def plot_outcomes(datafiles):
"""Plot the outcome of each scenario against the average assurance monitor value"""
# Compile AM values and outcomes of all data files with NO FAULT condition
no_fault_dfs = []
for df in datafiles:
# if len(df.fault_set) == 0 or fault_modes.SingularFaultModes.NO_FAULT in df.fault_set:
# no_fault_dfs.append(df)
no_fault_dfs.append(df)
top_occurred = np.zeros(len(no_fault_dfs))
collision_occurred = np.zeros(len(no_fault_dfs))
average_am = np.zeros(len(no_fault_dfs))
for i, df in enumerate(no_fault_dfs):
top_occurred[i] = df.top_occurred
collision_occurred[i] = df.collision_occurred
average_am[i] = df.am_avg_before_top
avg_prob = ((np.sum(top_occurred) + 1) / float(len(top_occurred) + 2))
print("Average TOP Probability: %f" % avg_prob)
# Fit a conditional binomial distribution to the data with max-likelihood
fit_results, _ = function_fitting.max_likelihood_conditional_fit(average_am, top_occurred)
x_range = np.linspace(np.min(average_am), np.max(average_am), 100)
y_sig_max_likelihood = function_fitting.bounded_sigmoid(x_range, *fit_results.x)
print("Sigmoid Fit: ", fit_results)
# Plot collisions and top events on same figure
# Plot conditional probability P(B | AM) and adjusted probability P(B | AM) / P(B)
fig = plt.figure(dpi=300)
ax1 = fig.add_subplot(1, 1, 1)
# ax1.scatter(average_am, top_occurred, label="Top Events")
# ax1.scatter(average_am, collision_occurred, label="Collision Events")
ax1.plot(x_range, y_sig_max_likelihood, label="MLE fit")
# ax1.plot(x_range, y_sig_max_likelihood/avg_prob, label="Relative to baseline")
ax1.axhline(avg_prob, linestyle="-.", label="baseline", color="green")
# ax1.axhline(1.0, linestyle="--", color="purple")
# ax1.set_xlabel("Average AM Log Martingale")
# ax1.set_ylabel("Outcome (True/False)")
ax1.set_xlabel("Assurance Monitor output")
ax1.set_ylabel("Barrier Failure Probability")
ax1.set_title("P(B1) vs. Assurance Monitor output")
ax1.legend(loc="best")
ax1.set_ylim([0.0, np.max(y_sig_max_likelihood/avg_prob) + 0.1])
plt.show()
def plot_outcomes(datafiles):
"""Plot the outcome of each scenario against the average assurance monitor value"""
# Compile outcomes of each sim run
top_occurred = np.zeros(len(datafiles))
collision_occurred = np.zeros(len(datafiles))
thruster_degradation = np.zeros(len(datafiles))
closest_approach = np.zeros(len(datafiles))
for i, df in enumerate(datafiles):
top_occurred[i] = df.top_occurred.astype(int)
collision_occurred[i] = df.consequence_occurred.astype(int)
thruster_degradation[i] = df.thruster_degradation_amount
closest_approach[i] = df.closest_approach
# Datasets with no top events or collisions can cause issues with max likelihood estimation. Print warning.
if np.sum(top_occurred) == 0:
print("WARNING: No Top events occurred in the provided datasets.")
if np.sum(collision_occurred) == 0:
print("WARNING: No Collision events occurred in the provided datasets.")
# Calculate some simple statistics
avg_top_prob = ((np.sum(top_occurred) + 1) / float(len(top_occurred) + 2))
avg_col_prob = ((np.sum(collision_occurred) + 1) / float(len(collision_occurred) + 2))
avg_col_prob_after_top = ((np.sum(collision_occurred) + 1) / float(np.sum(top_occurred) + 2))
avg_ca = np.sum(closest_approach) / float(len(closest_approach))
print("Average TOP Probability: %f" % avg_top_prob)
print("Average Collision Probability: %f" % avg_col_prob)
print("Average Collision Probability given TOP occurred: %f" % avg_col_prob_after_top)
print("Average closest approach (m): %f" % avg_ca)
print("\n\n")
# Fit a conditional binomial distribution to the data with max-likelihood for the TOP event
x_range = np.linspace(np.min(thruster_degradation), np.max(thruster_degradation), 100)
fit_results = None
try:
fit_results, _ = function_fitting.max_likelihood_conditional_fit(thruster_degradation, top_occurred)
except ValueError as e:
print("Exception occurred when fitting sigmoid for TOP event:\n%s" % str(e))
else:
y_sig_max_likelihood = function_fitting.bounded_sigmoid(x_range, *fit_results.x)
adjusted_likelihood = y_sig_max_likelihood/avg_top_prob
print("TOP Sigmoid Fit: ", fit_results)
print("\n\n")
# Divide results into bins and average each bin
bins = {"x": [], "y": {"top": [], "col": []}}
bins_avg = {"x": [], "y": {"top": [], "col": []}}
step_size = 0.2
allowed_skew = 0.01
for i in range(4):
bin_x = step_size * i
indices = np.logical_and(thruster_degradation > bin_x - allowed_skew, thruster_degradation < bin_x + allowed_skew)
bins["x"].append(thruster_degradation[indices])
bins["y"]["top"].append(top_occurred[indices])
bins["y"]["col"].append(collision_occurred[indices])
avg_deg = np.average(thruster_degradation[indices])
avg_top_prob = (np.sum(top_occurred[indices]) + 1) / float(len(top_occurred[indices]) + 2)
avg_col_prob = (np.sum(collision_occurred[indices]) + 1) / float(len(collision_occurred[indices]) + 2)
bins_avg["x"].append(avg_deg)
bins_avg["y"]["top"].append(avg_top_prob)
bins_avg["y"]["col"].append(avg_col_prob)
print("Bin %.2f contains %d entries.\n\tAverage probability of TOP: %.2f\n\tAverage probability of Collision: %.2f"
% (avg_deg, len(thruster_degradation[indices]), avg_top_prob, avg_col_prob))
print("\n\n")
# Plot conditional probability P(B | D) and adjusted probability P(B | D) / P(B) for TOP barrier
fig1 = plt.figure(dpi=300)
ax1 = fig1.add_subplot(1, 1, 1)
# ax1.scatter(thruster_degradation, top_occurred, label="Top Events")
# ax1.scatter(thruster_degradation, collision_occurred, label="Collision Events")
ax1.scatter(bins_avg["x"], bins_avg["y"]["top"], label="Top Event Prob.")
ax1.scatter(np.average(thruster_degradation), avg_top_prob, label="Average")
if fit_results is not None:
ax1.plot(x_range, y_sig_max_likelihood, label="Sigmoid MLE fit")
ax1.plot(x_range, adjusted_likelihood, label="Adjusted fit")
ax1.set_ylim([0.0, np.max(adjusted_likelihood)])
else:
ax1.set_ylim([0.0, 1.0])
ax1.set_xlabel("Thruster Degradation Percentage")
ax1.set_ylabel("TOP Event Probability")
ax1.legend(loc="best")
# Fit a conditional binomial distribution to the data with max-likelihood for the TOP event
fit_results = None
try:
fit_results, _ = function_fitting.max_likelihood_conditional_fit(thruster_degradation, collision_occurred)
except ValueError as e:
print("Exception occurred when fitting sigmoid for Collision event:\n%s" % str(e))
else:
collision_y_sigmoid = function_fitting.bounded_sigmoid(x_range, *fit_results.x)
collision_adj_likelihood = collision_y_sigmoid / avg_col_prob
print("Collision Sigmoid Fit: ", fit_results)
print("\n\n")
# Plot conditional probability P(B | D) and adjusted probability P(B | D) / P(B) for Collision barrier
fig2 = plt.figure(dpi=300)
ax2 = fig2.add_subplot(1, 1, 1)
# ax2.scatter(thruster_degradation, top_occurred, label="Top Events")
# ax2.scatter(thruster_degradation, collision_occurred, label="Collision Events")
ax2.scatter(bins_avg["x"], bins_avg["y"]["col"], label="Collision Event Prob.")
ax2.scatter(np.average(thruster_degradation), avg_col_prob, label="Average")
if fit_results is not None:
ax2.plot(x_range, collision_y_sigmoid, label="Sigmoid MLE fit")
ax2.plot(x_range, collision_adj_likelihood, label="Adjusted fit")
ax2.set_ylim([0.0, np.max(collision_adj_likelihood)])
else:
ax2.set_ylim([0.0, 1.0])
ax2.set_xlabel("Thruster Degradation Percentage")
ax2.set_ylabel("Collision Probability")
ax2.legend(loc="best")
# Show plots
plt.show()
def cont_analysis(cd):
# Calculate some simple statistics
avg_top_prob = ((np.sum(cd.top_events) + 1) /
float(len(cd.top_events) + 2))
avg_col_prob = ((np.sum(cd.consequences) + 1) /
float(len(cd.consequences) + 2))
avg_col_prob_after_top = ((np.sum(cd.consequences) + 1) /
float(np.sum(cd.top_events) + 2))
print("Average TOP Probability: %f" % avg_top_prob)
print("Average Consequence Probability: %f" % avg_col_prob)
print("Average Consequence Probability given TOP occurred: %f" %
avg_col_prob_after_top)
print("\n\n")
for key, values in cd.list_independent_vars_cont.items():
values_np = np.array(values)
top_np = np.array(cd.top_events)
cons_np = np.array(cd.consequences)
# # Fit a conditional binomial distribution to the data with max-likelihood for the TOP event
x_range = np.linspace(np.min(values), np.max(values), 100)
fit_results = None
try:
fit_results, _ = function_fitting.max_likelihood_conditional_fit(
values_np, top_np)
except ValueError as e:
print(
"Exception occurred when fitting sigmoid for TOP event:\n%s" %
str(e))
else:
y_sig_max_likelihood = function_fitting.bounded_sigmoid(
x_range, *fit_results.x)
adjusted_likelihood = y_sig_max_likelihood / avg_top_prob
print("TOP Sigmoid Fit: ", fit_results)
print("\n\n")
#todo add this back for plots
# Plot conditional probability P(B | D) and adjusted probability P(B | D) / P(B) for TOP barrier
fig1 = plt.figure(dpi=300)
ax1 = fig1.add_subplot(1, 1, 1)
ax1.scatter(values_np, top_np, label="Top Events")
ax1.scatter(values_np, cons_np, label="Collision Events")
# ax1.scatter(bins_avg["x"], bins_avg["y"]["top"], label="Top Event Prob.")
ax1.scatter(values_np, top_np, label="Top Events")
ax1.scatter(np.average(values_np), avg_top_prob, label="Average")
if fit_results is not None:
ax1.plot(x_range, y_sig_max_likelihood, label="Sigmoid MLE fit")
ax1.plot(x_range, adjusted_likelihood, label="Adjusted fit")
ax1.set_ylim([0.0, np.max(adjusted_likelihood)])
else:
ax1.set_ylim([0.0, 1.0])
ax1.set_xlabel("Indep. Var Percentage")
ax1.set_ylabel("TOP Event Probability")
ax1.legend(loc="best")
# Fit a conditional binomial distribution to the data with max-likelihood for the TOP event
fit_results = None
try:
fit_results, _ = function_fitting.max_likelihood_conditional_fit(
values_np, cons_np)
except ValueError as e:
print(
"Exception occurred when fitting sigmoid for Collision event:\n%s"
% str(e))
else:
collision_y_sigmoid = function_fitting.bounded_sigmoid(
x_range, *fit_results.x)
collision_adj_likelihood = collision_y_sigmoid / avg_col_prob
print("Collision Sigmoid Fit: ", fit_results)
print("\n\n")
# Plot conditional probability P(B | D) and adjusted probability P(B | D) / P(B) for Collision barrier
fig2 = plt.figure(dpi=300)
ax2 = fig2.add_subplot(1, 1, 1)
# ax2.scatter(thruster_degradation, top_occurred, label="Top Events")
ax2.scatter(values_np, cons_np, label="Collision Events")
# ax2.scatter(bins_avg["x"], bins_avg["y"]["col"], label="Collision Event Prob.")
ax2.scatter(np.average(values_np), avg_col_prob, label="Average")
if fit_results is not None:
ax2.plot(x_range, collision_y_sigmoid, label="Sigmoid MLE fit")
ax2.plot(x_range, collision_adj_likelihood, label="Adjusted fit")
ax2.set_ylim([0.0, np.max(collision_adj_likelihood)])
else:
ax2.set_ylim([0.0, 1.0])
ax2.set_xlabel("Indep. Var Percentage")
ax2.set_ylabel("Collision Probability")
ax2.legend(loc="best")
# Show plots
plt.show()