def __init__(self, dataContext, name):
clusteringResultsPath = getClusteringResultsPath(
) + getTimeSeriesToyDatasetName() + '/'
self.name = name
super().__init__(dataContext, clusteringResultsPath)
self.ownResourcesFolder = self.clusteringResultsPath + self.name + '/'
self.microClustersFolder = self.ownResourcesFolder
# obtain the data sets from the csv files
non_time_series_datasets = getDatasetsFromFolder(
getNonTimeSeriesDatasetsPath())
# clustering algorithms
kmeans = KMeans()
dbscan = DBSCAN(eps=0.2, min_samples=10)
algorithms = [(getmKMeansName(), kmeans), (getDbscanName(), dbscan)]
# iterate over the data sets
for datIndx in range(len(non_time_series_datasets)): # row index
X = non_time_series_datasets[datIndx]['dataset']
dName = non_time_series_datasets[datIndx]['name']
k = non_time_series_datasets[datIndx]['k']
baseFolder = getClusteringResultsPath() + dName + '/'
# normalize dataset for easier parameter selection
X = StandardScaler().fit_transform(X)
# iterate over the algorithms
for algIndx in range(len(algorithms)): # column index
algo = algorithms[algIndx][1]
algoName = algorithms[algIndx][0]
# if the algorithm is 'kmeans', k must be set specifically for the data set
if algoName.lower() == 'kmeans':
algo.set_params(n_clusters=k)
# assign labels
labels = algo.fit_predict(X)
print(" ", len(set(labels)), " clusters")
clusteringInfo(labels)
# result contains every scaled element with the corresponding label
result = np.c_[X, labels]
dataset = dataset[dataset['to_timestamp'].apply(validate_timestamp, date = datetime.datetime(2018, 10, 2, 10, 30))]
# data context
dataContext = [ # hardcodeamos los límites de sweden
BoundingBox(minimun=11.0273686052 , maximun=23.9033785336),
BoundingBox(minimun=55.3617373725, maximun=69.1062472602),
]
# dyclee
dyclee = Dyclee(dataContext = dataContext, relativeSize = params["relativeSize"], speed = params["speed"], lambd = params["lambd"],
periodicRemovalAt = params["periodicRemovalAt"], periodicUpdateAt = params["periodicUpdateAt"],
timeWindow = params["timeWindow"], closenessThreshold = params["closenessThreshold"])
tGlobal = 6000 # siguiendo lo del notebook, estoy tomando casi cada media hs
ac = 0 # represents amount of processed elements
folder = getClusteringResultsPath() + getRealDatasetName() + '/' + getDycleeName() + '/' # for saving the prepared resuls
if will_calculate_DBCV:
resetStorage(folder)
first = True # first clustering; 'end' picture will be added at the beggining (to understand the loop)
for ind, p in dataset.iterrows():
if end(p['to_timestamp']): # break; para no seguir, solo quiero ver los insights de las 11 y 16 del dia 2
print(p, ' ... (the end)')
break
if not validate_p(p): # me saco de encima los que estan por fuera de sweden, y no corresponden al dia que quiero analizar
continue
ac += 1
point = [p.latitude, p.longitude] # ahora esta completo el dataset, hay que subsetear
import json
import numpy as np
import os
from config import getClusteringResultsPath, getMbkName, getTimeSeriesDatasetsPath, getTimeSeriesToyDatasetName
import shutil
folder = getClusteringResultsPath() + getTimeSeriesToyDatasetName() + '/' + getMbkName() + '/'
def resetStorage():
if os.path.exists(folder):
shutil.rmtree(folder)
def createDirectoryIfNotExists(folder):
# check if resourcesFolder needs to be created
if not os.path.exists(folder):
os.makedirs(folder)
def storeResult(snapshot):
createDirectoryIfNotExists(folder)
time = snapshot.get("time")
result = snapshot.get("result")
targetFile = folder + str(time) + '.csv'
np.savetxt(targetFile, result, delimiter=',',)
def storeAlgoConfig(dict):
createDirectoryIfNotExists(folder)
def __init__(self, dataContext):
clusteringResultsPath = getClusteringResultsPath()
print('clusteringResultsPath', clusteringResultsPath)
super().__init__(dataContext, clusteringResultsPath)