forked from HelenaBlackmore/DISCUS_Hack_PoliticalSpending
-
Notifications
You must be signed in to change notification settings - Fork 0
/
match_invoices.py
229 lines (179 loc) · 8.84 KB
/
match_invoices.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
import pandas as pd
import numpy as np
def get_nth_key(dictionary, n=0):
if n < 0:
n += len(dictionary)
for i, key in enumerate(dictionary.keys()):
if i == n:
return key
raise IndexError("dictionary index out of range")
def get_totals(invoice_amounts,total_values,prediction):
totals = np.zeros_like(total_values)
for amount, cat in zip(invoice_amounts,prediction):
totals[cat] += amount
return totals
def get_error(invoice_amounts,total_values,prediction,loss='MSE',return_totals=False,return_diff=False):
diff = total_values - get_totals(invoice_amounts,total_values,prediction)
if loss == 'MSE':
error = np.square(diff)
else:
error = np.abs(diff)
r = [np.mean(error)]
if return_totals: r+= [totals]
if return_diff: r+= [diff]
return r
def random_solver(invoice_amounts,total_values,loss='MSE',max_iter=5,max_waiting_iter=1):
factor = invoice_amounts.shape[0] ** (len(total_values)-1)
max_iter = int(max_iter * factor)
max_waiting_iter = int(max_waiting_iter * factor)
pred = np.random.randint(low=0, high=len(total_values), size=invoice_amounts.shape[0], dtype=int)
error, diff = get_error(invoice_amounts,total_values,pred,loss,return_diff=True)
wait = 0
unvisited = list(range(len(pred)))
while len(unvisited) > 0:
idx = unvisited.pop(0)
cat = pred[idx]
#print(idx,unvisited)
error_, diff_, cat_ = error, diff, cat
for c in range(len(total_values)):
if c==cat: continue
pred[idx] = c
e, d = get_error(invoice_amounts,total_values,pred,loss,return_diff=True)
#print(f"{c=} {e=}")
if e<error_:
error_, diff_, cat_ = e, d, c
if error_ < error:
unvisited = list(range(len(pred)))
np.random.shuffle(unvisited)
#print(f"Improvement! {error_}")
pred[idx] = cat_
error, diff = error_, diff_
return pred
def genetic_algorithm(invoice_amounts,total_values,loss='MSE',N=100,max_iter=50,mutationRate=None,min_error=0.01):
def calcFit(prediction):
return get_error(invoice_amounts,total_values,prediction,loss=loss)[0]
No_Gen = invoice_amounts.shape[0]
No_Typ = len(total_values)
if mutationRate is None:
mutationRate = 1/No_Gen
#factor = No_Gen ** (No_Typ-1)
factor = 1
total_iter = int(max_iter * factor * N)
indeces = np.array(range(N))
pop = np.random.randint(low=0, high=No_Typ, size=(N,No_Gen), dtype=int)
fit = np.array([calcFit(pred) for pred in pop])
minFit_idx = np.argmin(fit)
minFit = fit[minFit_idx]
#print(f"Minimum: {minFit:.2f} (Idx = {minFit_idx}); solution: {pop[minFit_idx]}")
if minFit > min_error:
for it in range(total_iter*N):
champs = np.random.choice(indeces, size=2, replace=False)
# determine winner
if fit[champs[1]] < fit[champs[0]]:
champs[0],champs[1] = champs[1],champs[0]
# uniform crossover
for j in range(No_Gen):
# Mutation chance
if np.random.random() < mutationRate:
pop[champs[1],j] = np.random.randint(low=0, high=No_Typ, size=1, dtype=int)
# 50% chance of overwritting
elif np.random.random() < 0.5:
pop[champs[1],j] = pop[champs[0],j]
# determine fitness of newly combined individual
fit[champs[1]] = calcFit(pop[champs[1]])
if fit[champs[1]] < minFit:
minFit, minFit_idx = fit[champs[1]], champs[1]
#print(f"It {it} Minimum: {minFit:.2f} (Idx = {minFit_idx}); solution: {pop[minFit_idx]}")
if minFit < min_error:
break
# Print every N iterations
if (it+1)%N == 0:
print(f"{(it+1)//N}/{total_iter}: Fitness: {minFit:.2f} Solution: {pop[minFit_idx]}")
return pop[minFit_idx]
def run(solver='GA',N=100,verbose=False):
data_coding = pd.read_csv('data/processed/coding_both.csv')
data_invoice = pd.read_csv('data/processed/invoices.csv')
## TODO add CSV with manual set invoice amounts
#data_preset = pd.read_csv('data/input/invoices_manual.csv')
parties = pd.concat((data_coding['Party'], data_invoice['RegulatedEntityName'])).unique()
categories = data_coding.columns[5:]
error_acc = 0
row_list = []
for party in parties:
party_coding = data_coding[ data_coding['Party']==party ]
party_invoice = data_invoice[ data_invoice['RegulatedEntityName']==party ]
suppliers = pd.concat(( party_coding['Supplier'], party_invoice['SupplierName'] )).unique()
for supplier in suppliers:
sup_coding = party_coding[ party_coding['Supplier']==supplier ]
sup_invoice = party_invoice[ party_invoice['SupplierName']==supplier ]
sup_cats = {}
for cat in categories:
val = np.sum(sup_coding[cat].values)
if val >= 0.01:
sup_cats[cat] = val
if len(sup_cats) < 1:
continue
if verbose:
spent = np.round(sup_coding['Total Spend'].sum(),2)
filed = np.round(sup_invoice['TotalExpenditure'].sum(),2)
print(f"{party} - {supplier}")
print(f"Diff = {abs(spent-filed)} ({spent=} | {filed=}) accross {len(sup_cats)} categories with {sup_coding.shape[0]} codings and {sup_invoice.shape[0]} invoices.")
# Estimate a proper matching if there's more than one invoice
# random encoding which invoice belongs to which category
if sup_invoice.shape[0] > 1:
values = np.fromiter(sup_cats.values(), dtype=float)
invoices = sup_invoice['TotalExpenditure'].values
if len(values) <= 1:
prediction = np.zeros(invoices.shape[0],dtype=int)
else:
if solver == 'GA':
prediction = genetic_algorithm(invoices,values,loss='MSE',N=N,max_iter=len(values)**3,mutationRate=0.02)
else:
# Absolute error is fine for random solver as only one position is changed at the time
prediction = random_solver(invoices,values,loss='MAE')
# get non-squared error
error = get_error(invoices,values,prediction,loss='MAE')[0]
error_acc += error
for i, (idx, row) in enumerate(sup_invoice.iterrows()):
val = row['TotalExpenditure']
entry = {
'ECRef': row['ECRef'],
'InvoiceID': row['RedactedSupportingInvoiceId'],
'Supplier': supplier,
'Party': party,
'Total Spend': val,
'Error': error,
'Expense Category (Coding)': sup_coding['Expense Category'].mode()[0],
'Expense Category (Invoice)': row['ExpenseCategoryName'],
**{k:0 for k in categories}
}
cat = get_nth_key(sup_cats, n=prediction[i])
entry[cat]=val
row_list.append(entry)
else:
row = sup_invoice.iloc[0]
entry = {
'ECRef': row['ECRef'],
'InvoiceID': row['RedactedSupportingInvoiceId'],
'Supplier': supplier,
'Party': party,
'Total Spend': row['TotalExpenditure'],
'Error': 0.0,
'Expense Category (Coding)': sup_coding['Expense Category'].mode()[0],
'Expense Category (Invoice)': row['ExpenseCategoryName'],
**{k:0 for k in categories}
}
entry.update(sup_cats)
row_list.append(entry)
df = pd.DataFrame.from_dict(row_list)
df.to_csv('data/processed/invoices_matched.csv', index=False)
if verbose:
print("Error:",error_acc)
if __name__ == "__main__":
import argparse
parser = argparse.ArgumentParser()
parser.add_argument("-s","--solver", type=str, help="Options: \'GA\' for genetic algorithm; \'random\' for random solver", default='GA', dest='solver')
parser.add_argument("-N","--popsize", type=int, help="Number of individuals in the genetic algorithm", default=100, dest='N')
parser.add_argument("-v","--verbose", help="increase output verbosity", dest='verbose', action="store_true")
args = parser.parse_args()
run(solver=args.solver,N=args.N,verbose=args.verbose)