Page MenuHomec4science
Files F122126261

ann.py
No OneTemporary
Actions

File Metadata

Created
Tue, Jul 15, 23:15

ann.py
View Options

import pandas as pd
from torch import nn
from torch import optim
from daily.data import DailyPreprocessor, DailyDataset, DailyCrossValidation
class Ann:
"""Neural network optimized for daily prediction"""
def __init__(self, data: DailyPreprocessor, n_features, n_output, n_hidden):
"""Initialize the ANN with the best parameters found"""
self.model = nn.Sequential(
nn.Linear(n_features, n_hidden), nn.LeakyReLU(),
nn.Linear(n_hidden, n_output), nn.LeakyReLU()
)
self.data = data
self.criterionH = nn.MSELoss()
self.optimizer = optim.Adam(self.model.parameters(), lr=0.00001, betas=(0.9, 0.999), eps=1e-8)
def do_epoch(self):
"""Execute single epoch of learning on the training set"""
for feat, target in self.data.train_loader:
self.optimizer.zero_grad()
out = self.model(feat)
loss = self.criterionH(out, target)
loss.backward()
self.optimizer.step()
def fit(self, preprocessor: DailyPreprocessor):
"""Executes the optimum number of epochs to learn the model from the data of `preprocessor`"""
for j in range(5):
self.do_epoch()
print('Epoch:', j, 'Ln Q error on test:', preprocessor.evaluate(self.model, train=False))
def get_summer_days(df):
"""Returns warm days that should be removed when predicting heating needs"""
start_summer = pd.Timestamp(year=2017, month=3, day=15)
end_summer = pd.Timestamp(year=2017, month=11, day=15)
df = df[start_summer < df.index.map(pd.Timestamp)]
return df[df.index.map(pd.Timestamp) < end_summer].index
def run_daily_ann(building_features, weather_forecast, labels):
"""
Runs 4-fold cross validation of daily prediction for the best ANN model with provided data
:returns: Ln Q errors on testing set for each fold
"""
useful_weather = ['G_Dh_mean', 'G_Bn_mean', 'G_Dh_var', 'G_Bn_var', 'Ta_mean', 'Ta_var', 'Ta_min',
'FF_mean', 'FF_var', 'h_day']
weather_forecast = weather_forecast[useful_weather].copy()
summer = get_summer_days(weather_forecast)
weather_forecast.drop(summer, inplace=True)
labels.drop(summer, inplace=True)
dataset = DailyDataset(
building_features=building_features,
weather_forecast=weather_forecast,
labels=labels
)
ln_q = []
for k, data in enumerate(DailyCrossValidation(dataset, 4)):
print('Cross validation step:', k)
ann = Ann(data,
n_features=data.train.nb_features,
n_output=1,
n_hidden=50)
ann.fit(data)
ln_q.append(data.evaluate(ann.model, train=False))
return ln_q

Event Timeline

c4science · Help