def train(net, opt, trainBatcher, validBatcher, saver, minContext):
while True:
start = time.time()
#Run epochs
trainLoss, trainAcc = utils.runData(net,
opt,
trainBatcher,
trainable=True,
verbose=True)
validLoss, validAcc = utils.runData(net,
opt,
validBatcher,
minContext=minContext)
trainLoss, validLoss = np.exp(trainLoss), np.exp(validLoss)
#Print statistics
print('\nEpoch: ', saver.epoch(), ', Time: ', time.time() - start)
print('| Train Perp: ', trainLoss, ', Train Acc: ', trainAcc)
print('| Valid Perp: ', validLoss, ', Valid Acc: ', validAcc)
if np.isnan(validLoss) or np.isnan(trainLoss):
print('Got a bad update. Resetting epoch')
saver.refresh(net)
else:
saver.update(net, trainLoss, trainAcc, validLoss, validAcc)
def runToScore(model, dataset, data_paths, n=-1):
'''
Runs on n many train data and returns the score and the resultant image
_______
In:
model: The model being used
dataset: The ProbaVDataset
data_paths: List of paths of the train dataset
Out:
scores: The scores vs the baseline
results: The resultant images
'''
scores = []
results = []
model.to(device)
model.eval()
if n < 0:
n = len(dataset)
counter = 0
for d, _, _ in dataset:
if counter < n:
result = runData(model, d)
score = score_image(result, data_paths[counter])
scores.append(score)
results.append(result)
counter += 1
else:
break
return scores, results
def predictTest(model_file):
model = torch.load(model_file).to(device)
data_paths = all_scenes_paths(path_to_test)
dataset = pd.ProbaVDataset(data_paths, True)
images = []
model.eval()
for d in dataset:
images.append(runData(model, d))
return images
def test(net, batcher, minContext, name='Test'):
start = time.time()
loss, acc = utils.runData(net, None, batcher, minContext=minContext)
loss = np.exp(loss)
#Print statistics
print('Time: ', time.time() - start)
print(name, ' Perp: ', loss, ', Acc: ', acc)
self.som[mask] += self.alpha * update[...,None][mask]*self.diff[mask]
self.som[self.som > 1] = 1
self.som[self.som < 0] = 0
def updateLearnParams(self):
'''
Update the learning parameters for the next iteration.
'''
self.curIter += 1
self.alpha = self.alphaInit * exp(-self.curIter/self.delta)
omega = self.r * exp(-self.curIter/self.delta)
self.omega2 = omega * omega
if __name__ == '__main__':
from matplotlib import pyplot as pp
from matplotlib import animation as ani
from utils import runData, simpleView
colors = array([[0,0,0],
[1,0,0],
[0,1,0],
[0,0,1],
[1,0,1],
[1,1,0],
[0,1,1],
[1,1,1]])
som = SOM(3,shape=(20,20,20))
runData(som, colors, 1000)
simpleView(som, shape=(1,1), color=True)