def graph_utility_scaled(metrics: Metrics, path_prefix: str):
fig = plt.figure()
ax = fig.gca()
sources_utilities = {(b.source, b.capability) for b in metrics.buffers}
grouped_utility = {
(asrc, acap): [
(b.t, b.utility / b.max_utility)
for b in metrics.buffers
if asrc == b.source and acap == b.capability
]
for (asrc, acap) in sources_utilities
}
for ((src, cap), utilities) in sorted(grouped_utility.items(), key=lambda x: x[0]):
X, Y = zip(*utilities)
ax.plot(X, Y, label=f"{src} {cap}")
ax.set_ylim(0, 1)
ax.set_xlabel('Time (secs)')
ax.set_ylabel('Normalised Utility (\\%)')
ax.yaxis.set_major_formatter(ticker.PercentFormatter(xmax=1, symbol=''))
ax.legend(bbox_to_anchor=(1.5, 1), loc="upper right", ncol=2)
savefig(fig, f"{path_prefix}norm-utility.pdf")
plt.close(fig)
def graph_max_utility(metrics: Metrics, path_prefix: str):
fig = plt.figure()
ax = fig.gca()
sources_utilities = {(b.source, b.capability) for b in metrics.buffers}
grouped_utility = {
(asrc, acap): [
(b.t, b.max_utility)
for b in metrics.buffers
if asrc == b.source and acap == b.capability
]
for (asrc, acap) in sources_utilities
}
for ((src, cap), utilities) in sorted(grouped_utility.items(), key=lambda x: x[0]):
X, Y = zip(*utilities)
ax.plot(X, Y, label=f"{src} {cap}")
ax.set_ylim(0, 1)
ax.set_xlabel('Time (secs)')
ax.set_ylabel('Maximum Utility (\\%)')
ax.legend(bbox_to_anchor=(1.275, 1), loc="upper right", ncol=1)
savefig(fig, f"{path_prefix}max-utility.pdf")
plt.close(fig)
def graph_evictions(metrics: Metrics, path_prefix: str):
columns = ["crypto", "trust", "reputation", "stereotype"]
column_to_data = {
"crypto": metrics.evicted_crypto,
"trust": metrics.evicted_trust,
"reputation": metrics.evicted_reputation,
"stereotype": metrics.evicted_stereotype
}
agents = set(chain.from_iterable(
{a for (t, a, i) in data}
for data in column_to_data.values()
))
agents = list(sorted(agents))
def sanitise_i(column, i):
if column == "crypto" or column == "reputation":
return i[0]
else:
return f"{i[0]}-{i[1]}"
all_evictions = {
(agent, column): list(sorted([(t, sanitise_i(column, i)) for (t, a, i) in column_data if a == agent], key=lambda x: x[1]))
for agent in agents
for (column, column_data) in column_to_data.items()
}
fig, axs = plt.subplots(nrows=max(1, len(agents)), ncols=max(1, len(columns)), sharex=True, squeeze=False, figsize=(18,30))
for (agent, col) in itertools.product(agents, columns):
try:
evictions = all_evictions[(agent, col)]
except KeyError:
# Skip when there are no evictions
continue
# Skip when there were no evictions
if not evictions:
continue
X, Y = zip(*evictions)
ax = axs[agents.index(agent), columns.index(col)]
ax.scatter(X, Y)
ax.title.set_text(f"{agent} {col}")
ax.tick_params(axis='y', labelsize="small")
fig.subplots_adjust(hspace=0.35)
savefig(fig, f"{path_prefix}evictions.pdf")
plt.close(fig)
def graph_interactions(metrics: Metrics, path_prefix: str):
keys = {(b.target, b.capability) for b in metrics.buffers}
all_interactions = {
(target, capability): [
(b.t, f"{b.outcome.name} (Imp)" if np.isnan(b.utility) else b.outcome.name)
for b in metrics.buffers
if b.target == target and b.capability == capability
]
for (target, capability) in keys
}
agents, capabilities = zip(*all_interactions.keys())
agents = list(sorted(set(agents)))
capabilities = list(sorted(set(capabilities)))
fig, axs = plt.subplots(nrows=len(agents), ncols=len(capabilities), sharex=True, squeeze=False, figsize=(18,20))
yaxis_categories = [obs.name for obs in InteractionObservation]
for (agent, cap) in itertools.product(agents, capabilities):
try:
interactions = all_interactions[(agent, cap)]
except KeyError:
# Skip when there are no records
continue
# Skip when there were no interactions
if not interactions:
continue
X, Y = zip(*interactions)
ax = axs[agents.index(agent), capabilities.index(cap)]
ax.scatter(X, Y)
ax.title.set_text(f"{agent} {cap}")
#ax.set_xlim(min(X), max(X))
#ax.plot([min(X)-100+i for i in range(len(yaxis_categories))], yaxis_categories)
#ax.set_yticks(yaxis_categories)
#ax.set_ylim(0 - 0.5, len(yaxis_categories) - 1 + 0.5)
fig.subplots_adjust(hspace=0.35)
savefig(fig, f"{path_prefix}interactions.pdf")
plt.close(fig)
def graph_interactions_utility_hist(metrics: Metrics, path_prefix: str):
correct = [
b.utility
for b
in metrics.buffers
if b.outcome == InteractionObservation.Correct
]
incorrect = [
b.utility
for b
in metrics.buffers
if b.outcome == InteractionObservation.Incorrect
if not np.isnan(b.utility)
]
incorrect_imp = [
b.utility
for b
in metrics.buffers
if b.outcome == InteractionObservation.Incorrect
if np.isnan(b.utility)
]
bins = np.arange(0, 1, 0.02)
fig = plt.figure()
ax = fig.gca()
ax.hist([correct, incorrect, incorrect_imp], bins, histtype='bar', stacked=True,
label=[f"Correct ({len(correct)})", f"Incorrect ({len(incorrect)})", f"Incorrect (Imp) ({len(incorrect_imp)})"])
ax.legend()
ax.set_xlim(0, 1)
ax.set_xlabel('Utility (\\%)')
ax.set_ylabel('Interaction Count')
savefig(fig, f"{path_prefix}interactions-utility-hist.pdf")
plt.close(fig)
def graph_behaviour_state(metrics: Metrics, path_prefix: str):
agents, capabilities = zip(*metrics.behaviour_changes.keys())
agents = list(sorted(set(agents)))
capabilities = list(sorted(set(capabilities)))
fig, axs = plt.subplots(nrows=len(agents), ncols=len(capabilities), sharex=True, squeeze=False, figsize=(18,20))
yaxis_categories = [obs.name for obs in CapabilityBehaviourState]
for (agent, cap) in itertools.product(agents, capabilities):
try:
behaviour = metrics.behaviour_changes[(agent, cap)]
except KeyError:
# Skip when there are no records
continue
# Skip when there were no interactions
if not behaviour:
continue
X, Y = zip(*behaviour)
Y = [y.name for y in Y]
ax = axs[agents.index(agent), capabilities.index(cap)]
ax.scatter(X, Y)
ax.title.set_text(f"{agent} {cap}")
#ax.set_xlim(min(X), max(X))
#ax.plot([min(X)-100+i for i in range(len(yaxis_categories))], yaxis_categories)
#ax.set_yticks(yaxis_categories)
#ax.set_ylim(0 - 0.5, len(yaxis_categories) - 1 + 0.5)
fig.subplots_adjust(hspace=0.35)
savefig(fig, f"{path_prefix}behaviour_state.pdf")
plt.close(fig)
def graph_utility_scaled_cap_colour(metrics: Metrics, path_prefix: str):
fig = plt.figure()
ax = fig.gca()
sources_utilities = {(b.source, b.capability) for b in metrics.buffers}
grouped_utility = {
(asrc, acap): [
(b.t, b.utility / b.max_utility)
for b in metrics.buffers
if asrc == b.source and acap == b.capability
]
for (asrc, acap) in sources_utilities
}
sequential_cmaps = [seaborn.mpl_palette(name, n_colors=len(metrics.agent_names)) for name in ("Greens", "Purples")]
cmap_for_cap = {
c: sequential_cmaps[c]
for c in {int(c[1:]) for c in metrics.capability_names}
}
for ((src, cap), utilities) in sorted(grouped_utility.items(), key=lambda x: x[0]):
X, Y = zip(*utilities)
ax.plot(X, Y, label=f"{src} {cap}", color=cmap_for_cap[int(cap[1:])][metrics.agent_names.index(src)])
ax.set_ylim(0, 1)
ax.set_xlabel('Time (secs)')
ax.set_ylabel('Normalised Utility (\\%)')
ax.yaxis.set_major_formatter(ticker.PercentFormatter(xmax=1, symbol=''))
ax.legend(bbox_to_anchor=(1.5, 1), loc="upper right", ncol=2)
savefig(fig, f"{path_prefix}norm-utility-cc.pdf")
plt.close(fig)
def graph_interactions_performed(metrics: Metrics, path_prefix: str):
all_interactions = {
(agent, capability): [t for (t, a, c) in metrics.interaction_performed if a == agent and c == capability]
for agent in metrics.agent_names
for capability in metrics.capability_names
}
fig, axs = plt.subplots(nrows=len(metrics.agent_names), ncols=len(metrics.capability_names), sharex=True, squeeze=False, figsize=(18,30))
for (agent, col) in itertools.product(metrics.agent_names, metrics.capability_names):
try:
interactions = all_interactions[(agent, col)]
except KeyError:
# Skip when there are no evictions
continue
# Skip when there were no evictions
if not interactions:
continue
ax = axs[metrics.agent_names.index(agent), metrics.capability_names.index(col)]
bins = np.arange(min(interactions), max(interactions), 5)
ax.hist(interactions, bins, histtype='bar')
ax.title.set_text(f"{agent} {col}")
ax.tick_params(axis='y', labelsize="small")
fig.subplots_adjust(hspace=0.35)
savefig(fig, f"{path_prefix}interactions-performed.pdf")
plt.close(fig)