def evalPerturbationPercent(df):
interestedDim = df.index.get_level_values('interestedDim')[0]
interestedAgentID = df.index.get_level_values('interestedAgentID')[0]
varsNotInTraj = ['interestedDim', 'interestedAgentID']
DIRNAME = os.path.dirname(__file__)
trajectoryDirectory = os.path.join(DIRNAME, '..', '..', 'data', 'evaluatePerturbation', 'trajectories')
if not os.path.exists(trajectoryDirectory):
os.makedirs(trajectoryDirectory)
trajectoryFixedParameters = {'numWolves': 3, 'maxRunningStepsToSample': 101}
trajectoryExtension = '.pickle'
getTrajectorySavePath = GetSavePath(trajectoryDirectory, trajectoryExtension, trajectoryFixedParameters)
# Compute Statistics on the Trajectories
loadTrajectories = LoadTrajectories(getTrajectorySavePath, loadFromPickle)
trajVariables = readParametersFromDf(df)
[trajVariables.pop(var) for var in varsNotInTraj]
allTrajectories = loadTrajectories(trajVariables)
calculatePercentageActionChange = CalculatePercentageActionChange(interestedDim, interestedAgentID)
allMeasurements = np.array([calculatePercentageActionChange(trajectory) for trajectory in allTrajectories])
measurementMean = np.mean(allMeasurements, axis=0)
measurementSe = np.std(allMeasurements, axis=0) / np.sqrt(len(allTrajectories) - 1)
return pd.Series({'mean': measurementMean, 'se': measurementSe})
def main():
# manipulated variables
# manipulated variables
manipulatedVariables = OrderedDict()
manipulatedVariables['numSheep'] = [2, 4]
manipulatedVariables['shared'] = ['no', 'reward', 'agencyByIndividualWolf']#['no', 'reward', 'agencyBySharedWolf', 'agencyByIndividualWolf']
manipulatedVariables['sheepType'] = ['sheepWithIndividualRewardWolf']#['sheepWithSharedRewardWolf', 'copySheepWithSharedRewardWolf', 'sheepWithIndividualRewardWolf', 'copySheepWithIndividualRewardWolf']
levelNames = list(manipulatedVariables.keys())
levelValues = list(manipulatedVariables.values())
modelIndex = pd.MultiIndex.from_product(levelValues, names=levelNames)
toSplitFrame = pd.DataFrame(index=modelIndex)
productedValues = it.product(*[[(key, value) for value in values] for key, values in manipulatedVariables.items()])
parametersAllCondtion = [dict(list(specificValueParameter)) for specificValueParameter in productedValues]
DIRNAME = os.path.dirname(__file__)
trajectoryDirectory = os.path.join(DIRNAME, '..', '..', 'data', 'evaluateHierarchyPlanningEnvMADDPG',
'trajectories')
if not os.path.exists(trajectoryDirectory):
os.makedirs(trajectoryDirectory)
maxRunningSteps = 103
numWolves = 3
trajectoryFixedParameters = {'numWolves': numWolves, 'maxRunningSteps': maxRunningSteps}
trajectoryExtension = '.pickle'
getTrajectorySavePath = GetSavePath(trajectoryDirectory, trajectoryExtension, trajectoryFixedParameters)
# Compute Statistics on the Trajectories
loadTrajectories = LoadTrajectories(getTrajectorySavePath, loadFromPickle)
#loadTrajectoriesFromDf = lambda df: [trajectory for trajectory in loadTrajectories(readParametersFromDf(df)) if len(trajectory) < readParametersFromDf(df)['maxRunningSteps']]
loadTrajectoriesFromDf = lambda df: loadTrajectories(readParametersFromDf(df))
measureIntentionArcheivement = lambda df: lambda trajectory: np.sum(np.array(trajectory)[:, 3])
computeStatistics = ComputeStatistics(loadTrajectoriesFromDf, measureIntentionArcheivement)
statisticsDf = toSplitFrame.groupby(levelNames).apply(computeStatistics)
print(statisticsDf)
#__import__('ipdb').set_trace()
fig = plt.figure()
fig.set_dpi(120)
numColumns = len(manipulatedVariables['sheepType'])
numRows = 1#len(manipulatedVariables['sheepType'])
plotCounter = 1
sheepTypeTable = {'sheepWithSharedRewardWolf': 'No Copy Shared',
'copySheepWithSharedRewardWolf': 'Copy Shared',
'sheepWithIndividualRewardWolf': 'No Copy Individual',
'copySheepWithIndividualRewardWolf': 'Copy Individual'}
wolfTypeTable = {'no': 'Individual Reward', 'reward': 'Shared Reward',
'agencyBySharedWolf': 'Shared Agency',
'agencyByIndividualWolf': 'Shared Agency'}
for key, group in statisticsDf.groupby(['sheepType']):
group.index = group.index.droplevel(['sheepType'])
axForDraw = fig.add_subplot(numRows, numColumns, plotCounter)
if (plotCounter) % max(numColumns, 2) == 1:
axForDraw.set_ylabel('Average Bite')
#if plotCounter <= numColumns:
# axForDraw.set_title(str(sheepTypeTable[key]))
numWolvesLabels = ['2', '3']
for wolfType, grp in group.groupby('shared'):
grp['mean'] = grp['mean']-5
grp.index = grp.index.droplevel('shared')
grp.plot.line(ax = axForDraw, y = 'mean', yerr = 'se', label = wolfTypeTable[wolfType], ylim = (0, 15), marker = 'o', rot = 0 )
if int((plotCounter - 1) / numColumns) == numRows - 1:
axForDraw.xaxis.set_label_text('Number of Sheep')
else:
xAxis = axForDraw.get_xaxis()
xLabel = xAxis.get_label()
xLabel.set_visible(False)
plt.xticks([2, 4])
plotCounter = plotCounter + 1
plt.suptitle('3 Wolves')
#fig.text(x = 0.5, y = 0.92, s = 'Wolf Type', ha = 'center', va = 'center')
#fig.text(x = 0.05, y = 0.5, s = 'Sheep Type', ha = 'center', va = 'center', rotation=90)
plt.show()
def main():
# manipulated variables
manipulatedVariables = OrderedDict()
manipulatedVariables['numSheep'] = [2, 4, 8]
manipulatedVariables['hierarchy'] = [0, 1, 2]
manipulatedVariables['valuePriorEndTime'] = [-100, 0, 100]
levelNames = list(manipulatedVariables.keys())
levelValues = list(manipulatedVariables.values())
modelIndex = pd.MultiIndex.from_product(levelValues, names=levelNames)
toSplitFrame = pd.DataFrame(index=modelIndex)
productedValues = it.product(
*[[(key, value) for value in values]
for key, values in manipulatedVariables.items()])
parametersAllCondtion = [
dict(list(specificValueParameter))
for specificValueParameter in productedValues
]
DIRNAME = os.path.dirname(__file__)
trajectoryDirectory = os.path.join(DIRNAME, '..', '..', 'data',
'evaluateHierarchyPlanning',
'trajectories')
if not os.path.exists(trajectoryDirectory):
os.makedirs(trajectoryDirectory)
NNNumSimulations = 200
maxRunningSteps = 50
wolfPolicySoft = 2.5
sheepPolicySoft = 2.5
numWolves = 3
trajectoryFixedParameters = {
'numWolves': numWolves,
'wolfPolicySoft': wolfPolicySoft,
'sheepPolicySoft': sheepPolicySoft,
'maxRunningSteps': maxRunningSteps,
'NNNumSimulations': NNNumSimulations
}
trajectoryExtension = '.pickle'
getTrajectorySavePath = GetSavePath(trajectoryDirectory,
trajectoryExtension,
trajectoryFixedParameters)
# Compute Statistics on the Trajectories
loadTrajectories = LoadTrajectories(getTrajectorySavePath, loadFromPickle)
#loadTrajectoriesFromDf = lambda df: [trajectory for trajectory in loadTrajectories(readParametersFromDf(df)) if len(trajectory) < readParametersFromDf(df)['maxRunningSteps']]
loadTrajectoriesFromDf = lambda df: loadTrajectories(
readParametersFromDf(df))
priorIndexinTimestep = 4
composeInterpolateFunction = lambda df: Interpolate1dData(maxRunningSteps)
#getImaginedWeIdsForCrossEntropy = lambda df: [readParametersFromDf(df)['numSheep'], list(range(readParametersFromDf(df)['numSheep'] + 1, readParametersFromDf(df)['numSheep'] +
# readParametersFromDf(df)['numWolves']))]
#composeMeasureCrossEntropy = lambda df: MeasureCrossEntropy(getImaginedWeIdsForCrossEntropy(df), priorIndexinTimestep)
#measureFunction = lambda df: lambda trajectory: composeInterpolateFunction(df)(composeMeasureCrossEntropy(df)(trajectory))
getImaginedWeIdsForConvergeRate = lambda df: list(
range(
readParametersFromDf(df)['numSheep'],
readParametersFromDf(df)['numSheep'] + numWolves))
measureConvergeRate = lambda df: MeasureConvergeOnGoal(
getImaginedWeIdsForConvergeRate(df), priorIndexinTimestep,
sampleFromDistribution)
measureFunction = lambda df: lambda trajectory: composeInterpolateFunction(
df)(measureConvergeRate(df)(trajectory))
computeStatistics = ComputeStatistics(loadTrajectoriesFromDf,
measureFunction)
statisticsDf = toSplitFrame.groupby(levelNames).apply(computeStatistics)
fig = plt.figure()
numColumns = len(manipulatedVariables['hierarchy'])
numRows = len(manipulatedVariables['valuePriorEndTime'])
plotCounter = 1
hierarchyLabels = ['noHierarchy9', 'noHierarchy5', 'Hierarchy9']
for key, group in statisticsDf.groupby(['valuePriorEndTime', 'hierarchy']):
group.index = group.index.droplevel(['valuePriorEndTime', 'hierarchy'])
axForDraw = fig.add_subplot(numRows, numColumns, plotCounter)
#if plotCounter % numColumns == 1:
#axForDraw.set_ylabel('Cross Entropy')
axForDraw.set_ylabel('Converge Rate')
if plotCounter <= numColumns:
axForDraw.set_title(hierarchyLabels[key[1]])
for numSheep, grp in group.groupby('numSheep'):
df = pd.DataFrame(grp.values[0].tolist(),
columns=list(range(maxRunningSteps)),
index=['mean', 'se']).T
#df.plot.line(ax = axForDraw, label = 'Set Size of Intentions = {}'.format(numSheep), y = 'mean', yerr = 'se', ylim = (0, 3), rot = 0)
df.plot.line(ax=axForDraw,
label='numSheep = {}'.format(numSheep),
y='mean',
yerr='se',
ylim=(0, 1),
rot=0)
plotCounter = plotCounter + 1
plt.suptitle(str(numWolves) + 'Wolves')
#plt.legend(loc='best')
fig.text(x=0.5, y=0.04, s='hierarchy', ha='center', va='center')
fig.text(x=0.05,
y=0.5,
s='valuePriorEndTime',
ha='center',
va='center',
rotation=90)
plt.show()
def main():
# manipulated variables
manipulatedVariables = OrderedDict()
manipulatedVariables['numWolves'] = [2] # temp just 2
manipulatedVariables['numSheep'] = [1] # temp just 1
manipulatedVariables['inferenceSoft'] = [1.0]
manipulatedVariables['wolfPolicySoft'] = [2.5]
manipulatedVariables['otherCompeteRate'] = [
0.0, 0.5, 1.0
] # 0 never compete, 1 always compete
manipulatedVariables['competeDetectionRate'] = [
0.0, 0.5
] # 0 never detect compete, 1 only detect compete
levelNames = list(manipulatedVariables.keys())
levelValues = list(manipulatedVariables.values())
modelIndex = pd.MultiIndex.from_product(levelValues, names=levelNames)
toSplitFrame = pd.DataFrame(index=modelIndex)
productedValues = it.product(
*[[(key, value) for value in values]
for key, values in manipulatedVariables.items()])
parametersAllCondtion = [
dict(list(specificValueParameter))
for specificValueParameter in productedValues
]
DIRNAME = os.path.dirname(__file__)
trajectoryDirectory = os.path.join(DIRNAME, '..', '..', 'data',
'evaluateCompeteDetection',
'trajectories')
if not os.path.exists(trajectoryDirectory):
os.makedirs(trajectoryDirectory)
NNNumSimulations = 250
maxRunningSteps = 40
sheepPolicySoft = 1.0
competePolicy = 'heatseeking'
trajectoryFixedParameters = {
'competePolicy': competePolicy,
'sheepPolicySoft': sheepPolicySoft,
'maxRunningSteps': maxRunningSteps,
'NNNumSimulations': NNNumSimulations
}
heatseekingPrecesion = 50
trajectoryFixedParameters.update(
{'heatseekingPrecesion': heatseekingPrecesion})
print(trajectoryFixedParameters)
trajectoryExtension = '.pickle'
getTrajectorySavePath = GetSavePath(trajectoryDirectory,
trajectoryExtension,
trajectoryFixedParameters)
# Compute Statistics on the Trajectories
loadTrajectories = LoadTrajectories(getTrajectorySavePath, loadFromPickle)
loadTrajectoriesFromDf = lambda df: loadTrajectories(
readParametersFromDf(df))
maxSteps = 50
measureFunction = lambda df: lambda trajectory: int(
np.linalg.norm(trajectory[-1][0][0] - trajectory[-1][0][1]) <= 50
) - 0.00 * min(len(trajectory), maxSteps)
def measureSelfReward(trajectory):
stateLastTimeStep = trajectory[-1][0]
goalState, selfState, otherState = stateLastTimeStep
selfReward = int(np.linalg.norm(goalState - selfState) <= 50)
otherReward = int(np.linalg.norm(goalState - otherState) <= 50)
intentionLastTimeStep = trajectory[-1][4]
selfIntention, otherIntention = intentionLastTimeStep
selfShare = False
if (0, (1, 2)) in list(selfIntention):
if selfIntention[(0, (1, 2))] >= 0.8:
selfShare = True
otherShare = False
if (0, (1, 2)) in list(otherIntention):
if otherIntention[(0, (1, 2))] >= 0.8:
otherShare = True
if selfReward == 1 and selfShare:
selfReward = 0.5
if selfReward == 0 and otherShare:
selfReward = 0.5
selfCost = -0.02 * len(trajectory)
selfFinalGot = selfReward + selfCost
return selfFinalGot
measureFunction = lambda df: lambda trajectory: measureSelfReward(
trajectory)
computeStatistics = ComputeStatistics(loadTrajectoriesFromDf,
measureFunction)
statisticsDf = toSplitFrame.groupby(levelNames).apply(computeStatistics)
print(statisticsDf)
fig = plt.figure()
numColumns = len(manipulatedVariables['inferenceSoft'])
numRows = len(manipulatedVariables['wolfPolicySoft'])
plotCounter = 1
for key, group in statisticsDf.groupby(
['numWolves', 'numSheep', 'wolfPolicySoft', 'inferenceSoft']):
group.index = group.index.droplevel(
['numWolves', 'numSheep', 'wolfPolicySoft', 'inferenceSoft'])
axForDraw = fig.add_subplot(numRows, numColumns, plotCounter)
axForDraw.set_ylabel('Accumulated Reward')
#axForDraw.set_ylabel(str(numWolves))
#if plotCounter <= numColumns:
# axForDraw.set_title(str(key[3]) + 'inferenceSoft')
competeDetectionLabels = [
'Coorperation Only', 'Coorperation Or Competetion Detection',
'Competetion Only'
]
for competeDetection, grp in group.groupby('competeDetectionRate'):
grp.index = grp.index.droplevel('competeDetectionRate')
grp.plot.line(ax=axForDraw,
y='mean',
yerr='se',
label=competeDetectionLabels[int(competeDetection *
2)],
ylim=(-0.3, 0.4),
marker='o',
rot=0)
axForDraw.xaxis.set_label_text('Other Wolf Competing Rate')
plotCounter = plotCounter + 1
#plt.suptitle(str(competePolicy))
plt.show()
def main():
# manipulated variables
manipulatedVariables = OrderedDict()
manipulatedVariables['numWolves'] = [2]
manipulatedVariables['numSheep'] = [2, 4, 8]#, 4, 8]
manipulatedVariables['hierarchy'] = [0]
manipulatedVariables['valuePriorSoftMaxBeta'] = [0.0, 1.0, 2.0, 3.0, 4.0]
manipulatedVariables['valuePriorEndTime'] = [0]#[-100, 0, 100]
levelNames = list(manipulatedVariables.keys())
levelValues = list(manipulatedVariables.values())
modelIndex = pd.MultiIndex.from_product(levelValues, names=levelNames)
toSplitFrame = pd.DataFrame(index=modelIndex)
productedValues = it.product(*[[(key, value) for value in values] for key, values in manipulatedVariables.items()])
parametersAllCondtion = [dict(list(specificValueParameter)) for specificValueParameter in productedValues]
DIRNAME = os.path.dirname(__file__)
trajectoryDirectory = os.path.join(DIRNAME, '..', '..', 'data', 'evaluateHierarchyPlanning',
'trajectories')
if not os.path.exists(trajectoryDirectory):
os.makedirs(trajectoryDirectory)
NNNumSimulations = 250
maxRunningSteps = 51
wolfPolicySoft = 2.5
sheepPolicySoft = 2.5
trajectoryFixedParameters = {'wolfPolicySoft': wolfPolicySoft, 'sheepPolicySoft': sheepPolicySoft,
'maxRunningSteps': maxRunningSteps, 'NNNumSimulations': NNNumSimulations}
trajectoryExtension = '.pickle'
getTrajectorySavePath = GetSavePath(trajectoryDirectory, trajectoryExtension, trajectoryFixedParameters)
# Compute Statistics on the Trajectories
loadTrajectories = LoadTrajectories(getTrajectorySavePath, loadFromPickle)
loadTrajectoriesFromDf = lambda df: loadTrajectories(readParametersFromDf(df))
maxSteps = maxRunningSteps-26
measureIntentionArcheivement = lambda df: lambda trajectory: int(len(trajectory) < maxSteps) - 1 / maxSteps * min(len(trajectory), maxSteps)
computeStatistics = ComputeStatistics(loadTrajectoriesFromDf, measureIntentionArcheivement)
statisticsDf = toSplitFrame.groupby(levelNames).apply(computeStatistics)
print(statisticsDf)
fig = plt.figure()
numColumns = len(manipulatedVariables['numWolves'])
numRows = len(manipulatedVariables['hierarchy'])
plotCounter = 1
for key, group in statisticsDf.groupby(['valuePriorEndTime', 'hierarchy', 'numWolves']):
group.index = group.index.droplevel(['valuePriorEndTime', 'hierarchy', 'numWolves'])
axForDraw = fig.add_subplot(numRows, numColumns, plotCounter)
axForDraw.set_ylabel('Accumulated Reward')
#axForDraw.set_ylabel(str(numWolves))
#if plotCounter <= numColumns:
# axForDraw.set_title(str(key[1]) + 'Wolves')
#valuePriorLabels = ['No Value Prior', 'Value Prior']# First Time Step', 'Value Prior Every Time Step ']
for beta, grp in group.groupby('valuePriorSoftMaxBeta'):
grp.index = grp.index.droplevel('valuePriorSoftMaxBeta')
grp.plot.line(ax = axForDraw, y = 'mean', yerr = 'se', label = 'Value SofMax Beta = '+str(beta/2), ylim = (0.2, 0.7), marker = 'o', rot = 0 )
axForDraw.xaxis.set_label_text('Number of Sheep')
plotCounter = plotCounter + 1
#plt.suptitle('3 Wolves')
#fig.text(x = 0.5, y = 0.04, s = 'numWolves', ha = 'center', va = 'center')
#fig.text(x = 0.05, y = 0.5, s = 'valuePriorEndTime', ha = 'center', va = 'center', rotation=90)
plt.show()
def main():
# manipulated variables
manipulatedVariables = OrderedDict()
manipulatedVariables['numSheep'] = [1, 2, 4]
manipulatedVariables['wolfType'] = [
'individualReward', 'sharedReward',
'sharedAgencyByIndividualRewardWolf'
]
manipulatedVariables['sheepConcern'] = ['selfSheep']
levelNames = list(manipulatedVariables.keys())
levelValues = list(manipulatedVariables.values())
modelIndex = pd.MultiIndex.from_product(levelValues, names=levelNames)
toSplitFrame = pd.DataFrame(index=modelIndex)
DIRNAME = os.path.dirname(__file__)
trajectoryDirectory = os.path.join(DIRNAME, '..', '..', 'data',
'evaluateHierarchyPlanningEnvMADDPG',
'trajectories')
if not os.path.exists(trajectoryDirectory):
os.makedirs(trajectoryDirectory)
maxRunningSteps = 101
numWolves = 3
trajectoryFixedParameters = {
'numWolves': numWolves,
'maxRunningStepsToSample': maxRunningSteps
}
trajectoryExtension = '.pickle'
getTrajectorySavePath = GetSavePath(trajectoryDirectory,
trajectoryExtension,
trajectoryFixedParameters)
# Compute Statistics on the Trajectories
loadTrajectories = LoadTrajectories(getTrajectorySavePath, loadFromPickle)
loadTrajectoriesFromDf = lambda df: loadTrajectories(
readParametersFromDf(df))
measureIntentionArcheivement = lambda df: lambda trajectory: np.sum(
np.array(trajectory)[:, 3])
computeStatistics = ComputeStatistics(loadTrajectoriesFromDf,
measureIntentionArcheivement)
statisticsDf = toSplitFrame.groupby(levelNames).apply(computeStatistics)
print(statisticsDf)
fig = plt.figure()
fig.set_dpi(120)
numColumns = len(manipulatedVariables['sheepConcern'])
numRows = 1
plotCounter = 1
wolfTypeTable = {
'individualReward': 'Individual Reward',
'sharedReward': 'Shared Reward',
'sharedAgencyBySharedRewardWolf':
'Shared Agency By Shared Reward Wolf',
'sharedAgencyByIndividualRewardWolf': 'Shared Agency'
}
for key, group in statisticsDf.groupby(['sheepConcern']):
group.index = group.index.droplevel(['sheepConcern'])
axForDraw = fig.add_subplot(numRows, numColumns, plotCounter)
for wolfType, grp in group.groupby('wolfType'):
grp.index = grp.index.droplevel('wolfType')
grp.plot.line(ax=axForDraw,
y='mean',
yerr='se',
label=wolfTypeTable[wolfType],
ylim=(0, 18),
marker='o',
rot=0)
axForDraw.set_xlabel('Number of Biting Sheep')
if plotCounter % numColumns == 1:
axForDraw.set_ylabel('Number of Wolves = ' + str(key))
plt.xticks(manipulatedVariables['numSheep'])
plotCounter = plotCounter + 1
fig.text(x=0.03,
y=0.5,
s='Mean Episode Bite',
ha='center',
va='center',
rotation=90)
plt.suptitle('Compare Shared Agency with Individual/Sharing Rewards')
resultPath = os.path.join(DIRNAME, '..', 'evalResult')
if not os.path.exists(resultPath):
os.makedirs(resultPath)
plt.savefig(
os.path.join(
resultPath,
'RewardMADDPG3WolfSharedAgencyVSIndividualRewardVSSharedReward'))
plt.show()
def main():
# manipulated variables
manipulatedVariables = OrderedDict()
manipulatedVariables['numSheep'] = [2, 4]
#manipulatedVariables['rationalityBetaInInference'] = [0.0, 0.125, 0.25, 0.5, 1.0]
manipulatedVariables['rationalityBetaInInference'] = [
1.0
] #[0.0, 0.1, 0.2, 1.0]#[0.0, 0.025, 0.05, 0.1, 1.0]
manipulatedVariables['deviationFor2DAction'] = [3.0] #[0.3, 1.0, 3.0, 9.0]
levelNames = list(manipulatedVariables.keys())
levelValues = list(manipulatedVariables.values())
modelIndex = pd.MultiIndex.from_product(levelValues, names=levelNames)
toSplitFrame = pd.DataFrame(index=modelIndex)
productedValues = it.product(
*[[(key, value) for value in values]
for key, values in manipulatedVariables.items()])
parametersAllCondtion = [
dict(list(specificValueParameter))
for specificValueParameter in productedValues
]
DIRNAME = os.path.dirname(__file__)
trajectoryDirectory = os.path.join(DIRNAME, '..', '..', 'data',
'evaluateHierarchyPlanningEnvMADDPG',
'trajectories')
if not os.path.exists(trajectoryDirectory):
os.makedirs(trajectoryDirectory)
numWolves = 3
valuePriorSoftMaxBeta = 0.0
maxRunningSteps = 102
valuePriorEndTime = -100
trajectoryFixedParameters = {
'numWolves': numWolves,
'maxRunningSteps': maxRunningSteps,
'valuePriorEndTime': valuePriorEndTime,
'valuePriorSoftMaxBeta': valuePriorSoftMaxBeta
}
trajectoryExtension = '.pickle'
getTrajectorySavePath = GetSavePath(trajectoryDirectory,
trajectoryExtension,
trajectoryFixedParameters)
# Compute Statistics on the Trajectories
loadTrajectories = LoadTrajectories(getTrajectorySavePath, loadFromPickle)
#loadTrajectoriesFromDf = lambda df: [trajectory for trajectory in loadTrajectories(readParametersFromDf(df)) if len(trajectory) < readParametersFromDf(df)['maxRunningSteps']]
loadTrajectoriesFromDf = lambda df: loadTrajectories(
readParametersFromDf(df))
priorIndexinTimestep = 4
composeInterpolateFunction = lambda df: Interpolate1dData(maxRunningSteps)
#getImaginedWeIdsForCrossEntropy = lambda df: [readParametersFromDf(df)['numSheep'], list(range(readParametersFromDf(df)['numSheep'] + 1, readParametersFromDf(df)['numSheep'] +
# readParametersFromDf(df)['numWolves']))]
#composeMeasureCrossEntropy = lambda df: MeasureCrossEntropy(getImaginedWeIdsForCrossEntropy(df), priorIndexinTimestep)
#measureFunction = lambda df: lambda trajectory: composeInterpolateFunction(df)(composeMeasureCrossEntropy(df)(trajectory))
getImaginedWeIdsForConvergeRate = lambda df: list(
range(
readParametersFromDf(df)['numSheep'],
readParametersFromDf(df)['numSheep'] + numWolves))
measureConvergeRate = lambda df: MeasureConvergeOnGoal(
getImaginedWeIdsForConvergeRate(df), priorIndexinTimestep,
sampleFromDistribution)
measureFunction = lambda df: lambda trajectory: composeInterpolateFunction(
df)(measureConvergeRate(df)(trajectory))
computeStatistics = ComputeStatistics(loadTrajectoriesFromDf,
measureFunction)
statisticsDf = toSplitFrame.groupby(levelNames).apply(computeStatistics)
fig = plt.figure()
fig.set_dpi(120)
numColumns = len(manipulatedVariables['rationalityBetaInInference'])
numRows = len(manipulatedVariables['deviationFor2DAction'])
plotCounter = 1
for key, group in statisticsDf.groupby(
['deviationFor2DAction', 'rationalityBetaInInference']):
group.index = group.index.droplevel(
['deviationFor2DAction', 'rationalityBetaInInference'])
axForDraw = fig.add_subplot(numRows, numColumns, plotCounter)
#if (plotCounter) % max(numColumns, 2) == 1:
# axForDraw.set_ylabel('Observe Noise = '+str(key[0]))
#if plotCounter <= numColumns:
# axForDraw.set_title('Rationality Beta = '+str(key[1]))
for numSheep, grp in group.groupby('numSheep'):
df = pd.DataFrame(grp.values[0].tolist(),
columns=list(range(maxRunningSteps)),
index=['mean', 'se'
])[list(range(int(maxRunningSteps - 3)))]
dfTransPosed = df.T
#df.plot.line(ax = axForDraw, label = 'Set Size of Intentions = {}'.format(numSheep), y = 'mean', yerr = 'se', ylim = (0, 3), rot = 0)
dfTransPosed.plot.line(ax=axForDraw,
label='numSheep = {}'.format(numSheep),
y='mean',
yerr='se',
ylim=(0, 1.01),
rot=0)
plotCounter = plotCounter + 1
plt.suptitle('Shared Agency ' + str(numWolves) + 'Wolves Convege Rate')
#plt.legend(loc='best')
#fig.text(x = 0.5, y = 0.04, s = 'Rationality In Inference', ha = 'center', va = 'center')
#fig.text(x = 0.05, y = 0.5, s = 'valuePriorEndTime', ha = 'center', va = 'center', rotation=90)
plt.show()
def main():
# manipulated variables
# manipulated variables
manipulatedVariables = OrderedDict()
manipulatedVariables['numWolves'] = [3]
manipulatedVariables['numSheep'] = [1, 2, 4]
manipulatedVariables['deviationFor2DAction'] = [1.0, 3.0, 9.0]
#manipulatedVariables['rationalityBetaInInference'] = [0.0, 0.125, 0.25, 0.5, 1.0]
manipulatedVariables['rationalityBetaInInference'] = [0.0, 0.1, 0.2, 1.0]
levelNames = list(manipulatedVariables.keys())
levelValues = list(manipulatedVariables.values())
modelIndex = pd.MultiIndex.from_product(levelValues, names=levelNames)
toSplitFrame = pd.DataFrame(index=modelIndex)
productedValues = it.product(*[[(key, value) for value in values] for key, values in manipulatedVariables.items()])
parametersAllCondtion = [dict(list(specificValueParameter)) for specificValueParameter in productedValues]
DIRNAME = os.path.dirname(__file__)
trajectoryDirectory = os.path.join(DIRNAME, '..', '..', 'data', 'evaluateHierarchyPlanningEnvMADDPG',
'trajectories')
if not os.path.exists(trajectoryDirectory):
os.makedirs(trajectoryDirectory)
valuePriorSoftMaxBeta = 0.0
maxRunningSteps = 102
valuePriorEndTime = -100
trajectoryFixedParameters = {'maxRunningSteps': maxRunningSteps,'valuePriorSoftMaxBeta': valuePriorSoftMaxBeta,
'valuePriorEndTime': valuePriorEndTime}
trajectoryExtension = '.pickle'
getTrajectorySavePath = GetSavePath(trajectoryDirectory, trajectoryExtension, trajectoryFixedParameters)
# Compute Statistics on the Trajectories
loadTrajectories = LoadTrajectories(getTrajectorySavePath, loadFromPickle)
#loadTrajectoriesFromDf = lambda df: [trajectory for trajectory in loadTrajectories(readParametersFromDf(df)) if len(trajectory) < readParametersFromDf(df)['maxRunningSteps']]
loadTrajectoriesFromDf = lambda df: loadTrajectories(readParametersFromDf(df))
measureIntentionArcheivement = lambda df: lambda trajectory: np.sum(np.array(trajectory)[:, 3])
computeStatistics = ComputeStatistics(loadTrajectoriesFromDf, measureIntentionArcheivement)
statisticsDf = toSplitFrame.groupby(levelNames).apply(computeStatistics)
print(statisticsDf)
__import__('ipdb').set_trace()
fig = plt.figure()
fig.set_dpi(120)
numColumns = len(manipulatedVariables['rationalityBetaInInference'])
numRows = len(manipulatedVariables['deviationFor2DAction'])
plotCounter = 1
for key, group in statisticsDf.groupby(['deviationFor2DAction', 'rationalityBetaInInference']):
group.index = group.index.droplevel(['deviationFor2DAction', 'rationalityBetaInInference'])
axForDraw = fig.add_subplot(numRows, numColumns, plotCounter)
if (plotCounter) % max(numColumns, 2) == 1:
axForDraw.set_ylabel('Observe Noise = '+str(key[0]))
if plotCounter <= numColumns:
axForDraw.set_title('Rationality Beta = '+str(key[1]))
numWolvesLabels = ['2', '3']
for numWolves, grp in group.groupby('numWolves'):
grp.index = grp.index.droplevel('numWolves')
grp.plot.line(ax = axForDraw, y = 'mean', yerr = 'se', label = str(numWolves)+' Wolves', ylim = (0, 16), marker = 'o', rot = 0 )
if int((plotCounter - 1) / numColumns) == numRows - 1:
axForDraw.xaxis.set_label_text('Number of Sheep')
else:
xAxis = axForDraw.get_xaxis()
xLabel = xAxis.get_label()
xLabel.set_visible(False)
plotCounter = plotCounter + 1
plt.suptitle('Accumulated Reward')
#fig.text(x = 0.5, y = 0.04, s = 'numWolves', ha = 'center', va = 'center')
#fig.text(x = 0.05, y = 0.5, s = 'valuePriorEndTime', ha = 'center', va = 'center', rotation=90)
plt.show()
