Refactor:
Step 4 - Aggregating Neural Networks Step 5 - Training Neural Networks
This commit is contained in:
parent
9189759320
commit
4a81279b58
@ -4,16 +4,22 @@ import dill
|
||||
from tqdm import tqdm
|
||||
import copy
|
||||
|
||||
from tensorflow.python.keras import backend as K
|
||||
|
||||
from abc import ABC, abstractmethod
|
||||
|
||||
|
||||
class _BaseExperiment(ABC):
|
||||
class Experiment(ABC):
|
||||
|
||||
@staticmethod
|
||||
def from_dill(path):
|
||||
with open(path, "rb") as dill_file:
|
||||
return dill.load(dill_file)
|
||||
|
||||
@staticmethod
|
||||
def reset_model():
|
||||
K.clear_session()
|
||||
|
||||
def __init__(self, name=None, ident=None):
|
||||
self.experiment_id = f'{ident or ""}_{time.time()}'
|
||||
self.experiment_name = name or 'unnamed_experiment'
|
||||
@ -59,22 +65,27 @@ class _BaseExperiment(ABC):
|
||||
dill.dump(value, dill_file)
|
||||
|
||||
@abstractmethod
|
||||
def run_net(self, network, iterations, run_id=0):
|
||||
def run_net(self, net, trains_per_application=100, step_limit=100, run_id=0, **kwargs):
|
||||
raise NotImplementedError
|
||||
pass
|
||||
|
||||
def run_exp(self, network_generator, exp_iterations, prints=False, **kwargs):
|
||||
# INFO Run_ID needs to be more than 0, so that exp stores the trajectories!
|
||||
for run_id in range(exp_iterations):
|
||||
network = network_generator()
|
||||
self.run_net(network, 100, run_id=run_id + 1, **kwargs)
|
||||
self.historical_particles[run_id] = network
|
||||
if prints:
|
||||
print("Fixpoint? " + str(network.is_fixpoint()))
|
||||
self.reset_model()
|
||||
|
||||
class Experiment(_BaseExperiment):
|
||||
|
||||
def __init__(self, **kwargs):
|
||||
super(Experiment, self).__init__(**kwargs)
|
||||
pass
|
||||
|
||||
def run_net(self, network, iterations, run_id=0):
|
||||
pass
|
||||
def reset_all(self):
|
||||
self.reset_model()
|
||||
|
||||
|
||||
class FixpointExperiment(Experiment):
|
||||
if kwargs.get('logging', False):
|
||||
self.log(self.counters)
|
||||
|
||||
def __init__(self, **kwargs):
|
||||
kwargs['name'] = self.__class__.__name__ if 'name' not in kwargs else kwargs['name']
|
||||
@ -82,7 +93,7 @@ class FixpointExperiment(Experiment):
|
||||
self.counters = dict(divergent=0, fix_zero=0, fix_other=0, fix_sec=0, other=0)
|
||||
self.interesting_fixpoints = []
|
||||
|
||||
def run_net(self, net, step_limit=100, run_id=0):
|
||||
def run_net(self, net, step_limit=100, run_id=0, **kwargs):
|
||||
i = 0
|
||||
while i < step_limit and not net.is_diverged() and not net.is_fixpoint():
|
||||
net.self_attack()
|
||||
@ -105,31 +116,56 @@ class FixpointExperiment(Experiment):
|
||||
else:
|
||||
self.counters['other'] += 1
|
||||
|
||||
def reset_counters(self):
|
||||
for key in self.counters.keys():
|
||||
self.counters[key] = 0
|
||||
return True
|
||||
|
||||
def reset_all(self):
|
||||
super(FixpointExperiment, self).reset_all()
|
||||
self.reset_counters()
|
||||
|
||||
|
||||
class MixedFixpointExperiment(FixpointExperiment):
|
||||
|
||||
def run_net(self, net, trains_per_application=100, step_limit=100, run_id=0):
|
||||
|
||||
i = 0
|
||||
while i < step_limit and not net.is_diverged() and not net.is_fixpoint():
|
||||
def run_net(self, net, trains_per_application=100, step_limit=100, run_id=0, **kwargs):
|
||||
for i in range(step_limit):
|
||||
if net.is_diverged() or net.is_fixpoint():
|
||||
break
|
||||
net.self_attack()
|
||||
with tqdm(postfix=["Loss", dict(value=0)]) as bar:
|
||||
for _ in range(trains_per_application):
|
||||
loss = net.compiled().train()
|
||||
bar.postfix[1]["value"] = loss
|
||||
bar.update()
|
||||
i += 1
|
||||
if run_id:
|
||||
net.save_state()
|
||||
self.count(net)
|
||||
|
||||
|
||||
class SoupExperiment(Experiment):
|
||||
pass
|
||||
|
||||
def __init__(self, **kwargs):
|
||||
super(SoupExperiment, self).__init__(name=kwargs.get('name', self.__class__.__name__))
|
||||
|
||||
def run_exp(self, network_generator, exp_iterations, soup_generator=None, soup_iterations=0, prints=False):
|
||||
for i in range(soup_iterations):
|
||||
soup = soup_generator()
|
||||
soup.seed()
|
||||
for _ in tqdm(exp_iterations):
|
||||
soup.evolve()
|
||||
self.log(soup.count())
|
||||
self.save(soup=soup.without_particles())
|
||||
|
||||
def run_net(self, net, trains_per_application=100, step_limit=100, run_id=0, **kwargs):
|
||||
raise NotImplementedError
|
||||
pass
|
||||
|
||||
|
||||
class IdentLearningExperiment(Experiment):
|
||||
|
||||
def __init__(self):
|
||||
super(IdentLearningExperiment, self).__init__(name=self.__class__.__name__)
|
||||
pass
|
||||
|
||||
def run_net(self, net, trains_per_application=100, step_limit=100, run_id=0, **kwargs):
|
||||
pass
|
||||
|
@ -529,57 +529,46 @@ class TrainingNeuralNetworkDecorator:
|
||||
|
||||
|
||||
if __name__ == '__main__':
|
||||
def run_exp(network, prints=False):
|
||||
# INFO Run_ID needs to be more than 0, so that exp stores the trajectories!
|
||||
exp.run_net(network, 100, run_id=run_id + 1)
|
||||
exp.historical_particles[run_id] = network
|
||||
if prints:
|
||||
print("Fixpoint? " + str(network.is_fixpoint()))
|
||||
print("Loss " + str(loss))
|
||||
|
||||
if False:
|
||||
if True:
|
||||
# WeightWise Neural Network
|
||||
net_generator = ParticleDecorator(WeightwiseNeuralNetwork(width=2, depth=2).with_keras_params(activation='linear'))
|
||||
with FixpointExperiment() as exp:
|
||||
for run_id in tqdm(range(10)):
|
||||
net = ParticleDecorator(
|
||||
WeightwiseNeuralNetwork(width=2, depth=2).with_keras_params(activation='linear'))
|
||||
exp.run_exp(net)
|
||||
K.clear_session()
|
||||
exp.log(exp.counters)
|
||||
exp.run_exp(net_generator, 10, logging=True)
|
||||
exp.reset_all()
|
||||
|
||||
if False:
|
||||
# Aggregating Neural Network
|
||||
net_generator = ParticleDecorator(AggregatingNeuralNetwork(aggregates=4, width=2, depth=2).with_keras_params())
|
||||
with FixpointExperiment() as exp:
|
||||
for run_id in tqdm(range(10)):
|
||||
net = ParticleDecorator(
|
||||
AggregatingNeuralNetwork(aggregates=4, width=2, depth=2).with_keras_params())
|
||||
run_exp(net)
|
||||
K.clear_session()
|
||||
exp.log(exp.counters)
|
||||
exp.run_exp(net_generator, 10, logging=True)
|
||||
|
||||
exp.reset_all()
|
||||
|
||||
if False:
|
||||
# FFT Aggregation
|
||||
net_generator = lambda: ParticleDecorator(
|
||||
AggregatingNeuralNetwork(
|
||||
aggregates=4, width=2, depth=2, aggregator=AggregatingNeuralNetwork.aggregate_fft
|
||||
).with_keras_params(activation='linear'))
|
||||
with FixpointExperiment() as exp:
|
||||
for run_id in tqdm(range(10)):
|
||||
net = ParticleDecorator(
|
||||
AggregatingNeuralNetwork(
|
||||
aggregates=4, width=2, depth=2, aggregator=AggregatingNeuralNetwork.aggregate_fft
|
||||
).with_keras_params(activation='linear'))
|
||||
run_exp(net)
|
||||
K.clear_session()
|
||||
exp.log(exp.counters)
|
||||
exp.run_exp(net_generator, 1)
|
||||
exp.log(exp.counters)
|
||||
exp.reset_model()
|
||||
exp.reset_all()
|
||||
|
||||
if True:
|
||||
# ok so this works quite realiably
|
||||
with FixpointExperiment() as exp:
|
||||
run_count = 1000
|
||||
net = TrainingNeuralNetworkDecorator(ParticleDecorator(WeightwiseNeuralNetwork(width=2, depth=2)))
|
||||
net.with_params(epsilon=0.0001).with_keras_params(optimizer='sgd')
|
||||
run_count = 10000
|
||||
net_generator = TrainingNeuralNetworkDecorator(
|
||||
ParticleDecorator(WeightwiseNeuralNetwork(width=2, depth=2))
|
||||
).with_params(epsilon=0.0001).with_keras_params(optimizer='sgd')
|
||||
with MixedFixpointExperiment() as exp:
|
||||
for run_id in tqdm(range(run_count+1)):
|
||||
net.compiled()
|
||||
loss = net.train(epoch=run_id)
|
||||
exp.run_exp(net_generator, 1)
|
||||
if run_id % 100 == 0:
|
||||
run_exp(net)
|
||||
exp.run_net(net_generator, 1)
|
||||
K.clear_session()
|
||||
|
||||
if False:
|
||||
|
58
code/soup.py
58
code/soup.py
@ -109,39 +109,29 @@ class Soup(object):
|
||||
|
||||
|
||||
if __name__ == '__main__':
|
||||
if False:
|
||||
with SoupExperiment() as exp:
|
||||
for run_id in range(1):
|
||||
net_generator = lambda: WeightwiseNeuralNetwork(2, 2).with_keras_params(activation='linear').with_params()
|
||||
# net_generator = lambda: FFTNeuralNetwork(2, 2).with_keras_params(activation='linear').with_params()
|
||||
# net_generator = lambda: AggregatingNeuralNetwork(4, 2, 2).with_keras_params(activation='sigmoid')\
|
||||
# .with_params(shuffler=AggregatingNeuralNetwork.shuffle_random)
|
||||
# net_generator = lambda: RecurrentNeuralNetwork(2, 2).with_keras_params(activation='linear').with_params()
|
||||
soup = Soup(100, net_generator).with_params(remove_divergent=True, remove_zero=True)
|
||||
soup.seed()
|
||||
for _ in tqdm(range(1000)):
|
||||
soup.evolve()
|
||||
exp.log(soup.count())
|
||||
exp.save(soup=soup.without_particles())
|
||||
if True:
|
||||
net_generator = lambda: WeightwiseNeuralNetwork(2, 2).with_keras_params(activation='linear').with_params()
|
||||
soup_generator = Soup(100, net_generator).with_params(remove_divergent=True, remove_zero=True)
|
||||
exp = SoupExperiment()
|
||||
exp.run_exp(net_generator, 1000, soup_generator, 1, False)
|
||||
|
||||
# net_generator = lambda: FFTNeuralNetwork(2, 2).with_keras_params(activation='linear').with_params()
|
||||
# net_generator = lambda: AggregatingNeuralNetwork(4, 2, 2).with_keras_params(activation='sigmoid')\
|
||||
# .with_params(shuffler=AggregatingNeuralNetwork.shuffle_random)
|
||||
# net_generator = lambda: RecurrentNeuralNetwork(2, 2).with_keras_params(activation='linear').with_params()
|
||||
|
||||
if True:
|
||||
with SoupExperiment("soup") as exp:
|
||||
for run_id in range(1):
|
||||
net_generator = lambda: TrainingNeuralNetworkDecorator(WeightwiseNeuralNetwork(2, 2))\
|
||||
.with_keras_params(activation='linear').with_params(epsilon=0.0001)
|
||||
# net_generator = lambda: TrainingNeuralNetworkDecorator(AggregatingNeuralNetwork(4, 2, 2))
|
||||
# .with_keras_params(activation='linear')\
|
||||
# .with_params(shuffler=AggregatingNeuralNetwork.shuffle_random)
|
||||
# net_generator = lambda: TrainingNeuralNetworkDecorator(FFTNeuralNetwork(4, 2, 2))\
|
||||
# .with_keras_params(activation='linear')\
|
||||
# .with_params(shuffler=AggregatingNeuralNetwork.shuffle_random)
|
||||
# net_generator = lambda: RecurrentNeuralNetwork(2, 2).with_keras_params(activation='linear').with_params()
|
||||
soup = Soup(100, net_generator).with_params(remove_divergent=True, remove_zero=True, train=20)
|
||||
soup.seed()
|
||||
for _ in tqdm(range(100)):
|
||||
soup.evolve()
|
||||
exp.log(soup.count())
|
||||
# you can access soup.historical_particles[particle_uid].states[time_step]['loss']
|
||||
# or soup.historical_particles[particle_uid].states[time_step]['weights']
|
||||
# from soup.dill
|
||||
exp.save(soup=soup.without_particles())
|
||||
net_generator = lambda: TrainingNeuralNetworkDecorator(WeightwiseNeuralNetwork(2, 2)) \
|
||||
.with_keras_params(activation='linear').with_params(epsilon=0.0001)
|
||||
soup_generator = lambda: Soup(100, net_generator).with_params(remove_divergent=True, remove_zero=True, train=20)
|
||||
exp = SoupExperiment(name="soup")
|
||||
|
||||
exp.run_exp(net_generator, 100, soup_generator, 1, False)
|
||||
|
||||
# net_generator = lambda: TrainingNeuralNetworkDecorator(AggregatingNeuralNetwork(4, 2, 2))
|
||||
# .with_keras_params(activation='linear')\
|
||||
# .with_params(shuffler=AggregatingNeuralNetwork.shuffle_random)
|
||||
# net_generator = lambda: TrainingNeuralNetworkDecorator(FFTNeuralNetwork(4, 2, 2))\
|
||||
# .with_keras_params(activation='linear')\
|
||||
# .with_params(shuffler=AggregatingNeuralNetwork.shuffle_random)
|
||||
# net_generator = lambda: RecurrentNeuralNetwork(2, 2).with_keras_params(activation='linear').with_params()
|
||||
|
Loading…
x
Reference in New Issue
Block a user