Skip to content

GitLab

I
IBD-Analysis
Commit 4f8130d5 authored by Harald RINGBAUER's avatar Harald RINGBAUER
Browse files
Options

Upload new file

parent f7e3a103
Hide whitespace changes
Inline Side-by-side
Showing with 827 additions and 0 deletions
IBD-Simulations/multi_runs.py 0 → 100644
View file @ 4f8130d5
'''
Created on Mar 19, 2015
Package similar to main, but with the purpose to do multiple runs for statistical analysis and save the data.
Parameters are taken from grid.py and mle_estim_error.py!
@author: Harald Ringbauer
'''
from grid import factory_Grid
from analysis import Analysis, torus_distance
from random import shuffle
from itertools import combinations
from bisect import bisect_right
from math import pi
from scipy.special import kv as kv # Import Bessel functions of second kind
import cPickle as pickle
import numpy as np
import matplotlib.pyplot as plt
t = 200 # Generation time for a single run #t=200
nr_runs = 20 # How many runs
sample_sizes = (100, 270, 440, 625)
distances = [[2, 10], [10, 20], [20, 30], [30, 40], [40, 50], [50, 60]] # Distances to use for binning
# distances = [[1, 5], [5, 10], [10, 15], [15, 20], [20, 25], [25, 30]] # Distances to use for binning
intervals = [[4, 5], [5, 6.5], [6.5, 8], [8, 12]] # Bins for the block length binning
def single_run():
''' Do a single run, parameters are saved in grid'''
grid = factory_Grid() # Set grid
grid.reset_grid() # Delete everything
grid.set_samples()
grid.update_t(t) # Do the actual run
data = Analysis(grid) # Do Data-Analysis
data.fit_expdecay(show=False)
sigma = data.sigma_estimate
block_nr = len(grid.IBD_blocks)
return(sigma, block_nr)
def analysis_run():
grid = factory_Grid()
parameters = (grid.sigma, grid.gridsize, grid.sample_steps, grid.dispmode)
results = np.zeros((nr_runs, 2)) # Container for the data
'''Runs the statistical analysis'''
for i in range(0, nr_runs):
print("Doing run: %.1f" % i)
results[i, :] = single_run() # Do the run and save the results
print("RUN COMPLETE!!")
pickle.dump((results, parameters), open("Data1/stats_demes.p", "wb")) # Pickle the data
# save_name=raw_input("Save to what filename?")
# pickle.dump((results, parameters), open(save_name, "wb")) # Pickle the data
print("SAVED")
def run_var_samp(save_name):
'''Runs simulations for various sample sizes and saves estimates and parameters.'''
grid = factory_Grid()
results = np.zeros((len(sample_sizes), nr_runs, 2)) # Container for the data
sample_steps = grid.sample_steps
'''Actual runs:'''
row = 0
for k in sample_sizes:
position_list = [(i + sample_steps / 2, j + sample_steps / 2, 0) for i in range(0, grid.gridsize, sample_steps) for j in range(0, grid.gridsize, sample_steps)]
# position_list = [(i + sample_steps / 2, j + sample_steps / 2, 0) for i in range(0, sample_steps * k, sample_steps) for j in range(0, sample_steps * k, sample_steps)]
for i in range(0, nr_runs):
print("Doing run: %.1f for %.0f samples" % (i, k))
grid.reset_grid() # Delete everything
shuffle(position_list) # Randomize position List
print(position_list)
grid.set_chromosome(position_list[:k]) # Set the samples
grid.update_t(t) # Do the actual run
if grid.dispmode == "demes":
grid.update_IBD_blocks_demes(5) # Update position for deme analysis!!
data = Analysis(grid) # Do Data-Analysis
data.fit_expdecay(show=False)
sigma = data.sigma_estimate
block_nr = len(grid.IBD_blocks)
results[row, i, :] = (sigma, block_nr)
row += 1 # Go one down in the results_row
print("RUN COMPLETE!!")
parameters = (grid.sigma, grid.gridsize, sample_sizes, grid.dispmode)
pickle.dump((results, parameters), open(save_name, "wb")) # Pickle the data
print("SAVED")
def empirical_IBD_list(save_name):
'''Generate empirical IBD-list. Nr. of run times'''
results = [] # Container for the data
'''Actual runs:'''
for i in range(nr_runs):
print("Doing run: %i" % i)
grid = factory_Grid(growing=1) # No growing grid
grid.reset_grid()
grid.set_samples()
grid.update_t(t) # Do the actual run
# if grid.dispmode == "demes":
# grid.update_IBD_blocks_demes(5) # Update position for deme analysis!!
pair_dist, pair_IBD, pair_nr = grid.give_lin_IBD(bin_pairs=True) # Get the binned IBD-lists
results.append([pair_dist, pair_IBD, pair_nr]) # Save the empirical data
parameters = (grid.sigma, grid.gridsize, grid.sample_steps, grid.dispmode)
pickle.dump((results, parameters), open(save_name, "wb")) # Pickle the data
print("SAVED")
def get_normalization_factor(dist_bins, grid_size, sample_steps):
'''Calculates Normalization factor for binned distances and starting grid'''
position_list = [(i + sample_steps / 2, j + sample_steps / 2, 0) for i in range(0, grid_size, sample_steps) for j in range(0, grid_size, sample_steps)]
k = len(position_list)
print(k * (k - 1) / 2)
distance_bins = np.zeros(len(dist_bins) - 1) # Create bins for every element in List; len(bins)=len(counts)+1
dist_bins[-1] += 0.000001 # Hack to make sure that the distance exactly matching the max are counted
# Calculate Distance for every possible pair to get proper normalization factor:
for (x, y) in combinations(np.arange(len(position_list)), r=2):
dist = torus_distance(position_list[x][0], position_list[x][1], position_list[y][0], position_list[y][1], grid_size)
j = bisect_right(dist_bins, dist)
if j < len(dist_bins) and j > 0: # So it actually falls into somewhere
distance_bins[j - 1] += 1
return distance_bins
def get_normalization_lindata(dist_bins, pair_dist, pair_nr):
'''Gets the Nr. of pairs in each distance bin from linearized IBD-data'''
bl_nr = np.zeros(len(dist_bins)) # Creates zero array
for i in range(len(dist_bins)):
dist = dist_bins[i]
bl_nr[i] += np.sum([pair_nr[j] for j in range(len(pair_dist))
if dist[0] <= pair_dist[j] < dist[1]])
return bl_nr
def into_bins(pair_dist, pair_IBD, intervals, distances):
'''Return Matrix of blocks Nr in each interval and distance bin.'''
res = np.zeros((len(intervals), len(distances))).astype(np.int) # Create empty Int-Array.
for i in range(len(intervals)):
for j in range(len(distances)):
intv = intervals[i]
dist = distances[j]
for d in range(len(pair_dist)): # Iterate over every pair
if dist[0] <= pair_dist[d] < dist[1]:
res[i, j] += np.sum([1 for block in pair_IBD[d] if (intv[0] <= block <= intv[1])])
return np.array(res) # Return numpy array
def bessel_longer_l0(sigma, r, l0, b, D):
'''The formula for block sharing longer than l0. Measured in cM
If r vector return vector. b is growth rate parameter'''
l0 = l0 / 100.0 # Change to Morgan
G = 1.5 # Chromosome-Length
C = G / (D * pi * sigma ** 2) # First calculate the constant; D_e=1
y = C * 2 ** ((-5 - 3 * b) / 2.0) * (r / (np.sqrt(l0) * sigma)) ** (1 + b) * kv(1 + b, np.sqrt(2.0 * l0) * r / sigma)
# y = C * r / (sigma * np.sqrt(2 * l0)) * kv(1, np.sqrt(2.0 * l0) * r / sigma)
return y
def bessel_l(sigma, r, l, b, D):
'''Plots the exact Besseldecay for a block of length l. If r vector return vector'''
l = l / 100.0 # Change to Morgan
G = 1.5 # Chromosome-Length
C = G / (D * pi * sigma ** 2) # First calculate the constant; D_e=1
y = C * 2 ** (-3 - 3 * b / 2.0) * (r / (np.sqrt(l) * sigma)) ** (2 + b) * kv(2 + b, np.sqrt(2.0 * l) * r / sigma)
# y = C * r / (sigma * np.sqrt(2 * l0)) * kv(1, np.sqrt(2.0 * l0) * r / sigma)
return (y / 100.0) # Return density in cM
def get_theory_sharing(intervals, distances, sigma, b, D):
'''Gives back the theory sharing for the given Distance- and
block length intervals'''
res = np.zeros((len(intervals), len(distances))) # Array for results
for i in range(len(intervals)):
l = intervals[i]
int_len = l[1] - l[0]
for j in range(len(distances)):
d1 = distances[j]
dists = np.linspace(d1[0], d1[1], 20) # 20 Intervals for calculation
shr_bin = bessel_longer_l0(sigma, dists, l[0], b, D) - bessel_longer_l0(sigma, dists, l[1], b, D) # Get the sharing per length bin
res[i, j] = np.mean(shr_bin) / int_len # Mean sharing per cM averaged over dist bins
return res
def analyze_emp_IBD_list(save_name, show=True, b=0, D=1):
'''Plots summary of the empirical-IBD-list'''
(results, parameters) = pickle.load(open(save_name, "rb")) # Data2/file.p
print(parameters)
print(len(results))
sigma, _, _, _ = parameters
dist_means = np.array([np.mean(i) for i in distances]) # Mean distances
bl_nr = np.zeros((len(results), len(intervals), len(distances))).astype(np.int) # Container for block matrices
for i in range(len(results)):
pair_dist, pair_IBD, pair_nr = results[i] # Unpack the result arrays.
# bl_list = results[i]
# Generate Block-Info
# block_info = [[b[1], torus_distance(b[2][0], b[2][1], b[3][0], b[3][1], grid_size)] for b in bl_list]
bl_nr[i, :, :] = into_bins(pair_dist, pair_IBD, intervals, distances) # Get the empirical data matrix
# print(pair_dist)
# print(pair_nr)
nr_pairs_bins = get_normalization_lindata(distances, pair_dist, pair_nr) # Get Nr. factor: Pairs of inds
int_len = np.array([i[1] - i[0] for i in intervals]) # Length of the intervals - to normalize for that
mean_shr = np.mean(bl_nr, axis=0)
sts = np.std(bl_nr, axis=0)
emp_shr = mean_shr / nr_pairs_bins
emp_shr = emp_shr / int_len[:, None]
emp_sts = sts / nr_pairs_bins
emp_sts = emp_sts / int_len[:, None]
# print(emp_shr)
thr_shr = get_theory_sharing(intervals, distances, sigma, b, D) # Get predicted sharing
# print(thr_shr)
'''Now do the plotting'''
if show == True:
f, axarr = plt.subplots(2, 2, sharex=True, sharey=True) # Create sub-plots
for i in range(4): # Loop through interval list
curr_plot = axarr[i / 2, i % 2] # Set current plot
curr_plot.set_yscale('log') # Set Log-Scale
interval = intervals[i]
# int_len = interval[1] - interval[0] # Calculate the length of an interval
curr_plot.set_ylim([1.0 / 10 ** 6, 1.0 / 10])
curr_plot.set_xlim([0, 70])
l2, = curr_plot.semilogy(dist_means, thr_shr[i, :], 'r-.', linewidth=2) # Plot of exact fit
l1 = curr_plot.errorbar(dist_means, emp_shr[i, :], yerr=emp_sts[i, :], fmt='bo')
# curr_plot.set_ylim([min(y) / 3, max(y) * 3])
curr_plot.set_title("Interval: " + str(interval) + " cM")
# curr_plot.annotate("Blocks: %.0f" % self.total_bl_nr, xy=(0.4, 0.7), xycoords='axes fraction', fontsize=16)
f.text(0.5, 0.02, r'Distance [$\sigma$]', ha='center', va='center', fontsize=20)
f.text(0.025, 0.5, 'IBD-blocks per pair and cM', ha='center', va='center', rotation='vertical', fontsize=20)
f.legend((l1, l2), ('Simulated IBD-sharing', 'Theory'), loc=(0.7, 0.36))
plt.tight_layout()
plt.show()
dist_means = dist_means / sigma # Normalize The distance to Dispersal units
return((dist_means, thr_shr, emp_shr, emp_sts))
def analyze_mult_emp_lists(save_names):
emp_shr = []
emp_sts = []
for name in save_names: # Load the data
dist_means, thr_shr, shr, sts = analyze_emp_IBD_list(name, show=False)
emp_shr.append(shr)
emp_sts.append(sts)
# Do the actual plot: (Multiple Windows)
# f, axarr = plt.subplots(2, 2, sharex=True, sharey=True) # Create sub-plots
# for i in range(4): # Loop through interval list
# curr_plot = axarr[i / 2, i % 2] # Set current plot
# curr_plot.set_yscale('log') # Set Log-Scale
# interval = intervals[i]
# curr_plot.set_ylim([1.0 / 10 ** 6, 1.0 / 10])
# curr_plot.set_xlim([0, 30])
#
# l0, = curr_plot.semilogy(dist_means, thr_shr[i, :], 'r-', linewidth=2) # Plot of exact fit
# l1 = curr_plot.errorbar(dist_means - 1, emp_shr[0][i, :], yerr=emp_sts[0][i, :], fmt='yo', linewidth=2)
# l2 = curr_plot.errorbar(dist_means - 0.5, emp_shr[1][i, :], yerr=emp_sts[1][i, :], fmt='go', linewidth=2)
# l3 = curr_plot.errorbar(dist_means, emp_shr[2][i, :], yerr=emp_sts[2][i, :], fmt='ko', linewidth=2)
# l4 = curr_plot.errorbar(dist_means + 0.5, emp_shr[3][i, :], yerr=emp_sts[3][i, :], fmt='mo', linewidth=2)
# l5 = curr_plot.errorbar(dist_means + 1, emp_shr[4][i, :], yerr=emp_sts[4][i, :], fmt="co", linewidth=2)
# # curr_plot.set_ylim([min(y) / 3, max(y) * 3])
# curr_plot.set_title("Interval: " + str(interval) + " cM", fontsize=20)
# curr_plot.tick_params(axis='x', labelsize=20)
# curr_plot.tick_params(axis='y', labelsize=20)
#
# # curr_plot.annotate("Blocks: %.0f" % self.total_bl_nr, xy=(0.4, 0.7), xycoords='axes fraction', fontsize=16)
# f.text(0.5, 0.02, r'Distance [$\sigma$]', ha='center', va='center', fontsize=28)
# f.text(0.025, 0.5, 'IBD-blocks per pair and cM', ha='center', va='center', rotation='vertical', fontsize=28)
# f.legend((l0, l1, l2, l3, l4, l5), ('Theory', 'DiscSim', 'Laplace', 'Normal', 'Uniform', 'Demes'), loc=(0.83, 0.23))
# plt.tight_layout()
# plt.show()
plt.figure()
plt.yscale('log') # Set Log-Scale)
plt.ylim([1.0 / 10 ** 6, 1.0 / 10])
plt.xlim([0, 30])
styles = ['r-', 'r--', 'r-.', 'r:']
for i in range(len(intervals)):
interval = intervals[i]
plt.semilogy(dist_means, thr_shr[i, :], styles[i], linewidth=3, label=str(interval) + " cM") # Plot of exact fit
l1 = plt.errorbar(dist_means - 1, emp_shr[0][i, :], yerr=emp_sts[0][i, :], fmt='yo', linewidth=3)
l2 = plt.errorbar(dist_means - 0.5, emp_shr[1][i, :], yerr=emp_sts[1][i, :], fmt='go', linewidth=3)
l3 = plt.errorbar(dist_means, emp_shr[2][i, :], yerr=emp_sts[2][i, :], fmt='ko', linewidth=3)
l4 = plt.errorbar(dist_means + 0.5, emp_shr[3][i, :], yerr=emp_sts[3][i, :], fmt='mo', linewidth=3)
l5 = plt.errorbar(dist_means + 1, emp_shr[4][i, :], yerr=emp_sts[4][i, :], fmt="co", linewidth=3)
# curr_plot.set_ylim([min(y) / 3, max(y) * 3])
plt.tick_params(axis='x', labelsize=20)
plt.tick_params(axis='y', labelsize=20)
# curr_plot.annotate("Blocks: %.0f" % self.total_bl_nr, xy=(0.4, 0.7), xycoords='axes fraction', fontsize=16)
plt.xlabel(r'Distance [$\sigma$]', fontsize=28)
plt.ylabel('IBD-blocks per pair and cM', fontsize=28)
f1 = plt.legend((l1, l2, l3, l4, l5), ('DiscSim', 'Laplace', 'Normal', 'Uniform', 'Demes'), loc=(0.75, 0.67))
# Create a legend for the first line.
plt.gca().add_artist(f1) # Add the legend manually to the current Axes.
plt.legend(loc=(0.1, 0.1))
plt.tight_layout()
plt.show()
def analyze_var_density():
'''Analyze varying density data_sets'''
b = [1, -1, 0] # Growth rate parameters
D = [200, 1, 10]
save_names = ["growing20.p", "declining20.p", "const20.p"]
# distances = [[2, 10], [10, 20], [20, 30], [30, 40], [40, 50], [50, 100]] # Distances to use for binning
# dists = np.linspace(1, 80, 200) # x-Values for Theory
emp_shr = []
emp_sts = []
thr_shrs = []
for i in range(len(save_names)): # Load the data
name = save_names[i]
dist_means, thr_shr, shr, sts = analyze_emp_IBD_list(name, show=False, b=b[i], D=D[i])
thr_shrs.append(thr_shr)
emp_shr.append(shr)
emp_sts.append(sts)
# Do the actual plot:
f, axarr = plt.subplots(2, 2, sharex=True, sharey=True) # Create sub-plots
for i in range(4): # Loop through interval list
curr_plot = axarr[i / 2, i % 2] # Set current plot
curr_plot.set_yscale('log') # Set Log-Scale
interval = intervals[i]
curr_plot.set_ylim([1.0 / 10 ** 7, 1.0 / 10])
curr_plot.set_xlim([0, 30])
l0, = curr_plot.semilogy(dist_means, thr_shrs[0][i, :], 'r-', linewidth=2) # Plot of exact fit
curr_plot.errorbar(dist_means, emp_shr[0][i, :], yerr=emp_sts[0][i, :], fmt='ro')
# l2 = curr_plot.semilogy(dists, bessel_l(1.0, dists, np.sqrt(interval[0]*interval[1]), 1, 200), 'b-')
# l2 = curr_plot.errorbar(dist_means - 0.3, emp_shr[1][i, :], yerr=emp_sts[1][i, :], fmt='go')
l1, = curr_plot.semilogy(dist_means, thr_shrs[1][i, :], 'b-', linewidth=2) # Plot of exact fit
curr_plot.errorbar(dist_means, emp_shr[1][i, :], yerr=emp_sts[1][i, :], fmt='bo')
l2, = curr_plot.semilogy(dist_means, thr_shrs[2][i, :], 'm-', linewidth=2) # Plot of exact fit
curr_plot.errorbar(dist_means, emp_shr[2][i, :], yerr=emp_sts[2][i, :], fmt='mo')
curr_plot.set_title("Interval: " + str(interval) + " cM", fontsize=20)
curr_plot.tick_params(axis='x', labelsize=20)
curr_plot.tick_params(axis='y', labelsize=20)
# curr_plot.annotate("Blocks: %.0f" % self.total_bl_nr, xy=(0.4, 0.7), xycoords='axes fraction', fontsize=16)
f.text(0.5, 0.02, r'Distance [$\sigma$]', ha='center', va='center', fontsize=28)
f.text(0.025, 0.5, 'IBD-blocks per pair and cM', ha='center', va='center', rotation='vertical', fontsize=28)
f.legend((l0, l2, l1), ('Growing', 'Constant', 'Declining'), loc=(0.8, 0.35))
plt.tight_layout()
plt.show()
def analyze_var_samp():
'''Analyze saved results in Data2 produced by run_var_samp'''
(results_l, parameters_l) = pickle.load(open("Data2/laplace1.p", "rb"))
(results_n, parameters_n) = pickle.load(open("Data2/normal1.p", "rb")) # @UnusedVariable
(results_u, parameters_u) = pickle.load(open("Data2/uniform1.p", "rb")) # @UnusedVariable
(results_d, parameters_d) = pickle.load(open("Data2/demes3.p", "rb")) # @UnusedVariable
(results_c, parameters_c) = pickle.load(open("Data2/discsim2.p", "rb")) # @UnusedVariable
(results_l1, _) = pickle.load(open("Data2/laplace.p", "rb"))
(results_n1, _) = pickle.load(open("Data2/normal.p", "rb")) # @UnusedVariable
(results_u1, _) = pickle.load(open("Data2/uniform.p", "rb")) # @UnusedVariable
(results_d1, _) = pickle.load(open("Data2/demes.p", "rb")) # @UnusedVariable
(results_c1, _) = pickle.load(open("Data2/DISCSIM.p", "rb")) # @UnusedVariable
sample_sizes = parameters_l[2]
sigma_estimates_l = np.concatenate((results_l[:, :, 0], results_l1[:, :, 0]), axis=1)
sigma_estimates_n = np.concatenate((results_n[:, :, 0], results_n1[:, :, 0]), axis=1)
sigma_estimates_u = np.concatenate((results_u[:, :, 0], results_u1[:, :, 0]), axis=1)
sigma_estimates_d = results_d[:, :, 0]
sigma_estimates_c = np.concatenate((results_c[:, :, 0], results_c1[:, :, 0]), axis=1)
mean_sigma_est_l = np.mean(sigma_estimates_l, 1)
mean_sigma_est_n = np.mean(sigma_estimates_n, 1)
mean_sigma_est_u = np.mean(sigma_estimates_u, 1)
mean_sigma_est_d = np.mean(sigma_estimates_d, 1)
mean_sigma_est_c = np.mean(sigma_estimates_c, 1)
print("Sample size: %.1f" % np.size(sigma_estimates_u, 1))
std_sigma_est_l = np.std(sigma_estimates_l, 1)
std_sigma_est_n = np.std(sigma_estimates_n, 1)
std_sigma_est_u = np.std(sigma_estimates_u, 1)
std_sigma_est_d = np.std(sigma_estimates_d, 1)
std_sigma_est_c = np.std(sigma_estimates_c, 1)
plt.figure()
plt.errorbar(np.array(sample_sizes) - 10, mean_sigma_est_d, yerr=std_sigma_est_d, fmt='ko', label="Demes", linewidth=2)
plt.errorbar(np.array(sample_sizes) - 5, mean_sigma_est_u, yerr=std_sigma_est_u, fmt='bo', label="Uniform", linewidth=2)
plt.errorbar(sample_sizes, mean_sigma_est_l, yerr=std_sigma_est_l, fmt='mo', label="Laplace", linewidth=2)
plt.errorbar(np.array(sample_sizes) + 5, mean_sigma_est_n, yerr=std_sigma_est_n, fmt='ro', label="Normal", linewidth=2)
plt.errorbar(np.array(sample_sizes) + 10, mean_sigma_est_c, yerr=std_sigma_est_c, fmt='yo', label="DiscSim", linewidth=2)
plt.xlabel("Nr of samples", fontsize=20)
plt.ylabel(r"$\mathbf{\bar{\sigma}}$", fontsize=30)
plt.tick_params(axis='x', labelsize=15)
plt.tick_params(axis='y', labelsize=15)
plt.axhline(2, color='g', label="True Value", linewidth=2)
plt.legend()
plt.show()
def analyze_stats():
load_u = pickle.load(open("Data1/stats_uniform.p", "rb"))[0][:, 0]
load_n = pickle.load(open("Data1/stats_normal.p", "rb"))[0][:, 0]
load_l = pickle.load(open("Data1/stats_laplace.p", "rb"))[0][:, 0]
load_d = pickle.load(open("Data1/stats_demes.p", "rb"))[0][:, 0]
load_c = pickle.load(open("./Data1/disc_cstats.p", "rb"))[0][:, 0] / 1000.0
val_u = 1.987
val_n = 2.02
val_l = 2.02
val_d = 2
val_c = 2
# plt.figure()
# plt.hist(load_c, bins=20, alpha=0.5)
# plt.axvline(2, color='r', linestyle='dashed', linewidth=3)
# plt.xlabel("Estimation")
# plt.ylabel("Number")
# plt.show()
print(load_c.mean())
print("Rel. Bias %.6f" % (load_c.mean() / val_c - 1))
print("CV %.4f" % (load_c.std() / load_c.mean()))
# Plot Stuff
f, axarr = plt.subplots(3, 2, sharey=True, figsize=(5, 10))
# axarr.set_title("Estimated Dispersal rate")
axarr[0, 0].hist(load_u * 2 / val_u, bins=20, alpha=0.5)
axarr[0, 0].set_title('Uniform')
axarr[0, 0].axvline(2, color='r', linestyle='dashed', linewidth=3)
axarr[0, 0].axvline(np.mean(load_u), color='g', linestyle='dashed', linewidth=3)
axarr[0, 1].hist(load_n * 2 / val_n, bins=20, alpha=0.5)
axarr[0, 1].set_title('Normal')
axarr[0, 1].axvline(2, color='r', linestyle='dashed', linewidth=3)
axarr[0, 1].axvline(np.mean(load_n), color='g', linestyle='dashed', linewidth=3)
axarr[1, 0].hist(load_l * 2 / val_l, bins=20, alpha=0.5)
axarr[1, 0].set_title('Laplace')
axarr[1, 0].axvline(2, color='r', linestyle='dashed', linewidth=3)
axarr[1, 0].axvline(np.mean(load_l), color='g', linestyle='dashed', linewidth=3)
axarr[1, 1].hist(load_d * 2 / val_d, bins=20, alpha=0.5)
axarr[1, 1].set_title('Demes')
axarr[1, 1].axvline(2, color='r', linestyle='dashed', linewidth=3)
axarr[1, 1].axvline(np.mean(load_d), color='g', linestyle='dashed', linewidth=3)
axarr[1, 1].set_xlim(1.85, 2.15001)
axarr[2, 0].hist(load_c * 2 / val_c, bins=20, alpha=0.5)
axarr[2, 0].set_title('DISC-SIM')
axarr[2, 0].axvline(2, color='r', linestyle='dashed', linewidth=3)
axarr[2, 0].axvline(np.mean(load_c), color='g', linestyle='dashed', linewidth=3)
# for i in axarr[]:
# i.set_xlim
plt.delaxes(axarr[2, 1])
f.text(0.5, 0.04, 'Dispersal Estimates', ha='center')
f.text(0.04, 0.5, 'Counts', va='center', rotation='vertical')
# Fine-tune figure; hide x ticks for top plots and y ticks for right plots
# plt.setp([a.get_xticklabels() for a in axarr[0, :]], visible=False)
# plt.setp([a.get_yticklabels() for a in axarr[:, 1]], visible=False)
plt.show()
def run_var_samp1(file_name):
'''Runs MLE-estimates for various sample sizes and saves estimates and CIs.'''
grid = factory_Grid() # Create an empty Grid.
results = np.zeros((len(sample_sizes), nr_runs, 6)) # Container for the data
parameters = (grid.sigma, grid.gridsize, sample_sizes, grid.dispmode)
sample_steps, grid_size = grid.sample_steps, grid.gridsize
'''Actual runs:'''
row = 0
for k in sample_sizes:
position_list = [(i + sample_steps / 2, j + sample_steps / 2, 0) for i in range(0, grid_size, sample_steps) for j in range(0, grid_size, sample_steps)]
# position_list = [(i + sample_steps / 2, j + sample_steps / 2, 0) for i in range(0, sample_steps * k, sample_steps) for j in range(0, sample_steps * k, sample_steps)]
for i in range(0, nr_runs):
print("Doing run: %.1f for %.0f samples" % (i, k))
grid = factory_Grid()
grid.reset_grid() # Delete everything
shuffle(position_list) # Randomize position List
grid.set_samples(position_list[:k]) # Set the samples
grid.update_t(t) # Do the actual run
if grid.dispmode == "demes":
grid.update_IBD_blocks_demes(5) # Update position for deme analysis!!
# Do the maximum Likelihood estimation
mle_ana = grid.create_MLE_object(bin_pairs=True) # Create the MLE-object
mle_ana.create_mle_model("constant", grid.chrom_l, [1, 2])
mle_ana.mle_analysis_error()
d_mle, sigma_mle = mle_ana.estimates[0], mle_ana.estimates[1]
ci_s = mle_ana.ci_s
results[row, i, :] = (ci_s[1][0], ci_s[1][1], ci_s[0][0], ci_s[0][1], sigma_mle, d_mle)
row += 1 # Go one down in the results_row
print("RUN COMPLETE!!")
pickle.dump((results, parameters), open(file_name, "wb")) # Pickle the data
print("SAVED")
def analyze_var_samp1(file_name):
'''Analyze the results of the MLE-estimates for various sample size'''
(results, parameters) = pickle.load(open(file_name, "rb"))
print("Parameters used for Simulations: \n")
print(parameters)
result = 3 # Position of result to analyze
ci_lengths = results[result, :, 1] - results[result, :, 0]
sigmas_mles = results[result, :, 4]
d_mles = results[result, :, 5]
print("Mean CI length: %.4f" % np.mean(ci_lengths))
print("Mean sigma estimates: %.4f" % np.mean(sigmas_mles))
print("Standard Deviations sigma: %.4f" % np.std(sigmas_mles))
print("Mean D_e: %.4f" % (np.mean(d_mles)))
print("Standard Deviations D_e: %.4f" % np.std(d_mles))
k = len(results[:, 0, 0])
# Calculate Confidence Intervalls:
ci_lengths_s = [results[i, :, 1] - results[i, :, 0] for i in range(k)]
ci_lengths_d = [results[i, :, 3] - results[i, :, 2] for i in range(k)]
# Calculate Empirical Confidence Intervals:
ci_lengths_s1 = [np.percentile(results[i, :, 4], 97.5) - np.percentile(results[i, :, 4], 2.5)
for i in range(k)]
ci_lengths_d1 = [np.percentile(results[i, :, 5], 97.5) - np.percentile(results[i, :, 5], 2.5)
for i in range(k)]
print("\n Mean Length of est. Confidence Intervals (Sigma/D)")
print(np.mean(ci_lengths_s, axis=1))
print(np.mean(ci_lengths_d, axis=1))
print("\n Empirical Confidence Intervals:")
print(ci_lengths_s1)
print(ci_lengths_d1)
# Now do the correlation of estimates:
print("Correlation of Estimates")
print([np.corrcoef(results[i, :, 4], results[i, :, 5])[0, 1] for i in range(k)])
# Plot Sigma Estimate
plt.figure()
x_dist = np.linspace(0, 3, num=len(sigmas_mles))
ist = np.argsort(results[0, :, 4])
plt.plot(0 + x_dist, results[0, ist, 4], 'mo', label="MLE")
plt.vlines(0 + x_dist, results[0, ist, 0], results[0, ist, 1], 'r', label="Confidence Interval")
ist = np.argsort(results[1, :, 4])
plt.vlines(3.5 + x_dist, results[1, ist, 0], results[1, ist, 1], 'r')
plt.plot(3.5 + x_dist, results[1, ist, 4], 'mo')
ist = np.argsort(results[2, :, 4])
plt.vlines(7 + x_dist, results[2, ist, 0], results[2, ist, 1], 'r')
plt.plot(7 + x_dist, results[2, ist, 4], 'mo')
ist = np.argsort(results[3, :, 4])
plt.vlines(10.5 + x_dist, results[3, ist, 0], results[3, ist, 1], 'r')
# plt.scatter(11 + x_dist, results[3, :, 0], c='b')
plt.plot(10.5 + x_dist, results[3, ist, 4], 'mo')
plt.xlabel("Sample Size", fontsize=20)
plt.ylabel("Estimated " + r"$\mathbf{\sigma}$", fontsize=20)
plt.xticks([1.5, 5, 8.5, 12], ["100", "270", "440", "625"], fontsize=20)
plt.hlines(2, -0.5, 14, label="True " + r"$\mathbf{\sigma}$", color='k', linewidth=2)
plt.legend()
plt.ylim([0, 3.5])
plt.show()
# Plot density estimate
plt.figure()
x_dist = np.linspace(0, 3, num=len(sigmas_mles))
ist = np.argsort(results[0, :, 5])
plt.plot(0 + x_dist, results[0, ist, 5], 'mo', label="MLE")
plt.vlines(0 + x_dist, results[0, ist, 2], results[0, ist, 3], 'r', label="Confidence Interval")
ist = np.argsort(results[1, :, 5])
plt.vlines(3.5 + x_dist, results[1, ist, 2], results[1, ist, 3], 'r')
plt.plot(3.5 + x_dist, results[1, ist, 5], 'mo')
ist = np.argsort(results[2, :, 5])
plt.vlines(7 + x_dist, results[2, ist, 2], results[2, ist, 3], 'r')
plt.plot(7 + x_dist, results[2, ist, 5], 'mo')
ist = np.argsort(results[3, :, 5