print("end loading\n")
df['WEEK_END'] = df['DAY_WE_DS'].apply(lambda x: int(x >= 5))
print(df.head())
print(len(df))
#limits for each assignment from the start to the week w
limits = [[0 for k in range(28)] for v in range(12)]
#finding maximal value and average value
for i in range(28):
limits[w][i] = df[(df.ASS_ASSIGNMENT == i)].CSPL_RECEIVED_CALLS.max(
axis=0)
#0 special case for Evenements
limits[centre('Evenements')][w] = 0
assign = np.unique(df.ASS_ASSIGNMENT)
print('end !')
print(time.time() - start_time)
start_time = time.time()
# In[ ]:
#start of training
clf = [[] for i in range(12)]
for i in assign:
clf[w].append(
ensemble.GradientBoostingRegressor(
loss='ls',
max_depth=param[i]['max_depth'],
import numpy as np
import pandas as pd
from Calculhoraire_exclusif import holidays
from Calculhoraire_exclusif import centre
from Calculhoraire_exclusif import intervalle
data_name = "train_2011_2012_2013.csv"
cols = ['DATE', 'ASS_ASSIGNMENT', 'CSPL_RECEIVED_CALLS']
print("loading data...")
df = pd.read_csv(data_name, usecols=cols, sep=";")
print("end loading\n")
df['TIME_SLOT'] = df['DATE'].apply(lambda x: intervalle(x))
df['ASS_ASSIGNMENT'] = df['ASS_ASSIGNMENT'].apply(lambda x: centre(x))
from datetime import datetime
df['DATE'] = df['DATE'].apply(
lambda x: datetime.strptime(x, "%Y-%m-%d %H:%M:%S.%f"))
df['DAY_WE_DS'] = df['DATE'].dt.weekday
df['short_DATE'] = df['DATE'].dt.date
df['short_DATE'] = df['short_DATE'].apply(
lambda x: datetime.strftime(x, "%Y-%m-%d"))
df['HO'] = df['short_DATE'].apply(lambda x: holidays(str(x)))
df["MONTH"] = df['short_DATE'].apply(lambda x: int(x.split("-")[1]))
df["YEAR"] = df['short_DATE'].apply(lambda x: int(x.split("-")[0]))
df["DAY"] = df['short_DATE'].apply(lambda x: int(x.split("-")[2]))
df2 = df.groupby([
'short_DATE', 'YEAR', 'MONTH', 'DAY', 'DAY_WE_DS', 'TIME_SLOT', 'HO',
dsub['HO'] = dsub['short_DATE'].apply(lambda x: holidays(x))
print("End of conversion!")
dsub['DAY_WE_DS'] = dsub['DATE'].dt.weekday
dsub['WEEK_END'] = dsub['DAY_WE_DS'].apply(lambda x: int(x >= 5))
dsub['DAY'] = dsub['DATE'].dt.day
dsub['MONTH'] = dsub['DATE'].dt.month
dsub['YEAR'] = dsub['DATE'].dt.year
dsub['TIME_SLOT'] = dsub['DATE'].apply(lambda x: intervalle(x))
# In[16]:
prediction = []
for row in dsub.iterrows():
ass = row[1]['ASS_ASSIGNMENT']
if limits[centre(ass)] == 0:
prediction.append(0)
continue
time = row[1]['TIME_SLOT']
wk = row[1]['WEEK_END']
dayds = row[1]['DAY_WE_DS']
month = row[1]['MONTH']
year = row[1]['YEAR']
day = row[1]['DAY']
holidays = row[1]['HO']
pred = clf[centre(ass)].predict(
[[year, month, day, time, holidays, dayds, wk]])
pred = max(0, pred)
pred = min(limits[centre(ass)] / 2, pred)
pred = 2 * pred
prediction.append(pred)
dsub['MONTH'] = dsub['DATE'].dt.month
dsub['YEAR'] = dsub['DATE'].dt.year
dsub['TIME_SLOT'] = dsub['DATE'].apply(lambda x: intervalle(x))
# In[ ]:
prediction = []
for row in dsub.iterrows():
ass = row[1]['ASS_ASSIGNMENT']
time = row[1]['TIME_SLOT']
wk = row[1]['DAY_WE_DS']
month = row[1]['MONTH']
year = row[1]['YEAR']
day = row[1]['DAY']
holidays = row[1]['HO']
pred = clf[centre(ass)].predict([[wk, time, month, year, day, holidays]])
prediction.append(max(0, pred))
#arrondit à l'entier le plus proche
prediction = [int(x) for x in prediction]
#mise à jour
dsub['prediction'] = prediction
dsub['DATE'] = dsub['DATE'].apply(
lambda x: datetime.strftime(x, "%Y-%m-%d %H:%M:%S"))
dsub['DATE'] = dsub['DATE'] + '.000'
header = ['DATE', 'ASS_ASSIGNMENT', 'prediction']
#Ecriture
dsub.to_csv("sortie.txt", columns=header, sep='\t', index=False)
# In[ ]:
# In[4]:
errors = [0 for i in range(len(assignments))]
from datetime import timedelta
for row in dsub.iterrows():
ass = row[1]['ASS_ASSIGNMENT']
date = row[1]['DATE']
date = date + timedelta(days=-7)
date = date.strftime("%Y-%m-%d %H:%M:%S")
date += '.000'
prediction = row[1]['prediction']
yreal = dreal2.get((date, ass), 0)
errors[centre(ass)] += (exp(0.1 * (yreal - prediction)) - 0.1 *
(yreal - prediction) - 1)
errors
# In[5]:
sum(errors)
# In[7]:
error = sum(errors) / 82909
error
# In[ ]:
bestmult = [0 for i in range(len(assignments))]
# In[26]:
for i in range(5):
tmperrors = [0 for j in range(len(assignments))]
for row in dsub.iterrows():
ass = row[1]['ASS_ASSIGNMENT']
date = row[1]['DATE']
date = date + timedelta(days=-7)
date = date.strftime("%Y-%m-%d %H:%M:%S")
date += '.000'
prediction = row[1]['prediction']
yreal = dreal2.get((date, ass), 0)
tmperrors[centre(ass)] += (exp(0.1 * (yreal - (i + 1) * prediction)) -
0.1 * (yreal - (i + 1) * prediction) - 1)
print(tmperrors)
for a in range(len(assignments)):
if (errors[a] > tmperrors[a]):
errors[a] = tmperrors[a]
bestmult[a] = i + 1
# In[27]:
print(errors)
print(bestmult)
# In[28]:
# In[5]:
