def generateDataFramesRangeSaving(StartDate, EndDate, DateStep, Gender='None', SaveEvery=1, File='OBOextract', Path='Data/Raw/', Overwrite = 1): """Generate the dataframes from a gender and datelist""" import os import pandas as pd import urllib3 retry = urllib3.util.Retry(total=100, read=100, connect=100, backoff_factor=1) timeout = urllib3.util.Timeout(connect=4.0, read=8.0) http=urllib3.PoolManager(retry=retry, timeout=timeout, maxsize=5) _Iteration = 0 DummyFrame = pd.DataFrame(columns=c2n.generateCategories()) if Overwrite is 0: # Does the file exist if os.path.isfile(Path + File+'.csv'): try: DummyFrame = pd.DataFrame.from_csv(Path + File+'.csv') _TempStartDate = int(DummyFrame.tail(1).values[0,0] + DateStep) if _TempStartDate >= StartDate: StartDate = _TempStartDate + DateStep except: DummyFrame = pd.DataFrame(columns=c2n.generateCategories()) for _startDate in range(StartDate,EndDate+1,DateStep): _Iteration += 1 _endDate= _startDate+DateStep-1 print('Generating row for {} to {}, Gender: {}'.format(_startDate,_endDate,Gender)) DummyFrame.loc[len(DummyFrame)] = generateRowRange(_startDate, _endDate, Gender) if _Iteration >= SaveEvery: _Iteration = 0 DummyFrame.to_csv(Path + File+'.csv') return DummyFrame
def generateHeadings(OffenceArray=c2n.offcat): """Generate the pairwise comparisons on an array, OffenceArray Returns ------- array of strings """ OffenceArrayLen=len(OffenceArray) HeadingArray=[] for ci,C in enumerate(OffenceArray): for P in range(ci+1,OffenceArrayLen): HeadingArray.append(c2n.upcaseFirstLetter(C) +c2n.upcaseFirstLetter(OffenceArray[P])) return HeadingArray
def generateDummyDataFrame(Delta=5): """ Generate a DataFrame of dummy data to mimic an Emperical Categories frame. We want to produce dummy data to test our model selection. We have two trials in one: Detect the optimal delta, and detect the optimal partitionings. We do this with one data set where we have some small number of slightly different distributions, repeated the number of times for our. We only need distinct distributions for offences, repeated for all 7 punishments Arbitrarily we will choose deltat = 5 and three distinct partitions. One of size one, One of size two, One of size seve """ offenceDistributions = [ [0.05,0.2,0.05,0.05,0.25,0.05,0.05,0.25,0.05], # C,A,C,C,B,C,C,B,C [0.25,0.05,0.05,0.05,0.25,0.05,0.05,0.2,0.05], # B,C,C,C,B,C,C,A,C [0.05,0.25,0.25,0.05,0.05,0.05,0.05,0.05,0.2], # C,B,B,C,C,C,C,C,A [0.05,0.2,0.05,0.05,0.25,0.05,0.05,0.25,0.05] # C,A,C,C,B,C,C,B,C ] probabilities = [] for offDis in offenceDistributions: probabilities.append([x for x in [z/7 for z in offDis] for y in [1,2,3,4,5,6,7]]) dummyEmp = pd.DataFrame(index=list(range(1674,(1674+4*Delta))), columns=c2n.generateCategories()[1:64]) for offenDist in range(4): for delta in range(Delta): dummyEmp.iloc[offenDist*Delta+delta] = discreteRandomSamples(probabilities[offenDist]) return dummyEmp
def generateDataFramesParallel(Date_List, Gender): """Generate the dataframes from a gender and datelist""" import workerpool import json import pandas as pd Gender = 'None' DummyFrame = pd.DataFrame(columns=c2n.generateCategories()) NUM_SOCKETS = 3 NUM_WORKERS = 5 # We want a few more workers than sockets so that they have extra # time to parse things and such. workers = workerpool.WorkerPool(size=NUM_WORKERS) class MyJob(workerpool.Job): def __init__(self, dAte, Gender): self.dAte = dAte self.Gender = Gender def run(self): print('Generating row for {}'.format(self.dAte)) DummyFrame.loc[len(DummyFrame)] = generateRow(self.dAte,self.Gender,NUM_SOCKETS) for daTe in Date_List: workers.put(MyJob(daTe,Gender)) # Send shutdown jobs to all threads, and wait until all the jobs have been completed # (If you don't do this, the script might hang due to a rogue undead thread.) workers.shutdown() workers.wait() return DummyFrame
def deathornot(CatEmp): """Create the raw data for death or not as punishment per offence. Use generateDependentModelLaplace(deathornot) to generate the probability estimates """ #Grouping arrays, statically as it is easier to understand. 63 columns, combine by category for Death and Not: Groupings = [ c2n.upcaseFirstLetter(x)+y for x in c2n.offcat for y in ['Not']*2 + ['Death'] + ['Not']*4] DeathOrNotEmp = CatEmp.groupby(Groupings,axis=1,sort=False).sum() return DeathOrNotEmp
def generateDataFrames(Date_List, Gender): """Generate the dataframes from a gender and datelist""" import pandas as pd DummyFrame = pd.DataFrame(columns=c2n.generateCategories()) for dAte in Date_List: print('Generating row for {}'.format(dAte)) DummyFrame.loc[len(DummyFrame)] = generateRow(dAte,Gender) return DummyFrame
def validatePartitioning(): """ Validate the partition searching. Parameters ---------- DumyEmp : pandas DataFrame, Dummy emperical data frame as generated by generateDummyDataFrame() BestPartition: bool, Whether to return the partition AIC scores or the best partition. Returns: pandas DataFrame, Partition AIC scores, or the best partition Try also: x = [15,85,44,56,49,51,46,54,50,50,50,50,56,44,46,54,4,96] NNNNope """ #Death Or Not test #For partitions of typ: A, B, A, C, A, A, A, C, A, where we should get # Which is [[0,2,4,5,6,8],[1],[3,7]] #A = [ 'breakingPeace','deception', 'miscellaneous', 'royalOffences', 'sexual', 'violentTheft'] #B = [ 'damage' ] #C = [ 'kill', 'theft'] # We also ensure that the occurence of the offences are different within the same partition TestRow = [40, 160, 70, 30, 20, 80, 50, 50, 20, 80, 20, 80, 10, 40, 100, 100, 20, 80] #For not Death or Not, have 7 punishmnets A = [ 10, 20, 30, 40, 50, 60, 70 ] B = [ 70, 60, 50, 40, 30, 20, 10 ] C = [ 20, 10, 40, 30, 60, 50, 70 ] #For 9 offences #A, B, A, C, A, A, A, C, A as above TestFullRowA = A+listMul(B,2)+listMul(A,3)+C+A+A+listMul(A,2)+C+A #C, A, A, B, B, C, A, A, C TestFullRowB = C+A+listMul(A,2)+B+listMul(B,3)+C+A+listMul(A,3)+C TestFrame = pd.DataFrame([TestRow,TestRow], columns=list(range(18)), index=[0,1]) TestFullFrame = pd.DataFrame([TestFullRowA,TestFullRowA,TestFullRowB,TestFullRowB,TestFullRowB], columns=list(range(63)), index=list(range(5))) partitions = partition.Partition([c2n.upcaseFirstLetter(x) for x in c2n.offcat]) print('Testing Death Or Not partitioning') DeathAICtable = oboP.generateAICtable(TestFrame) DeathAICmin = DeathAICtable.idxmin(axis=1).apply(lambda x: partitions[int(x)]) print('Found minimal partitions:') print(DeathAICmin) print('Testing full partitioning') AICtable = oboP.generateAICtable(TestFullFrame) AICmin = AICtable.idxmin(axis=1).apply(lambda x: partitions[int(x)]) print('Found minimal partitions:') print(AICmin)
def hyphenateCategories(DataFrame): """Hyphenate runtogether offence-punishment column labels. Parameters ---------- DataFrame : pandas DataFrame, a Categories frame with 63 columns Returns ------- HyphenateFrame : pandas DataFrame, with 9 Offence Category Columns """ DataFrame.columns = c2n.generateCategoriesHyphenated() return DataFrame
def generateDataFramesRollingRangeSaving(StartDate, EndDate, DateStep, Delta=1000, Gender='None', SaveEvery=1, File='OBOextract', Path='Data/Raw/', Overwrite = 1): """Generate the dataframes from a gender and datelist For dates from StartDate to EndDate stepped through by DateStep, data is extracted for the date to date+Delta-1. The data is saved in a Pandas DataFrame and cumulativly updated, and saved every SaveEvery retrievals to the file File. If Overwrite=1 then the process will attempt to restart from interruption based on the xistingesaeved DataFdramed.dd""" import os import pandas as pd import urllib3 retry = urllib3.util.Retry(total=100, read=100, connect=100, backoff_factor=1) timeout = urllib3.util.Timeout(connect=4.0, read=8.0) http=urllib3.PoolManager(retry=retry, timeout=timeout, maxsize=5) _Iteration = 0 DummyFrame = pd.DataFrame(columns=c2n.generateCategories()) if Overwrite is 0: # Does the file exist if os.path.isfile(Path + File+'.csv'): try: DummyFrame = pd.DataFrame.from_csv(Path + File+'.csv') _TempStartDate = int(DummyFrame.tail(1).values[0,0] + DateStep) if _TempStartDate >= StartDate: StartDate = _TempStartDate + DateStep except: DummyFrame = pd.DataFrame(columns=c2n.generateCategories()) for _startDate in range(StartDate,EndDate+1,DateStep): _Iteration += 1 _endDate= _startDate+Delta-1 print('Generating row for {} to {}, Gender: {}'.format(_startDate,_endDate,Gender)) DummyFrame.loc[len(DummyFrame)] = generateRowRange(_startDate, _endDate, Gender) if _Iteration >= SaveEvery: _Iteration = 0 DummyFrame.to_csv(Path + File+'.csv') return DummyFrame
def generateDataFrameInChunks(Date_List,Gender,ChunkSize=10, File='OBOextract', Path='Data/Raw/', Overwrite = 1, Start=0, Stop=0): """ Generate DataFrames, but save to csv every so often""" import pandas as pd if Stop <= Start: Stop=len(Date_List) DummyFrame = pd.DataFrame(columns=c2n.generateCategories()) for i in range(Start,Stop,ChunkSize): for dAte in Date_List[i:i+ChunkSize-1]: print('Generating row for date {}, and gender {}'.format(dAte,Gender)) DummyFrame.loc[len(DummyFrame)] = generateRow(dAte,Gender) DummyFrame.to_csv(File+'.csv') return DummyFrame
def generateDataFramesRange(StartDate, EndDate, DateStep, Gender='None'): """Generate the dataframes from a gender and datelist""" import pandas as pd import urllib3 retry = urllib3.util.Retry(total=1000, read=200, connect=200, backoff_factor=0.5) timeout = urllib3.util.Timeout(connect=2.0, read=4.0) http=urllib3.PoolManager(retry=retry, timeout=timeout, maxsize=10) DummyFrame = pd.DataFrame(columns=c2n.generateCategories()) for _startDate in range(StartDate,EndDate+1,DateStep): _endDate= _startDate+DateStep-1 print('Generating row for {} to {}:'.format(_startDate,_endDate)) DummyFrame.loc[len(DummyFrame)] = generateRowRange(_startDate, _endDate, Gender) return DummyFrame
def initialiseEmptyArray(): """Initialise an empty array the same length as the Categories""" return [0]*len(c2n.generateCategories())
def initialiseDataFrame(): """Create an empty dataframe with nothing but categories""" import pandas as pd return pd.DataFrame(columns=c2n.generateCategories())
def generateRowRange(StartDate, EndDate, Gender='None', http='None',Sockets=2): """Generate one row of data for a given date and defendent gender We are not atomising the code to generalise URL gets such that the socket opened by urllib3 can stay open. This may be messy looking... """ import CategoryToNumberAssignment as c2n import urllib3 import json if http is 'None': http=urllib3.PoolManager(maxsize=Sockets) sStartDate = str(StartDate) sEndDate = str(EndDate) _Columns = c2n.generateCategories() #Generate empty columns Row = [sStartDate] """Get not breakdown of punishments by category and subcategory for the period""" # Categories for Category in c2n.offcat: _TempCategories = initialiseEmptyCatArray() #Get the Json data #print("Generating URL: {}".format(generateURLCategoryRange(sStartDate,sEndDate,Category, Gender))) _Json = URLtoJSON(generateURLCategoryRange(sStartDate,sEndDate,Category, Gender),http) #Find the punishment totals and place in the correct position in the array # adding 1 place for the Not guilty for Totals in _Json['breakdown']: _TempCategories[c2n.puncat.index(Totals['term'])+1]=Totals['total'] # Append _Temps to the Rows Row = Row + _TempCategories #Associate Offence Subcategories with Punishment Subategories for Category in c2n.offsubcat: _TempCategories = initialiseEmptySubCatArray() #Get the Json data _Json = URLtoJSON(generateURLSubCategoryRange(sStartDate,sEndDate,Category, Gender),http) #Find the punishment totals and place in the correct position in the array # adding 1 place for the Not guilty for Totals in _Json['breakdown']: _TempCategories[c2n.punsubcat.index(Totals['term'])+1]=Totals['total'] # Append _Temps to the Rows Row = Row + _TempCategories """Get not guilties by category and subcategory for the period""" #Get not guilties: _JsonNotGuiltyCat = URLtoJSON(generateURLCategoryNotGuiltyRange(sStartDate,sEndDate,Gender),http) _JsonNotGuiltySubCat = URLtoJSON(generateURLSubCategoryNotGuiltyRange(sStartDate,sEndDate,Gender),http) # Place values associated with locations in Row: for Totals in _JsonNotGuiltyCat['breakdown']: Row[_Columns.index(c2n.upcaseFirstLetter(Totals['term'])+'NotGuilty')] = Totals['total'] for Totals in _JsonNotGuiltySubCat['breakdown']: Row[_Columns.index(c2n.upcaseFirstLetter(Totals['term'])+'NotGuilty')] = Totals['total'] return Row
import pandas as pd import numpy as np from ast import literal_eval from partitionsets import partition import CategoryToNumberAssignment as c2n import OBOModelling as oboM #The following dependency is from CythonGSL from https://github.com/twiecki/CythonGSL # The interfaces must be installed as: # sudo python3 setup_interface.py install # in the CythonGSL directory. (you will need gcc and libgsl-dev or equivalent installed) import probability_distribution as gslPDD Deltas = [1,2,3,4,5,10,50,100,240] #Initilise partitions: partitions = partition.Partition([c2n.upcaseFirstLetter(x) for x in c2n.offcat]) partitioN = partition.Partition(list(range(0,9))) def partitionAIC(EmpFrame, part, OffenceEstimateFrame = [], ReturnDeathEstimate=False, BlockPunishment='Death', Verbose=True): """Calculate AIC score between the EmpFrame and the model where offences are partitioned as `part'. Parameters: ----------- EmpFrame : DataFrame DataFrame of emperical data, pre processed, maybe. part : nested list, 2 levels Partition formatted as: [[0, 3], [1, 2, 6, 7], [4, 5, 8]] ReturnDeathEstimate : bool Whether to return the DeathEstimate frame Returns: