We split the training set into 10 roughly equal sized parts.
The random training set split tool (set test size for the data: 0.2) was then used to split the training set and test set.
We split the training dataset into half: the first half is for training the expert networks and the other half is for training the gating network.
To split the trains into these four different groups a different label value is assigned to each of the trains from their start points.
In order to perform unbiased comparisons, we split the training data into a smaller training set of 2000 utterances and a hold-out set of 1233 utterances.
Since the speaker identities were known, we could split the training data into several clusters (male interviewees, female interviewees and interviewers) before the actual discriminative training adaptation (DT--see [15] for details) to enhance the method's effectiveness.
To this end, we first split the training set into three equally sized subsets by randomly assigning training dataset molecules to one of the subsets (S1 and S2 consist of 59 molecules, S3 consists of 58 molecules).
In our recommendation task, there is only one group of features for all instances, thus we simply split the training data into two parts with bootstrap, namely ({mathbf{X}_{d_1},mathbf{Y}_{d_1}^0}) and ({mathbf{X}_{d_2},mathbf{Y}_{d_2}^0}), and then train the base MIML learner on these two parts and obtain the (f_{d_1}^0) and (f_{d_2}^0), respectively.
When there is only a small amount of data, HDFS needs to split the training data and store the distribution in the data nodes; therefore, the pre-processing time of MR-DIDC is far higher than the operation time of the algorithm and results in the lower total time efficiency of MR-DIDC compared to that of SPRINT.
Therefore, we split the training set into 3-folds.
