def marius():
# initialize Marius parameters to those specified in configuration file or default
# values by calling parseConfig(config_path)
config_path = "examples/training/configs/fb15k_cpu.ini"
config = m.parseConfig(config_path)
# initialize the training and evaluation sets
train_set, eval_set = m.initializeDatasets(config)
# initialize the model with the encoder and decoder type specified in the config
# Note: if encoder and decoder are already initialized, you can initialize the model
# with a constructor call, i.e. model = m.Model(encoder, decoder). See other Python
# example file (custom_models_example.py) for how to define custom models
model = m.initializeModel(config.model.encoder_model,
config.model.decoder_model)
# initialize the trainer and evaluator using the data and model
trainer = m.SynchronousTrainer(train_set, model)
evaluator = m.SynchronousEvaluator(eval_set, model)
# train and evaluate
for epoch in range(config.training.num_epochs):
trainer.train(1)
evaluator.evaluate(True)
def marius():
# initialize Marius parameters to those specified in configuration file or default
# values by calling parseConfig(config_path)
config_path = "examples/training/configs/fb15k_cpu.ini"
config = m.parseConfig(config_path)
# Here we define a custom model to use during the training and evaluation process
# A Model consists of both 1) Encoder and 2) Decoder
# A Decoder consists of 1) Comparator, 2) Relation Operator, and 3) Loss Function
# initialize the encoder to Empty
encoder = m.EmptyEncoder()
# initialize the loss function to built-in SoftMax
loss_function = m.SoftMax()
# initialize custom Relation Operator and Comparator implemented in Python
rel_op = translation()
comp = PyDotCompare()
# comp = L2()
# initialize the decoder with our Loss Function and custom Relation Operator and Comparator
decoder = m.LinkPredictionDecoder(comp, rel_op, loss_function)
# initialize the custom model with our Encoder and Decoder
custom_model = transE(encoder, decoder)
# model instantiation and train for one epoch
train_set, eval_set = m.initializeDatasets(config)
trainer = m.SynchronousTrainer(train_set, custom_model)
evaluator = m.SynchronousEvaluator(eval_set, custom_model)
trainer.train(1)
evaluator.evaluate(True)
def test_one_epoch(self):
#preprocess.fb15k(output_dir="output_dir/")
config_path = "examples/training/configs/fb15k_cpu.ini"
config = m.parseConfig(config_path)
train_set, eval_set = m.initializeDatasets(config)
model = m.initializeModel(config.model.encoder_model,
config.model.decoder_model)
trainer = m.SynchronousTrainer(train_set, model)
evaluator = m.SynchronousEvaluator(eval_set, model)
trainer.train(1)
evaluator.evaluate(True)
def test_comparator(self):
#preprocess.fb15k(output_dir="output_dir/")
config_path = "examples/training/configs/fb15k_cpu.ini"
config = m.parseConfig(config_path)
encoder = m.EmptyEncoder()
comp = PyDotCompare()
rel_op = m.HadamardOperator()
loss_function = m.SoftMax()
decoder = m.LinkPredictionDecoder(comp, rel_op, loss_function)
model = m.Model(encoder, decoder)
train_set, eval_set = m.initializeDatasets(config)
trainer = m.SynchronousTrainer(train_set, model)
evaluator = m.SynchronousEvaluator(eval_set, model)
trainer.train(1)
evaluator.evaluate(True)