MetaDataLoader

MetaDataLoader is the key component for transforming dataset into task form. As usual, we consider each user as a task. Here, a batch of data refers to a batch of tasks.

Overall, we extend AbstractDataLoader to MetaDataLoader. The extended modification can be listed briefly as following:

• [Add] self.transformToTaskFormat(): Generate Task dict from dataset.

• [Add] self.generateSingleTaskForTrain(uid,v): Format a single task.

• [Add] self.getTaskIdList(): Generate the list of task ids in this dataset.

• [Override] self._init_batch_size_and_step(): Initialize train_batch_size.

• [Override] self.pr_end(): Get the number of tasks.

• [Override] self._shuffle(): Shuffle the task.

• [Override] self._next_batch_data(): Generate a batch of tasks iteratively.

self.transformToTaskFormat()

This function is used to generate task dict from dataset. It will return ‘taskDict’ for this MetaDataLoader. During the process of this method, it will call self.generateSingleTaskForTrain(uid,v) to deal with a single task(user).

[return] finalTaskDict(dict)

A dict whose keys are user_id and values are corresponding Task object.

self.generateSingleTaskForTrain(uid,v)

We use this function to generate a task.

uid

uid from function transformToTaskFormat().

v

value from function transformToTaskFormat().

[return]

An object of class Task.

self.getTaskIdList()

This function can generate the list of task ids in this dataset.

[return] userlist (1D numpy.ndarray)

an 1D array of user list.

self._init_batch_size_and_step()

This function is used to initialize train_batch_size. The train_batch_size indicates the number of tasks for each training batch.

self.pr_end()

Get the number of tasks(users).

self._shuffle()

Shuffle the task.

self._next_batch_data()

This function is used to generate a batch of tasks iteratively. :[return] taskBatch(list): A list of task. The length is train_batch_size.