def test_SplitRandom_ratios():
train, val, test = range(1000) >> SplitRandom(ratio=(0.6, 0.3, 0.1))
assert len(train) == 600
assert len(val) == 300
assert len(test) == 100
with pytest.raises(ValueError) as ex:
range(1000) >> SplitRandom(ratio=(0.6, 0.7))
assert str(ex.value).startswith('Ratios must sum up to one')
def test_SplitRandom_constraint():
same_letter = lambda t: t[0]
data = zip('aabbccddee', range(10))
train, val = data >> SplitRandom(
rand=rnd.Random(0), ratio=0.6, constraint=same_letter) >> Map(sorted)
assert train == [('a', 0), ('a', 1), ('b', 2), ('b', 3), ('c', 4),
('c', 5)]
assert val == [('d', 6), ('d', 7), ('e', 8), ('e', 9)]
def test_SplitRandom_constraint():
same_letter = lambda t: t[0]
data = zip('aabbccddee', range(10))
train, val = data >> SplitRandom(rand=StableRandom(0), ratio=0.6,
constraint=same_letter) >> Collect()
print(train)
print(val)
assert train == [('a', 1), ('a', 0), ('d', 7), ('b', 2), ('d', 6), ('b', 3)]
assert val == [('c', 5), ('e', 8), ('e', 9), ('c', 4)]
def test_SplitRandom_constraint():
same_letter = lambda t: t[0]
data = zip('aabbccddee', range(10))
train, val = data >> SplitRandom(
rand=None, ratio=0.6, constraint=same_letter) >> Collect()
train.sort()
val.sort()
assert train == [('a', 0), ('a', 1), ('b', 2), ('b', 3), ('d', 6),
('d', 7)]
assert val == [('c', 4), ('c', 5), ('e', 8), ('e', 9)]
def train():
from keras.metrics import categorical_accuracy
rerange = TransformImage(0).by('rerange', 0, 255, 0, 1, 'float32')
build_batch = (BuildBatch(BATCH_SIZE).by(0, 'image', 'float32').by(
1, 'one_hot', 'uint8', NUM_CLASSES))
p = 0.1
augment = (AugmentImage(0).by('identical', 1.0).by(
'elastic', p, [5, 5], [100, 100],
[0, 100]).by('brightness', p,
[0.7, 1.3]).by('color', p, [0.7, 1.3]).by(
'shear', p,
[0, 0.1]).by('fliplr', p).by('rotate', p, [-10, 10]))
plot_eval = PlotLines((0, 1), layout=(2, 1))
print('creating network...')
network = create_network()
print('loading data...')
train_samples, test_samples = load_samples()
train_samples, val_samples = train_samples >> SplitRandom(0.8)
print('training...', len(train_samples), len(val_samples))
for epoch in range(NUM_EPOCHS):
print('EPOCH:', epoch)
t_loss, t_acc = (train_samples >> PrintProgress(train_samples) >>
Pick(PICK) >> augment >> rerange >> Shuffle(100) >>
build_batch >> network.train() >> Unzip())
t_loss, t_acc = t_loss >> Mean(), t_acc >> Mean()
print("train loss : {:.6f}".format(t_loss))
print("train acc : {:.1f}".format(100 * t_acc))
v_loss, v_acc = (val_samples >> rerange >> build_batch >>
network.validate() >> Unzip())
v_loss, v_acc = v_acc >> Mean(), v_acc >> Mean()
print('val loss : {:.6f}'.format(v_loss))
print('val acc : {:.1f}'.format(100 * v_acc))
network.save_best(v_acc, isloss=False)
plot_eval((t_acc, v_acc))
print('testing...', len(test_samples))
e_acc = (test_samples >> rerange >> build_batch >> network.evaluate(
[categorical_accuracy]))
print('test acc : {:.1f}'.format(100 * e_acc))
def test_SplitRandom_seed():
split1 = range(10) >> SplitRandom(rand=StableRandom(0))
split2 = range(10) >> SplitRandom(rand=StableRandom(0))
split3 = range(10) >> SplitRandom(rand=StableRandom(1))
assert split1 == split2
assert split1 != split3
def test_SplitRandom_stable_default():
split1 = range(10) >> SplitRandom()
split2 = range(10) >> SplitRandom()
assert split1 == split2
def test_SplitRandom_split():
train, val = range(1000) >> SplitRandom(ratio=0.7)
assert len(train) == 700
assert len(val) == 300
assert not set(train).intersection(val)
def load_samples() -> (Samples, Samples, Samples):
read = ReadLabelDirs(CFG.datadir, '*.json')
label2int = MapCol(1, int)
return read >> label2int >> SplitRandom(ratio=CFG.ratios)