from training_base import training_base
from Losses import loss_NLL
#also does all the parsing
train = training_base(testrun=False)
if not train.modelSet():
from DeepJet_models_ResNet import resnet_model
train.setModel(resnet_model)
train.compileModel(learningrate=0.0004,
loss=['categorical_crossentropy', loss_NLL],
metrics=['accuracy'],
loss_weights=[1., 0.000000000000001])
model, history = train.trainModel(nepochs=50,
batchsize=10000,
stop_patience=300,
lr_factor=0.5,
lr_patience=10,
lr_epsilon=0.0001,
lr_cooldown=2,
lr_minimum=0.0001,
maxqsize=100)
# In[5]:
import setGPU
#os.environ['CUDA_VISIBLE_DEVICES'] = '1'
from training_base import training_base
from Losses import loss_NLL
import sys
args = MyClass()
args.inputDataCollection = '/cms-sc17/convert_20170717_ak8_deepDoubleB_db_sv_train_val/dataCollection.dc'
args.outputDir = 'train_deep_sv_removals_test'
TrainBool = True
EvalBool = True
train = training_base(testrun=False, args=args)
if TrainBool:
#also does all the parsing
if not train.modelSet():
from DeepJet_models_removals import deep_model_removal_sv as model
from DeepJet_models_removals import Slicer1D
train.setModel(model)
train.compileModel(learningrate=0.001,
loss=['categorical_crossentropy'],
metrics=['accuracy'])
model, history, callbacks = train.trainModel(nepochs=2,
from training_base import training_base
from MultiDataCollection import MultiDataCollection
from pdb import set_trace
#also does all the parsing
train = training_base(testrun=False, collection_class=MultiDataCollection)
print 'Inited'
sizes = train.train_data.sizes
norm = float(
sizes[2]) / sizes[1] #normalization because samples have different sizes
train.train_data.setFlags([[1, 0], [0, norm], [0, 1]])
train.train_data.addYs([[0], [1], [0]])
evt = train.train_data.generator().next()
set_trace()
train.val_data.setFlags([[1, 0], [0, norm], [0, 1]])
train.val_data.addYs([[0], [1], [0]])
if not train.modelSet():
from models import dense_model_gradientReversal
print 'Setting model'
train.setModel(dense_model_gradientReversal, dropoutRate=0.1)
train.compileModel(
learningrate=0.003,
loss=['categorical_crossentropy', 'binary_crossentropy'],
#loss_weights=[1., 0.000000000001],
metrics=['accuracy'])
model, history = train.trainModel(nepochs=50,
batchsize=5000,
from training_base import training_base
from Losses import loss_NLL
#also dows all the parsing
train = training_base(testrun=True)
from models import convolutional_model_broad_map_reg
train.setModel(convolutional_model_broad_map_reg, dropoutRate=0.1)
train.compileModel(learningrate=0.005,
loss=['categorical_crossentropy', loss_NLL],
metrics=['accuracy'])
model, history = train.trainModel(nepochs=5,
batchsize=250,
stop_patience=300,
lr_factor=0.5,
lr_patience=10,
lr_epsilon=0.0001,
lr_cooldown=2,
lr_minimum=0.0001,
maxqsize=10)
from DeepJet_models_final import conv_model_final as trainingModel
from training_base import training_base
from eval_funcs import loadModel, makeRoc, _byteify, makeLossPlot, makeComparisonPlots, makeMetricPlots
trainDir = dayinfo + "_train" + opts.n
inputTrainDataCollection = trainDataCollection
inputTestDataCollection = testDataCollection
inputDataset = sampleDatasets_pf_cpf_sv
if TrainBool:
args = MyClass()
args.inputDataCollection = inputTrainDataCollection
args.outputDir = trainDir
#also does all the parsing
train = training_base(splittrainandtest=0.9, testrun=False, args=args)
if not train.modelSet():
train.setModel(trainingModel, inputDataset, removedVars)
train.compileModel(
learningrate=0.001,
loss=[
'binary_crossentropy'
], #other losses: categorical_crossentropy, kullback_leibler_divergence and many other in https://keras.io/losses/
metrics=['accuracy', 'binary_accuracy', 'MSE', 'MSLE'],
loss_weights=[1.])
model, history, callbacks = train.trainModel(nepochs=1,
batchsize=1024,
stop_patience=1000,
lr_factor=0.7,