def trainModel(self, cfg):
callbacks = [self.log, \
cb.MyModelCheckpointInterval(cfg), \
cb.MyLearningRateScheduler(cfg), \
cb.SaveLog2File(cfg), \
cb.PrintCallBack()]
if cfg.include_eval:
callbacks.append(cb.EvaluateTest(self.genTest, m.EvalResults, cfg))
if cfg.include_validation:
callbacks.append(cb.MyEarlyStopping(cfg))
callbacks.append(cb.MyModelCheckpointBest(cfg))
self.model.fit_generator(generator = self.genTrain.begin(), \
steps_per_epoch = self.genTrain.nb_batches, \
validation_data = self.genVal.begin(), \
validation_steps = self.genVal.nb_batches, \
epochs = cfg.epoch_end, initial_epoch=cfg.epoch_begin, callbacks=callbacks)
else:
self.model.fit_generator(generator = self.genTrain.begin(), \
steps_per_epoch = self.genTrain.nb_batches, \
verbose = 2,\
epochs = cfg.epoch_end, initial_epoch=cfg.epoch_begin, callbacks=callbacks)
# data
genTrain = DataGenerator(imagesMeta=data.trainGTMeta,
cfg=cfg,
data_type='train',
do_meta=False)
# models
Models = methods.AllModels(cfg, mode='train', do_hoi=True)
_, _, model_hoi = Models.get_models()
sys.stdout.flush()
#if False:
# train
callbacks = [callbacks.MyModelCheckpointInterval(cfg), \
callbacks.MyLearningRateScheduler(cfg), \
callbacks.MyModelCheckpointWeightsInterval(cfg),\
callbacks.SaveLog2File(cfg), \
callbacks.PrintCallBack()]
if cfg.dataset == 'TUPPMI':
model_hoi.fit_generator(generator = genTrain.begin(), \
steps_per_epoch = genTrain.nb_batches, \
verbose = 2,\
epochs = cfg.epoch_end, initial_epoch=cfg.epoch_begin, callbacks=callbacks)
else:
genTest = DataGenerator(imagesMeta=data.valGTMeta,
cfg=cfg,
data_type='test',
do_meta=False,