def predict(self, business_forecast, training_data=pd.DataFrame()):
sum_bins = np.zeros(6)
for i in business_forecast:
sched = ActualSchedule(training_data[\
training_data.user_tz == i['user_tz']])
stuff = (sched.bins() + self.smoother) / (sched.bins() + self.smoother).sum()
sum_bins += stuff * i['frequency'] * i['schedule_type']
return sum_bins
def predict(self, business_forecast, training_data=pd.DataFrame()):
sum_bins = np.zeros(6)
smoother = np.array([0.01, 0.01, 0.01, 0.01, 0.01, 0.01])
for i in business_forecast:
sched = ActualSchedule(self.training_data[\
self.training_data.schedule_type == i['schedule_type']])
stuff = (sched.bins() + smoother) / (sched.bins() + smoother).sum()
sum_bins += stuff * i['frequency'] * i['schedule_type']
return sum_bins
def predict(self, business_forecast, training_data=pd.DataFrame()):
# try company - show chi-square test
# try timezone - show chi-square test
# timezone generally is not statistically significant (2x and 3x).
# Significant for 4x and 5x but
# try company and timezone
# too difficult. Already pretty accurate
actual_schedule = ActualSchedule(self.filtered_training_data(business_forecast))
bins = actual_schedule.bins()
b = bins / bins.sum() * self.num_business_forecast_lessons(business_forecast)
return b
def errors(self):
indices = [] # index of a row, formatted YYYY-M (e.g. 2016-8, 2016-9)
predictions = []
errors = [] # errors. Lower is better
test_data_size = []
actuals = []
for i in range(0, self.num_months() - self.training_data_span_months):
m = self._model()
training_data = self.training_data(i,
self.training_data_span_months)
m.fit(training_data)
td = self.test_data(i + self.training_data_span_months)
print "Generating prediction..."
prediction = m.predict(\
BusinessForecast(\
td).convert(),
training_data=training_data)
actual = ActualSchedule(\
td).bins()
print "prediction "
print prediction
print "\n"
print "actual"
print actual
print "\n"
error = mean_absolute_error(actual, prediction)
errors.append(error)
index = self.year_month_index(i + self.training_data_span_months)
indices.append(index)
print "time: {}, error: {}".format(index, error)
predictions.append(prediction)
actuals.append(actual)
test_data_size.append(actual.sum())
return pd.DataFrame({
'errors': errors,
'test_data_size': test_data_size,
'predictions': predictions,
'actuals': actuals
}).set_index([indices])
def errors(self):
indices = [] # index of a row, formatted YYYY-M (e.g. 2016-8, 2016-9)
predictions = []
errors = [] # errors. Lower is better
test_data_size = []
actuals = []
for i in range(0,self.num_months()-self.training_data_span_months):
m = self._model()
training_data = self.training_data(i, self.training_data_span_months)
m.fit(training_data)
td = self.test_data(i+self.training_data_span_months)
print "Generating prediction..."
prediction = m.predict(\
BusinessForecast(\
td).convert(),
training_data=training_data)
actual = ActualSchedule(\
td).bins()
print "prediction "
print prediction
print "\n"
print "actual"
print actual
print "\n"
error = mean_absolute_error(actual, prediction)
errors.append(error)
index = self.year_month_index(i+self.training_data_span_months)
indices.append(index)
print "time: {}, error: {}".format(index, error)
predictions.append(prediction)
actuals.append(actual)
test_data_size.append(actual.sum())
return pd.DataFrame({
'errors': errors,
'test_data_size': test_data_size,
'predictions': predictions,
'actuals': actuals
}).set_index([indices])
def converts_it_to_an_actual_schedule(self):
user_tz = ["Brasilia",
"Brasilia",
"Pacific (US & Canada)",
"Eastern (US & Canada)",
"Brasilia"
]
l1_time = range(0, 5)
l1_day = range(0, 5)
l2_time = range(1, 6)
l2_day = range(0, 5)
l3_time = range(2, 7)
l3_day = range(0, 5)
l4_time = range(3, 8)
l4_day = range(0, 5)
schedule_type = [4,4,4,4,4]
args = {
'user_tz': user_tz,
'l1_time': l1_time,
'l1_day': l1_day,
'l2_time': l2_time,
'l2_day': l2_day,
'l3_time': l3_time,
'l3_day': l3_day,
'l4_time': l4_time,
'l4_day': l4_day,
'schedule_type': schedule_type
}
unique_user_summaries = pd.DataFrame(args)
schedule = ActualSchedule(unique_user_summaries)
bins = schedule.bins()
expect(bins[0]).to.equal(10)
expect(bins[1]).to.equal(10)
expect(bins[2]).to.equal(0)
expect(bins[3]).to.equal(0)
expect(bins[4]).to.equal(0)
expect(bins[5]).to.equal(0)
def converts_it_to_an_actual_schedule(self):
user_tz = [
"Brasilia", "Brasilia", "Pacific (US & Canada)",
"Eastern (US & Canada)", "Brasilia"
]
l1_time = range(0, 5)
l1_day = range(0, 5)
l2_time = range(1, 6)
l2_day = range(0, 5)
l3_time = range(2, 7)
l3_day = range(0, 5)
l4_time = range(3, 8)
l4_day = range(0, 5)
schedule_type = [4, 4, 4, 4, 4]
args = {
'user_tz': user_tz,
'l1_time': l1_time,
'l1_day': l1_day,
'l2_time': l2_time,
'l2_day': l2_day,
'l3_time': l3_time,
'l3_day': l3_day,
'l4_time': l4_time,
'l4_day': l4_day,
'schedule_type': schedule_type
}
unique_user_summaries = pd.DataFrame(args)
schedule = ActualSchedule(unique_user_summaries)
bins = schedule.bins()
expect(bins[0]).to.equal(10)
expect(bins[1]).to.equal(10)
expect(bins[2]).to.equal(0)
expect(bins[3]).to.equal(0)
expect(bins[4]).to.equal(0)
expect(bins[5]).to.equal(0)