def basic_spatial_grid(ctrl):
ax = graphs.spatial_grid(ctrl, unit='m', s=12)
ax.set_xlabel('Meters')
ax.set_ylabel('Meters')
ax.set_aspect('equal')
graphs.save('basic_spatial_grid')
graphs.show()
def zero_padded_crosscorr():
fontsize_tmp = graphs.FONTSIZE
graphs.change_fontsize(15)
p = pinit64_no_pulse_shape()
ax = graphs.crosscorr(p, savename='latex_figures/discrete_crosscorr');
graphs.show()
graphs.change_fontsize(fontsize_tmp)
return ax
def interd_cavg_vs_nodecount():
# 150 sample per point
graphs.GRAPH_OUTPUT_FORMAT = 'png'
graphs.change_fontsize(graphs.FONTSIZE + 2)
alldates = db.fetch_dates([20160825101515414, 20160825135628487]) # cavg vs nodecount
dates = [alldates[0], alldates[-1]]
ax = graphs.scatter_range(dates, ['nodecount', 'good_link_ratio'], color='k')
ax.set_xlabel("Number of nodes $M$")
ax.set_ylabel("$C$")
graphs.save('interd_cavg_vs_nodecount')
graphs.show()
def interd_compare_quiet():
graphs.GRAPH_OUTPUT_FORMAT = 'png'
graphs.change_fontsize(graphs.FONTSIZE + 2)
alldates = db.fetch_dates([20160822172521531, 20160822215536865]) # 960sims quiet compare
dates = [alldates[0], alldates[-1]]
labels = []
labels.append('Random')
labels.append('Clustering')
labels.append('Sensing')
ax = graphs.scatter_range(dates, ['max_dist_from_origin', 'good_link_ratio'], multiplot='quiet_selection', legend_labels=labels);
ax.set_xlabel("Side of square area (m)")
ax.set_ylabel("$C$")
graphs.save('interd_compare_quiet')
graphs.show()
def interd_dpll_vs_r12():
# 150 sample per point
graphs.GRAPH_OUTPUT_FORMAT = 'png'
graphs.change_fontsize(graphs.FONTSIZE + 2)
alldates = db.fetch_dates([20160828200017740, 20160828205450849]) # r12 vs dpll
dates = [alldates[0], alldates[-1]]
labels = []
labels.append('R12')
labels.append('DPLL')
ax = graphs.scatter_range(dates, ['nodecount', 'good_link_ratio'], multiplot='peak_detect', legend_labels=labels)
ax.set_xlabel("Number of nodes (M)")
ax.set_ylabel("$C$")
graphs.save('interd_dpll_vs_r12')
graphs.show()
def interd_cavg_vs_distance():
# 150 sample per point
graphs.GRAPH_OUTPUT_FORMAT = 'png'
graphs.change_fontsize(graphs.FONTSIZE + 2)
alldates = db.fetch_dates([20160825141108531, 20160825183253474]) # cavg vs dist
dates = [alldates[0], alldates[-1]]
ax = graphs.scatter_range(dates, ['max_dist_from_origin', 'good_link_ratio'], color='k')
ax.set_xlabel("Side of square area (Meters)")
ax.set_ylabel("$C$")
tick_locs = [500,750,1000,1250,1500]
tick_lbls = list(map(str, tick_locs))
ax.set_xticks(tick_locs)#, minor=False)
ax.set_xticklabels(tick_lbls)
graphs.save('interd_cavg_vs_distance')
graphs.show()
def kfold_train(learner_list,
data,
targets,
folds,
show_hair=True,
display=True):
"""Trains the list of learners in parallel. PARALLEL NOT IMPLEMENTED YET
All learners must be objects having the following function signature:
<learner>.train(x_train, y_train, x_test, y_test, num_epochs, train_batchsize, display=False)
learner_list: list of tuples: (learner_obj, args, kwargs)
"""
if type(learner_list) != type(list()):
raise ValueError(
'Expected a list of learners as a first argument, even for single learners'
)
for l in learner_list:
all_errors = []
accuracies = []
k = 0
tf = time.clock
t0 = tf()
for x_train, y_train, x_test, y_test in stratified_folds(data,
targets,
folds=folds):
if display:
if k != 0:
logger.info('Fold ' + str(k - 1) + ': done in ' + "%2.2f" %
(tf() - t0) + ' sec')
logger.info('Working on <' + type(l[0]).__name__ + '> fold ' +
str(k))
t0 = tf()
# If x/ydata is a string, assume it is a path and load it
epochs_errors, test_acc = l[0].train(x_train, y_train, x_test,
y_test, *l[1], **l[2])
all_errors.append(epochs_errors)
accuracies.append(test_acc)
k += 1
logger.debug(accuracies)
if show_hair:
graphs.train_and_test(all_errors,
legend_labels=('Training folds',
'Testing folds'))
graphs.show()
def interd_quiet_grids():
graphs.GRAPH_OUTPUT_FORMAT = 'png'
def init_simwrap():
sim = SimWrap(ctrl, p, cdict, pdict);
sim.set_all_nodisp()
sim.TO_show_clusters = False
sim.ctrl.saveall = False
return sim
def graph_mods():
ax.set_xlabel('Meters')
ax.set_ylabel('Meters')
ax.set_aspect('equal')
# Random grid
graphs.change_fontsize(graphs.FONTSIZE + 2)
ctrl, p, cdict, pdict = dec_grids()
ctrl.quiet_selection = 'random'
ctrl.quiet_nodes = ctrl.nodecount - 5
ctrl.steps = 1
sim = init_simwrap()
with warnings.catch_warnings():
warnings.filterwarnings("ignore")
sim.simulate()
ax = graphs.spatial_grid(ctrl, unit='m', s=12, show_broadcast=True)
graph_mods()
graphs.save('interd_grid_random')
graphs.show()
# cluster grid
ctrl, p, cdict, pdict = dec_grids()
ctrl.quiet_selection = 'kmeans'
ctrl.quiet_nodes = ctrl.nodecount - 5
ctrl.steps = 1
sim = init_simwrap()
with warnings.catch_warnings():
warnings.filterwarnings("ignore")
sim.simulate()
ax = graphs.spatial_grid(ctrl, unit='m', s=12, show_broadcast=True)
graph_mods()
graphs.save('interd_grid_kmeans')
graphs.show()
# sensing grid
ctrl, p, cdict, pdict = dec_grids()
ctrl.quiet_selection = 'contention'
ctrl.steps = 40
sim = SimWrap(ctrl, p, cdict, pdict);
sim.TO_show_clusters = False
sim.ctrl.saveall = False
with warnings.catch_warnings():
warnings.filterwarnings("ignore")
sim.simulate()
ax = graphs.spatial_grid(ctrl, unit='m', s=12, show_broadcast=True)
graph_mods()
graphs.save('interd_grid_sensing')
graphs.show()
def interd_dpll_final_compare():
#alldates = db.fetch_dates([20160829122809568, 20160829152128098]) # plain dpll vs contention
graphs.GRAPH_OUTPUT_FORMAT = 'png'
graphs.change_fontsize(15)
alldates = db.fetch_dates([20160828200017740, 20160828205450849]) # r12 vs dpll
dates = [alldates[0], alldates[-1]]
#extra ( add to scatter_range)
#sens_dates = db.fetch_dates([20160829122809568, 20160829152128098]) # plain dpll vs contention
#new_fetch_dict = {'date':sens_dates, 'quiet_selection':['contention']}
#raw_data = db.fetch_matching(new_fetch_dict, collist)
#datalist.append(np.array(raw_data))
#labels.append('Sensing')
labels = []
labels.append('R12')
labels.append('DPLL')
labels.append('DPLL-S')
ax = graphs.scatter_range(dates, ['nodecount', 'good_link_ratio'], multiplot='peak_detect', legend_labels=labels)
ax.set_xlabel("Number of nodes (M)")
ax.set_ylabel("$C$")
graphs.save('interd_dpll_final_compare')
graphs.show()
def ml_full_3d(noise_var=1, fct=lib.ll_redux_2d):
"""Graphs a 3d surface of the specified ML fct"""
p = ml_pinit_no_pulse_shape()
points = 10
t0 = 0
d0 = 0
t_halfwidth = 10
d_halfwidth = 10
t_min = t0-t_halfwidth
t_max = t0+t_halfwidth
d_min = d0-d_halfwidth
d_max = d0+d_halfwidth
t_step = t_halfwidth*2/points # int(round()) because integer samples
theta_range = np.arange(t_min, t_max, t_step)
deltaf_range = np.arange(d_min, d_max,d_halfwidth*2/points)*1e-6 # Deltaf in ppm
#deltaf_range = np.zeros(len(deltaf_range))
loglike = fct(p,0,0,theta_range,deltaf_range, var_w=noise_var)
#deltaf_range = np.arange(-1,1.1, 0.1)*1e-6
# Plot preparations
x = theta_range
y = deltaf_range*1e6
z = loglike*1e-3
# Plot prameters
fig, ax = graphs.surface3d(x, y, z, density=40)
ax.set_xlabel('Time offset (samples)')
ax.set_ylabel('CFO (ppm)')
ax.set_zlabel('Log likelihood (1e3)')
graphs.plt.tight_layout()
graphs.show()
return r[0]
df = pd.read_csv('https://sololearn.com/uploads/files/titanic.csv')
df['male'] = df['Sex'] == 'male'
X = df[[
'Pclass', 'male', 'Age', 'Siblings/Spouses', 'Parents/Children', 'Fare'
]].values
y = df['Survived'].values
X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=5)
model = LogisticRegression()
model.fit(X_train, y_train)
# Adjusting the threshhold
# y_pred = model.predict_proba(X_test)[:, 1] > 0.75
y_pred_proba = model.predict_proba(X_test)
fpr, tpr, thresholds = roc_curve(y_test, y_pred_proba[:, 1])
plt.plot(fpr, tpr)
plt.plot([0, 1], [0, 1], linestyle='--')
plt.xlim([0.0, 1.0])
plt.ylim([0.0, 1.0])
plt.xlabel('1 - specificity')
plt.ylabel('sensitivity')
plt.show()
# print("sensitivity:", sensitivity_score(y_test, y_pred))
# print("specificity:", specificity_score(y_test, y_pred))
nr = ("number of relations", "Number of similarity relations")
# Similarities at highest in-degree threshold
s1 = selectedSimilarities(data)[-1]
pl.init()
wordSimScatterPlot(s1, ws353, "$1 - L_1$", "WS-353")
wordSimScatterPlot(s1, mt287, "$1 - L_1$", "MT-287")
wordSimScatterPlot(s1, mt771, "$1 - L_1$", "MT-771")
wordSimScatterPlot(s1, simlex999, "$1 - L_1$", "SIMLEX-999")
level0Plot(data, md, rt)
level0Plot(data, md, nv)
level0Plot(data, md, ne)
histograms(data, "relative l1 norm", "Relative $\mathrm{L_{1}}$ norm")
histograms(data, "max error", "Error bound")
plotLevel0VsLevel1Stats(data, "max in-degree", "relative l1 norm",
"In-degree threshold",
"Relative $\mathrm{L_{1}}$ norm")
plotLevel0VsLevel1Stats(data, "max in-degree", "max error",
"In-degree threshold", "Error bound")
correlationsVsLevel0Prop(data, "max in-degree", "In-degree threshold",
[ws353, mt287, mt771, simlex999],
["WS-353", "MT-287", "MT-771", "SIMLEX-999"])
if show:
pl.show()
# Save figures under figure path, will not work if figures have already been displayed with .show()
for i in plt.get_fignums():
plt.figure(i)
plt.savefig(figurePath + '/' + experiment_name + '-figure%d.pdf' % i)
def highlited_regimes():
"""Highlights the converging vs the drift regime"""
compare_fontsize = 15
compare_aspect=12
props = dict(boxstyle='round', facecolor='wheat', alpha=0.5)
fontsize_tmp = graphs.FONTSIZE
def run_hair_graph(aspect='auto'):
hair_kwargs = {'y_label':r'$\theta_i$ $(T_0)$', 'show_clusters':False, 'savename':''}
hair_args = (
(ctrl.sample_inter , ctrl.theta_inter, ctrl),
hair_kwargs,
)
ax = graphs.hair(*hair_args[0], **hair_args[1])
ax.set_aspect(aspect)
return ax
# DRIFT REGIME
graphs.change_fontsize(compare_fontsize)
ctrl, p, cdict, pdict = dec_sample_theta()
sim = SimWrap(ctrl, p, cdict, pdict)
sim.conv_min_slope_samples = 15
sim.ctrl.keep_intermediate_values = True
sim.simulate()
ax = run_hair_graph(aspect=compare_aspect)
ax.text(0.85, 0.9, 'No DC', transform=ax.transAxes, fontsize=compare_fontsize, verticalalignment='top', bbox=props)
xmid = 12
xmax = ctrl.steps-13
xmin = 0
ymin, ymax = ax.get_ylim()
ya = ymin-0.06
ax.set_xlim([xmin,xmax])
ax.set_ylim([ya,ymax])
fname = 'latex_figures/init_theta'
graphs.save(fname)
graphs.show()
del ax
graphs.change_fontsize(15)
ax = run_hair_graph()
# Draw biarrows
ax.annotate('', xy=(xmid, ya), xycoords='data',
xytext=(xmin, ya), textcoords='data',
arrowprops=dict(arrowstyle="<->"))
ax.annotate('', xy=(xmax, ya), xycoords='data',
xytext=(xmid, ya), textcoords='data',
arrowprops=dict(arrowstyle="<->"))
ax.plot([xmid, xmid], [ya-1, ymax], 'k--')
# text
ax.text(xmid/2, ya, 'Transient', ha='center', va='bottom' )
ax.text((xmax-xmid)/2 +xmid, ya, 'Drifting', ha='center', va='bottom' )
ax.set_xlim([xmin,xmax])
ax.set_ylim([ya-0.03,ymax])
fname = 'latex_figures/highlighted_regimes'
graphs.save(fname)
graphs.show()
# THETA EXAMPLE
graphs.change_fontsize(compare_fontsize)
ctrl, p, cdict, pdict = dec_sample_theta()
ctrl.prop_correction = True
sim = SimWrap(ctrl, p, cdict, pdict)
sim.conv_min_slope_samples = 15
sim.ctrl.keep_intermediate_values = True
sim.show_CFO = False
sim.show_TO = False
sim.make_cat = False
sim.simulate()
ax = run_hair_graph(aspect=compare_aspect)
ax.text(0.85, 0.9, 'Q = 7', transform=ax.transAxes, fontsize=compare_fontsize, verticalalignment='top', bbox=props)
ax.set_xlim([xmin,xmax])
ax.set_ylim([ya,ymax])
fname = 'latex_figures/example_theta'
graphs.save(fname)
graphs.show()
# THETA EXAMPLE with Q=14
graphs.change_fontsize(compare_fontsize)
ctrl, p, cdict, pdict = dec_sample_theta()
ctrl.prop_correction = True
ctrl.pc_b, ctrl.pc_a = lib.hipass_avg(14)
sim = SimWrap(ctrl, p, cdict, pdict)
sim.conv_min_slope_samples = 15
sim.ctrl.keep_intermediate_values = True
sim.show_CFO = False
sim.show_TO = False
sim.make_cat = False
sim.simulate()
ax = run_hair_graph(aspect=compare_aspect)
ax.text(0.85, 0.9, 'Q = 14', transform=ax.transAxes, fontsize=compare_fontsize, verticalalignment='top', bbox=props)
ax.set_xlim([xmin,xmax])
ax.set_ylim([ya,ymax])
fname = 'latex_figures/example_theta_q14'
graphs.save(fname)
graphs.show()
graphs.change_fontsize(fontsize_tmp)
def ml_full_one(variable='CFO',noise_var=1):
"""Graphs only 1 axis of the log ML"""
p = ml_pinit()
points = 1000
t0 = 10
d0 = 0
t_halfwidth = 100
d_halfwidth = 1
t_min = t0-t_halfwidth
t_max = t0+t_halfwidth
d_min = d0-d_halfwidth
d_max = d0+d_halfwidth
t_step = max(int(round(t_halfwidth*2/points)),1) # int(round()) because integer samples
theta_range = np.arange(t_min, t_max, t_step)
deltaf_range = np.arange(d_min, d_max,d_halfwidth*2/points)*1e-6 # Deltaf in ppm
# x & z preparation
if variable == 'CFO':
loglike = lib.loglikelihood_1d_CFO(p,t0,d0,deltaf_range, var_w=noise_var)
x = deltaf_range*1e6
xlabel = 'CFO (ppm)'
elif variable == 'TO':
loglike = lib.loglikelihood_1d_TO(p,t0,d0,theta_range, var_w=noise_var)
x = theta_range
xlabel = 'TO (samples)'
z = loglike*1e-3
# Plot prametersZ
ax = graphs.continuous(x, z)
ax.set_xlabel(xlabel)
ax.set_ylabel('Log likelihood (1e3)')
graphs.plt.tight_layout()
graphs.show()
p = ml_pinit()
points = 40
t0 = 0
d0 = 0
t_halfwidth = 10
d_halfwidth = 1
t_min = t0-t_halfwidth
t_max = t0+t_halfwidth
d_min = d0-d_halfwidth
d_max = d0+d_halfwidth
t_step = t_halfwidth*2/points # int(round()) because integer samples
theta_range = np.arange(t_min, t_max, t_step)
deltaf_range = np.arange(d_min, d_max,d_halfwidth*2/points)*1e-6 # Deltaf in ppm
#deltaf_range = np.zeros(len(deltaf_range))
loglike = lib.ll_redux_2d(p,0,0,theta_range,deltaf_range, var_w=noise_var)
#deltaf_range = np.arange(-1,1.1, 0.1)*1e-6
# Plot preparations
x = theta_range
y = deltaf_range*1e6
z = loglike*1e-3
# Plot prameters
fig, ax = graphs.surface3d(x, y, z, density=40)
ax.set_xlabel('Time offset (samples)')
ax.set_ylabel('CFO (ppm)')
ax.set_zlabel('Log likelihood (1e3)')
graphs.plt.tight_layout()
graphs.show()
