesting()
sim_net.sigma_in_sq = 0
sim_net.sigma_te_sq = 0
sim_net.one_shot = False
num_genes = 200
prob_conn = 0.05
spec_rad = 0.8
rate_list = [1, 2, 3, 4, 6]
bhatta_list = []
for rate in rate_list:
sim_net.samp_times = int(12 / rate)
bhatta_list.append(
sim_net.sim_er_bhatta(num_genes,
prob_conn,
spec_rad,
num_sims,
stationary=True,
skip=rate - 1)[0])
plt.figure()
plt.plot(rate_list, bhatta_list)
plt.xlabel("sampling rate")
plt.ylabel("Bhattacharyya coefficient")
plt.savefig(
"/Users/veggente/Documents/workspace/python/sampcomp/bhatta_v_rate_n{}.eps"
.format(num_sims))
def plot_bhatta_w_time(num_sims: int = 20) -> None:
"""Plot Bhattacharyya coefficient with sampling times.
Args:
num_sims: Number of simulations.
Returns:
Save plot.
"""
sim_net = sampcomp.NetworkHypothesisTesting()
sim_net.sigma_in_sq = 0
sim_net.sigma_te_sq = 0
sim_net.one_shot = False
num_genes = 200
prob_conn = 0.05
spec_rad = 0.8
time_list = [2, 4, 6, 8, 10]
bhatta_list = []
for samp_time in time_list:
sim_net.samp_times = samp_time
bhatta_list.append(
sim_net.sim_er_bhatta(num_genes,
prob_conn,
spec_rad,
num_sims,
stationary=True)[0])
plt.figure()
plt.semilogy(time_list, bhatta_list)
plt.xlabel("number of sampling times")
plt.ylabel("Bhattacharyya coefficient")
plt.savefig(
"/Users/veggente/Documents/workspace/python/sampcomp/bhatta_v_time_n{}.eps"
.format(num_sims))
def plot_lb_w_spec_rad(num_sims: int = 20, sigma_te_sq: float = 0) -> None:
"""Plot lower bounds with spectral radius.
Args:
num_sims: Number of simulations.
sigma_te_sq: Technical variation.
Returns:
Save plot.
"""
sim_net = sampcomp.NetworkHypothesisTesting()
sim_net.sigma_in_sq = 0
sim_net.sigma_te_sq = sigma_te_sq
num_genes = 200
prob_conn = 10 / num_genes
spec_rad_list = [0.1, 0.2, 0.4, 0.8]
samp_time = 10
num_cond = 200
lower_bounds = []
sim_net.samp_times = samp_time
for spec_rad in spec_rad_list:
lower_bounds.append(
sim_net.sim_er_genie_bhatta_lb(num_genes, prob_conn, spec_rad,
num_sims, num_cond)[1])
lower_bounds = np.asarray(lower_bounds)
plt.figure()
plt.errorbar(spec_rad_list, lower_bounds[:, 0], yerr=lower_bounds[:, 1])
plt.xlabel("spectral radius")
plt.ylabel("lower bound on average error probability")
plt.savefig(
"/Users/veggente/Documents/workspace/python/sampcomp/pe_v_spec_rad_n{}_c{}_t{}.eps"
.format(num_sims, num_cond, sim_net.sigma_te_sq))
def bhatta_vs_skipped_step_size(sims: int, samp_times: int, total_time: float,
base_eta: float):
"""Computes Bhattacharyya coefficient with varying skipped step sizes.
Follows the setting in [Bento, Ibrahimi, Montanari 2010].
Args:
sims: Number of simulations.
samp_times: Number of actual samples for each skip value.
total_time: Total time interval.
base_eta: Base step size.
Returns:
Prints Bhattacharyya coefficients.
"""
sim_net = sampcomp.NetworkHypothesisTesting()
sim_net.sigma_in_sq = 0
sim_net.sigma_te_sq = 0
sim_net.one_shot = False
sim_net.samp_times = samp_times
skips = list(range(10))
bhatta_fixed_duration = []
for this_skip in skips:
bhatta = sim_net.sim_er_bhatta(
16,
1 / 4,
None,
sims,
stationary=True,
step_size=base_eta,
memory=True,
skip=this_skip,
)[0]
bhatta_fixed_duration.append(bhatta**(total_time / base_eta /
samp_times / (this_skip + 1)))
output_prefix = "/Users/veggente/Data/research/flowering/soybean-rna-seq-data/sampcomp/bhatta_v_skipped_step_unscaled_t{}_n{}_s{}_e{}".format( # pylint: disable=line-too-long
total_time,
sims,
samp_times,
base_eta,
)
with open(output_prefix + ".data", "w") as f:
for bhatta in bhatta_fixed_duration:
f.write(str(bhatta) + "\n")
plt.figure()
plt.plot(skips, bhatta_fixed_duration, "-o")
plt.xlabel("skips")
plt.ylabel("Bhattacharyya coefficient")
plt.savefig(output_prefix + ".eps")
def stb_lb(spec_rad: float,
samp_times: int,
sigma_te_sq: float,
and_upper: bool = False,
**kwargs) -> Union[float, Tuple[float, float]]:
"""Calculates lower bound on average error rate.
Average is on the cross edges.
Args:
spec_rad: Spectral radius.
samp_times: Number of sampling times.
sigma_te_sq: Technical variance.
and_upper: Also compute upper bounds.
**stationary: bool
Start from stationary distribution.
Returns:
Bhattacharyya lower bound (or lower and upper bounds) on average error rate.
"""
num_genes = 200
sigma_en_sq = 1
sigma_in_sq = 0
num_rep = 1
prob_conn = 0.05
num_sims = 10
num_cond = 2000 / samp_times
network_ht = sampcomp.NetworkHypothesisTesting()
network_ht.sigma_en_sq = sigma_en_sq
network_ht.sigma_in_sq = sigma_in_sq
network_ht.sigma_te_sq = sigma_te_sq
network_ht.samp_times = samp_times
network_ht.num_rep = num_rep
network_ht.one_shot = False
sim_lower_bound = network_ht.sim_er_genie_bhatta_lb(
num_genes,
prob_conn,
spec_rad,
num_sims,
num_cond,
bayes=False,
and_upper=and_upper,
**filter_kwargs(kwargs, network_ht.sim_er_genie_bhatta_lb))
if and_upper:
return sim_lower_bound[1][0], sim_lower_bound[3][0]
return sim_lower_bound[1][0]
def bhatta_vs_step_size(sims: int):
"""Computes Bhattacharyya coefficient with varying step sizes.
Roughly follows the setting in [Bento, Ibrahimi, Montanari 2010].
Args:
sims: Number of simulations.
Returns:
Prints Bhattacharyya coefficients.
"""
sim_net = sampcomp.NetworkHypothesisTesting()
sim_net.sigma_in_sq = 0
sim_net.sigma_te_sq = 0
sim_net.one_shot = False
sim_net.samp_times = 10
eta_list = np.linspace(0.02, 0.2, 10)
bhatta_list = []
for eta in eta_list:
bhatta_list.append(
sim_net.sim_er_bhatta(16,
1 / 4,
None,
sims,
stationary=True,
step_size=eta,
memory=True))
total_time = 1
bhatta_fixed_duration = [
bhatta**(total_time / eta_list[0] / 10) for bhatta, _ in bhatta_list
]
with open(
"/Users/veggente/Data/workspace/python/sampcomp/bhatta_v_step_unscaled.data",
"w",
) as f:
for bhatta in bhatta_fixed_duration:
f.write(str(bhatta))
plt.figure()
plt.plot(eta_list, bhatta_fixed_duration, "-o")
plt.xlabel(r"$\eta$")
plt.ylabel("Bhattacharyya coefficient")
plt.savefig(
"/Users/veggente/Data/workspace/python/sampcomp/bhatta_v_step_unscaled.eps"
)
def plot_sub_samp(seed: int = 0):
"""Plot subsampling results."""
np.random.seed(seed)
er_graph, weight = sampcomp.erdos_renyi(10, 0.2, 0.8)
network_ht = sampcomp.NetworkHypothesisTesting()
adj_mat_pair = list(
network_ht.genie_hypotheses(er_graph, (0, 1), weight, 0.8))
bhatta_list = []
for skip in range(30):
cov_mat = [
sampcomp.gen_cov_mat(adj_mat, 0, 1, 10, 1, False, 1, skip)
for adj_mat in adj_mat_pair
]
bhatta_list.append(sampcomp.bhatta_coeff(*cov_mat))
print(adj_mat_pair)
plt.figure()
plt.plot(np.arange(1, 31), bhatta_list)
plt.xlabel("sampling interval T")
plt.ylabel("Bhattacharyya coefficient")
plt.savefig("subsampling_s{}.eps".format(seed))
