"How do I split the data into training, validation and testing subsets?"

Curious

How do I split the data into training, validation and testing subsets? (Not just training and testing)

MartinLiebig

Hi,

Split Data is your friend

BR,

Martin

MartinLiebig

Hi,

Split Data is your friend

BR,

Martin

varunm1

Hi @Curious

As @mschmitz informed you can split using split data operator. You can provide the ratio of splits like 0.7 for training, 0.1 for validation and 0.2 for testing. You can see the sample code. The order in which you give this ratio defines the order of outputs are well.

<?xml version="1.0" encoding="UTF-8"?><process version="9.1.000">
  <context>
    <input/>
    <output/>
    <macros/>
  </context>
  <operator activated="true" class="process" compatibility="9.1.000" expanded="true" name="Process">
    <parameter key="logverbosity" value="init"/>
    <parameter key="random_seed" value="2001"/>
    <parameter key="send_mail" value="never"/>
    <parameter key="notification_email" value=""/>
    <parameter key="process_duration_for_mail" value="30"/>
    <parameter key="encoding" value="SYSTEM"/>
    <process expanded="true">
      <operator activated="true" class="generate_data" compatibility="9.1.000" expanded="true" height="68" name="Generate Data" width="90" x="45" y="85">
        <parameter key="target_function" value="random"/>
        <parameter key="number_examples" value="100"/>
        <parameter key="number_of_attributes" value="5"/>
        <parameter key="attributes_lower_bound" value="-10.0"/>
        <parameter key="attributes_upper_bound" value="10.0"/>
        <parameter key="gaussian_standard_deviation" value="10.0"/>
        <parameter key="largest_radius" value="10.0"/>
        <parameter key="use_local_random_seed" value="false"/>
        <parameter key="local_random_seed" value="1992"/>
        <parameter key="datamanagement" value="double_array"/>
        <parameter key="data_management" value="auto"/>
      </operator>
      <operator activated="true" class="split_data" compatibility="9.1.000" expanded="true" height="124" name="Split Data" width="90" x="246" y="136">
        <enumeration key="partitions">
          <parameter key="ratio" value="0.7"/>
          <parameter key="ratio" value="0.1"/>
          <parameter key="ratio" value="0.2"/>
        </enumeration>
        <parameter key="sampling_type" value="automatic"/>
        <parameter key="use_local_random_seed" value="false"/>
        <parameter key="local_random_seed" value="1992"/>
      </operator>
      <connect from_op="Generate Data" from_port="output" to_op="Split Data" to_port="example set"/>
      <connect from_op="Split Data" from_port="partition 1" to_port="result 1"/>
      <connect from_op="Split Data" from_port="partition 2" to_port="result 2"/>
      <connect from_op="Split Data" from_port="partition 3" to_port="result 3"/>
      <portSpacing port="source_input 1" spacing="0"/>
      <portSpacing port="sink_result 1" spacing="0"/>
      <portSpacing port="sink_result 2" spacing="0"/>
      <portSpacing port="sink_result 3" spacing="0"/>
      <portSpacing port="sink_result 4" spacing="0"/>
    </process>
  </operator>
</process>

Thanks,
Varun

Telcontar120

But based on modern data science best practice, you should be using Cross Validation and not Split Validation...there are very few cases left (if any) where a simple split validation is better than cross validation for estimating future model performance.  Split validation provides a single point estimate based on one test sample only.  Cross validation uses all the data for training and testing.  

varunm1

Hi @Telcontar120

I just want to clarify if there is any use of validation set when we apply cross validation? I get this question a lot in deep learning when i skip validation set in training because I apply cross validation most of the time. As the main use of validation set is not to overfit during training but I think cross validation reduces over fitting as well. 

Thanks,
Varun

IngoRM

Hi,

If there is a cross validation as the most outer step including all preprocessing and modeling, then an additional validation set would indeed not be necessary.  However, this is not always feasible - most often for runtime reasons, sometimes the complexity of the processes gets a bit out of control.

In those cases, I would still keep some fraction of the original data (before I do anything to it!) as a validation set to make sure that I did not accidentally leak any information as part of my data processing.

Hope this helps,

Ingo

Curious

Thank you so much everyone!