Page MenuHomec4science
Files F123817019

ann.py
No OneTemporary
Actions

File Metadata

Created
Tue, Jul 29, 17:51

ann.py
View Options

from torch import nn
from torch import optim
from annual.preprocessor import DataPreProcessor
class Ann:
"""Best ANN version for annual predictions of heating needs"""
def __init__(self, data: DataPreProcessor, n_features: int, n_output: int, n_hidden: int):
"""Initialize an ANN for the given data and with provided number of neurons in each layer"""
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=0.01, betas=(0.9, 0.999), eps=1e-8)
def do_epoch(self, evaluate=True):
"""
Do one epoch of learning
If `evaluate` is True, will return the Ln Q error for training and testing set at the end of the epoch
"""
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()
if evaluate:
train_predictions = self.model(self.data.train.features_t).detach().numpy()
test_predictions = self.model(self.data.test.features_t).detach().numpy()
ln_q_tr = self.data.evaluate(self.data.train, train_predictions)
ln_q_te = self.data.evaluate(self.data.test, test_predictions)
return ln_q_tr, ln_q_te
else:
return None, None
def run_ann(data: DataPreProcessor):
"""
Runs the best ANN model for yearly heating prediction,
:returns: the average Ln Q error for training and testing at the end
"""
ann = Ann(data,
n_features=data.train.features.shape[1],
n_output=1,
n_hidden=8)
for j in range(5000):
ln_q_tr, ln_q_te = ann.do_epoch(j % 1000 == 0)
if j % 1000 == 0:
print('Epoch %d/5000: Ln Q error on test: %f' % (j, ln_q_te))
ln_q_tr, ln_q_te = ann.do_epoch(True)
return ln_q_tr, ln_q_te

Event Timeline

c4science · Help