Merge remote-tracking branch 'origin/vis' into develop

# Conflicts:
#	.gitignore
#	DAS/simulator.py
#	study.py
This commit is contained in:
Csaba Kiraly 2023-03-02 01:32:25 +01:00
commit 35d1790429
No known key found for this signature in database
GPG Key ID: 0FE274EE8C95166E
5 changed files with 160 additions and 14 deletions

2
.gitignore vendored
View File

@ -3,3 +3,5 @@
results/*
myenv
doc/_build
!results/plots.py
Frontend/

View File

@ -1,3 +1,4 @@
from DAS.simulator import *
from DAS.configuration import *
from DAS.shape import *
from DAS.visualizer import *

View File

@ -42,7 +42,6 @@ class Simulator:
def initNetwork(self):
"""It initializes the simulated network."""
self.shape.netDegree = 6
rowChannels = [[] for i in range(self.shape.blockSize)]
columnChannels = [[] for i in range(self.shape.blockSize)]
for v in self.validators:

138
DAS/visualizer.py Normal file
View File

@ -0,0 +1,138 @@
#!/bin/python3
import os, sys
import time
import xml.etree.ElementTree as ET
import matplotlib.pyplot as plt
import numpy as np
import seaborn as sns
from itertools import combinations
class Visualizer:
def __init__(self, execID):
self.execID = execID
self.folderPath = "results/"+self.execID
self.parameters = ['run', 'blockSize', 'failureRate', 'numberValidators', 'netDegree', 'chi']
self.minimumDataPoints = 2
def plottingData(self):
#Store data with a unique key for each params combination
data = {}
#Loop over the xml files in the folder
for filename in os.listdir(self.folderPath):
#Loop over the xmls and store the data in variables
if filename.endswith('.xml'):
tree = ET.parse(os.path.join(self.folderPath, filename))
root = tree.getroot()
run = int(root.find('run').text)
blockSize = int(root.find('blockSize').text)
failureRate = int(root.find('failureRate').text)
numberValidators = int(root.find('numberValidators').text)
netDegree = int(root.find('netDegree').text)
chi = int(root.find('chi').text)
tta = int(root.find('tta').text)
# Loop over all possible combinations of length 4 of the parameters
for combination in combinations(self.parameters, 4):
# Get the indices and values of the parameters in the combination
indices = [self.parameters.index(element) for element in combination]
selectedValues = [run, blockSize, failureRate, numberValidators, netDegree, chi]
values = [selectedValues[index] for index in indices]
names = [self.parameters[i] for i in indices]
keyComponents = [f"{name}_{value}" for name, value in zip(names, values)]
key = tuple(keyComponents[:4])
#Get the names of the other 2 parameters that are not included in the key
otherParams = [self.parameters[i] for i in range(6) if i not in indices]
#Append the values of the other 2 parameters and the ttas to the lists for the key
otherIndices = [i for i in range(len(self.parameters)) if i not in indices]
#Initialize the dictionary for the key if it doesn't exist yet
if key not in data:
data[key] = {}
#Initialize lists for the other 2 parameters and the ttas with the key
data[key][otherParams[0]] = []
data[key][otherParams[1]] = []
data[key]['ttas'] = []
if otherParams[0] in data[key]:
data[key][otherParams[0]].append(selectedValues[otherIndices[0]])
else:
data[key][otherParams[0]] = [selectedValues[otherIndices[0]]]
if otherParams[1] in data[key]:
data[key][otherParams[1]].append(selectedValues[otherIndices[1]])
else:
data[key][otherParams[1]] = [selectedValues[otherIndices[1]]]
data[key]['ttas'].append(tta)
print("Getting data from the folder...")
return data
def similarKeys(self, data):
#Get the keys for all data with the same x and y labels
filteredKeys = {}
for key1, value1 in data.items():
subKeys1 = list(value1.keys())
filteredKeys[(subKeys1[0], subKeys1[1])] = [key1]
for key2, value2 in data.items():
subKeys2 = list(value2.keys())
if key1 != key2 and subKeys1[0] == subKeys2[0] and subKeys1[1] == subKeys2[1]:
try:
filteredKeys[(subKeys1[0], subKeys1[1])].append(key2)
except KeyError:
filteredKeys[(subKeys1[0], subKeys1[1])] = [key2]
print("Getting filtered keys from data...")
return filteredKeys
def formatLabel(self, label):
#Label formatting for the figures
result = ''.join([f" {char}" if char.isupper() else char for char in label])
return result.title()
def formatTitle(self, key):
#Title formatting for the figures
name = ''.join([f" {char}" if char.isupper() else char for char in key.split('_')[0]])
number = key.split('_')[1]
return f"{name.title()}: {number} "
def plotHeatmaps(self):
#Plot and store the 2D heatmaps in subfolders
data = self.plottingData()
filteredKeys = self.similarKeys(data)
print("Plotting heatmaps...")
#Create the directory if it doesn't exist already
heatmapsFolder = self.folderPath + '/heatmaps'
if not os.path.exists(heatmapsFolder):
os.makedirs(heatmapsFolder)
#Plot
for labels, keys in filteredKeys.items():
for key in keys:
xlabels = np.sort(np.unique(data[key][labels[0]]))
ylabels = np.sort(np.unique(data[key][labels[1]]))
if len(xlabels) < self.minimumDataPoints or len(ylabels) < self.minimumDataPoints:
continue
hist, xedges, yedges = np.histogram2d(data[key][labels[0]], data[key][labels[1]], bins=(len(xlabels), len(ylabels)), weights=data[key]['ttas'])
hist = hist.T
fig, ax = plt.subplots(figsize=(10, 6))
sns.heatmap(hist, xticklabels=xlabels, yticklabels=ylabels, cmap='Purples', cbar_kws={'label': 'Time to block availability'}, linecolor='black', linewidths=0.3, annot=True, fmt=".2f", ax=ax)
plt.xlabel(self.formatLabel(labels[0]))
plt.ylabel(self.formatLabel(labels[1]))
filename = ""
title = ""
paramValueCnt = 0
for param in self.parameters:
if param != labels[0] and param != labels[1]:
filename += f"{key[paramValueCnt]}"
formattedTitle = self.formatTitle(key[paramValueCnt])
title += formattedTitle
paramValueCnt += 1
title_obj = plt.title(title)
font_size = 16 * fig.get_size_inches()[0] / 10
title_obj.set_fontsize(font_size)
filename = filename + ".png"
targetFolder = os.path.join(heatmapsFolder, f"{labels[0]}Vs{labels[1]}")
if not os.path.exists(targetFolder):
os.makedirs(targetFolder)
plt.savefig(os.path.join(targetFolder, filename))
plt.close()
plt.clf()

View File

@ -23,23 +23,25 @@ def study():
start = time.time()
for run in range(config.numberRuns):
for fr in range(config.failureRateStart, config.failureRateStop+1, config.failureRateStep):
for chi in range(config.chiStart, config.chiStop+1, config.chiStep):
for blockSize in range(config.blockSizeStart, config.blockSizeStop+1, config.blockSizeStep):
for nv in range(config.nvStart, config.nvStop+1, config.nvStep):
for netDegree in range(config.netDegreeStart, config.netDegreeStop+1, config.netDegreeStep):
for nv in range(config.nvStart, config.nvStop+1, config.nvStep):
for blockSize in range(config.blockSizeStart, config.blockSizeStop+1, config.blockSizeStep):
for fr in range(config.failureRateStart, config.failureRateStop+1, config.failureRateStep):
for netDegree in range(config.netDegreeStart, config.netDegreeStop+1, config.netDegreeStep):
for chi in range(config.chiStart, config.chiStop+1, config.chiStep):
if not config.deterministic:
random.seed(datetime.now())
shape = Shape(blockSize, nv, fr, chi, netDegree, run)
sim.resetShape(shape)
sim.initValidators()
sim.initNetwork()
result = sim.run()
sim.logger.info("Shape: %s ... Block Available: %d in %d steps" % (str(sim.shape.__dict__), result.blockAvailable, len(result.missingVector)), extra=sim.format)
results.append(copy.deepcopy(result))
simCnt += 1
# Network Degree has to be an even number
if netDegree % 2 == 0:
shape = Shape(blockSize, nv, fr, chi, netDegree, run)
sim.resetShape(shape)
sim.initValidators()
sim.initNetwork()
result = sim.run()
sim.logger.info("Shape: %s ... Block Available: %d in %d steps" % (str(sim.shape.__dict__), result.blockAvailable, len(result.missingVector)), extra=sim.format)
results.append(copy.deepcopy(result))
simCnt += 1
end = time.time()
sim.logger.info("A total of %d simulations ran in %d seconds" % (simCnt, end-start), extra=sim.format)
@ -49,6 +51,10 @@ def study():
res.dump(execID)
sim.logger.info("Results dumped into results/%s/" % (execID), extra=sim.format)
visualization = 1
if visualization:
vis = Visualizer(execID)
vis.plotHeatmaps()
study()