def build_temp_calendar(temps, year, month, day):
day_of_year = np.zeros(temps.size)
for i_row in range(temps.size):
day_of_year[i_row] = tools.find_day_of_year(year[i_row], month[i_row],
day[i_row])
median_temp_calendar = np.zeros(366)
ten_day_medians = np.zeros(temps.size)
for i_day in range(0, 365):
# create 10-day window
low_day = i_day - 5
high_day = i_day + 4
if low_day < 0:
low_day += 365
if high_day > 365:
high_day += -365
if low_day < high_day:
i_window_days = np.where(
np.logical_and(day_of_year >= low_day,
day_of_year <= high_day))
else:
i_window_days = np.where(
np.logical_or(day_of_year >= low_day, day_of_year <= high_day))
ten_day_median = np.median(temps[i_window_days])
median_temp_calendar[i_day] = ten_day_median
ten_day_medians[np.where(day_of_year == i_day)] = ten_day_median
#Handle '364' years, to be handled the same as normal years, to make things easier
if i_day == 364:
ten_day_medians[np.where(day_of_year == 365)] = ten_day_median
median_temp_calendar[365] = ten_day_median
return median_temp_calendar
def build_temp_calendar(temps, year, month, day):
"""
Create an arrary of typical temperatures by day-of-year.
Day 0 = Jan 1, etc.
Parameters
----------
temps: arrary of floats
year, month, day: arrary of ints
Returns
-------
median_temp_calendar: array of floats length 366
"""
# start with an array of zero's according to the proper temp. size
day_of_year = np.zeros(temps.size)
for i_row in range(temps.size):
day_of_year[i_row] = tools.find_day_of_year(year[i_row], month[i_row],
day[i_row])
# Create 10-day medians for each day of the year
#we use median instead of mean to avoid outliers skewing our calcs
median_temp_calendar = np.zeros(366)
ten_day_medians = np.zeros(temps.size)
for i_day in range(0, 365):
# create 10-day window
low_day = i_day - 5
high_day = i_day + 4
if low_day < 0:
low_day += 365
if high_day > 365:
high_day += -365
if low_day < high_day:
i_window_days = np.where(
np.logical_and(day_of_year >= low_day,
day_of_year <= high_day))
else:
i_window_days = np.where(
np.logical_or(day_of_year >= low_day, day_of_year <= high_day))
ten_day_median = np.median(temps[i_window_days])
median_temp_calendar[i_day] = ten_day_median
ten_day_medians[np.where(day_of_year == i_day)] = ten_day_median
#Handle '364' years, to be handled the same as normal years, to make things easier
if i_day == 364:
ten_day_medians[np.where(day_of_year == 365)] = ten_day_median
median_temp_calendar[365] = ten_day_median
#print(i_day, low_day, high_day, i_window_days[0].size)
#print(ten_day_medians.size, np.unique(ten_day_medians), ten_day_medians)
return median_temp_calendar
def find_seasonal_temp(self, year, month, day):
"""
For a given day, month, and year, find the seasonal
high temperature for Fort Lauderdale Beach.
Parameters
----------
year, month, day: int
The date of interest
Returns
-------
seasonal_temp: float
"""
doy = tools.find_day_of_year(year, month, day)
seasonal_temp = self.temp_calendar[doy]
return seasonal_temp
def build_temp_calendar(self, temps, year, month, day):
"""
Create an array of typical temperatures by day-of-year.
Day 0 = Jan 1, etc.
Parameters
----------
temps: array of floats
year, month, day: array of ints
Returns
-------
median_temp_calendar: array of floats of length 366
"""
day_of_year = np.zeros(temps.size)
for i_row in range(temps.size):
day_of_year[i_row] = tools.find_day_of_year(
year[i_row], month[i_row], day[i_row])
## Create 10-day medians for each day of the year.
median_temp_calendar = np.zeros(366)
for i_day in range(0, 365):
low_day = i_day - 5
high_day = i_day + 4
if low_day < 0:
low_day += 365
if high_day > 365:
high_day += -365
if low_day < high_day:
i_window_days = np.where(
np.logical_and(day_of_year >= low_day,
day_of_year <= high_day))
else:
i_window_days = np.where(
np.logical_or(day_of_year >= low_day,
day_of_year <= high_day))
ten_day_median = np.median(temps[i_window_days])
median_temp_calendar[i_day] = ten_day_median
if i_day == 364:
median_temp_calendar[365] = ten_day_median
return median_temp_calendar
def test():
"""
Run through the data history, calculating the prediction
error for each day.
"""
# Create and initialize a new Predictor
predictor = Predictor()
temps, year, month, day = predictor.get_data()
deseasonalized_temps = np.zeros(temps.size)
doy = np.zeros(temps.size, dtype=np.int)
for i, temp in enumerate(temps):
seasonal_temp = predictor.find_seasonal_temp(year[i], month[i], day[i])
deseasonalized_temps[i] = temp - seasonal_temp
doy[i] = tools.find_day_of_year(year[i], month[i], day[i])
# Make predictions for three days into the future.
deseasonalized_predictions = np.zeros(temps.size)
for i, temp in enumerate(deseasonalized_temps):
deseasonalized_predictions[i] = predictor.predict_deseasonalized(temp)
predictions = np.zeros(temps.size - 3)
for i, temp in enumerate(deseasonalized_predictions[:-3]):
predictions[i] = predictor.reseasonalize(temp, doy[i + 3])
residuals = temps[3:] - predictions
print('MEA:', np.mean(np.abs(residuals)))
actuals = temps[3:]
# Compare predictions three days away vs. actuals for above/below 85.
sensitivity = []
targets = np.arange(84, 90)
for target in targets:
i_warm = np.where(actuals > 85)[0]
i_warm_predictions = np.where(predictions > target)[0]
n_true_positives = np.intersect1d(i_warm, i_warm_predictions).size
n_false_negatives = np.setdiff1d(i_warm, i_warm_predictions).size
n_false_positives = np.setdiff1d(i_warm_predictions, i_warm).size
n_true_negatives = (actuals.size - n_true_positives -
n_false_positives - n_false_negatives)
sensitivity.append(n_true_positives /
(n_true_positives + n_false_positives))
def predict(self, year, month, day, past_temp):
"""
Parameters
----------
year, month, day: ints
past_temp: float
The temperature from 3 days before the date of interest.
"""
# Make a prediction for a single day.
doy = tools.find_day_of_year(year, month, day)
doy_past = doy - 3
if doy_past < 0:
doy_past += 365
deseasonalized_temp = self.deseasonalize(past_temp, doy_past)
# Make predictions for three days into the future.
deseasonalized_prediction = self.predict_deseasonalized(
deseasonalized_temp)
prediction = self.reseasonalize(deseasonalized_prediction, doy)
return prediction
def predict(year, month, day, temperature_calendar):
"""
For a given day, month, and year, predict
the high temperature for Fort Lauderdale Beach.
Parameters
----------
year, month, day: int
The date of interest
temperature_calendar: arrary of floats
The typical temperature for each day of the year
Jan 1 = 0,etc.
Returns
-------
prediction: float
"""
day = tools.find_day_of_year(year, month, day)
# Looks up the associated temperature, and returns that as a prediction
prediction = temperature_calendar[day]
return prediction
def predict(year, month, day, temperature_calendar):
day = tools.find_day_of_year(year, month, day)
prediction = temperature_calendar[day]
return prediction