Config

class config.Config

In this class, we set the configuration parameters, adopt C library for data and memory processing. In the following, we train models and test models.

Getting Statistics of Dataset

Config.get_ent_total()

This method gets the entity total of knowledge base.

Config.get_rel_total()

This method gets the relation total of knowledge base.

Setting Configuration Parameters

Config.set_alpha(alpha)

This mothod sets the learning rate for gradient descent.

Parameters:alpha (float) – the learning rate.

Config.set_margin(margin)

This method sets the margin for the widely used pairwise margin-based ranking loss.

Parameters:margin (float) – margin for margin-based ranking function

Config.set_bern(bern)

This method sets the strategy for negative sampling.

Parameters:bern – “bern” or “unif”

Config.set_dimension(dim)

This method sets the entity dimension and relation dimension at the same time.

Parameters:dim (int) – the dimension of entity and relation.

Config.set_ent_dimension(dim)

This method sets the dimension of entity.

Parameters:dim (int) – the dimension of entity.

Config.set_rel_dimension(dim)

This method sets the dimension of relation.

Parameters:dim (int) – the dimension of relation.

Config.set_train_times(times)

This method sets the rounds for training.

Parameters:times (int) – rounds for training.

Config.set_nbatches(nbatches)

This method sets the number of batch.

Parameters:nbatches (int) – number of batch.

Config.set_work_threads(threads)

We can use multi-threading trainning for accelaration. This method sets the numebr of threads.

Parameters:threads (int) – number of working threads.

Config.set_ent_neg_rate(rate)

the number of negatives generated per positive training sample influnces the experiment results. This method sets the number of negative entities constructed per positive sample.

Parameters:rate (int) – the number of negative entities per positive sample.

Config.set_rel_neg_rate(rate)

This method sets the number of negative relations per positive sample.

Parameters:rate (int) – the number of negative relations per positive sample.

Config.set_lr_decay(lr_decay)

This method sets the learning rate decay for Adagrad optim method.

Parameters:lr_decay (float) – learning rate decay

Config.set_weight_decay(weight_decay)

This method sets the weight decay for Adagrad optim method.

Parameters:weight_decay (float) – weight decay for Adagrad.

Config.set_opt_method(method)

This method sets the optimizer for your model.

Parameters:optimizer – SGD Adagrad Adam and Adadelta can be chosen for optimizing.

Config.set_log_on(flag)

This method sets whether to log on the loss value.

Parameters:flag (bool) – if True, logs on the loss value when training.

Setting Inpath and Outpath

Config.set_in_path(path)

This method sets the path of benchmark.

Config.set_out_files(path)

This method sets where to emport embedding matrix.

Config.set_import_files(path)

Model paramters are exported automatically every few rounds. This method sets the path to find exported model parameters.

Parameters:path – path to automatically exported model parameters.

Config.set_export_files(path)

Model parameters will be exported to this path automatically.

Parameters:path – files that model parameters will be exported to.

Config.set_export_steps(steps)

This method sets that every few steps the model paramters will be exported automatically.

Parameters:steps (int) – Models will be exported via torch.save() automatically every few rounds

Saving and Loading Models

Config.save_pytorch()

This method saves the model paramters to self.exportName which was set by set_export_files().

Config.restore_pytorch()

This method restore model through torch.load

Config.export_variables(path=None)

This method export model paramters through torch.save.

Parameters:path – If None, this function euquals to save_pytorch(), else save paramters to path

Config.import_variables(path=None)

This method export model paramters through torch.load.

Parameters:path – If None, this function euquals to restore_pytorch(), else save paramters to path

Config.get_parameters(mode='numpy')

This method gets the model paramters.

Parameters:mode – if numpy, returns model parameters as numpy array, if list, returns those as list

Config.save_parameters(path=None)

This method save model parameters as json files when training finished.

Parameters:path – if None, save parameters to self.out_path which was set by set_out_files().

Setting Models

Config.set_model(model)

This method sets the traing model and optimizer method.

Parameters:model – training model. We can choose from :class:models.TransE :class:models.TransH :class:models.TransR :class:models.TransD :class:models.RESCAL :class:models.DistMult and :class:models.ComplEx

Config.sampling()

In this function, we choose positive samples and construct negative samples.

Training Models

Config.run()

In this function, we train the model

Testing Models

Config.set_test_flag(flag)

This method sets whether we test our model.

Parameters:flag (bool) – if True, we test the model.

Note

Note that test_flag must be set after all the other configuration parameters are set.

Config.test()

In this function, we test the model.