The Altair Community is migrating to a new platform to provide a better experience for you. In preparation for the migration, the Altair Community is on read-only mode from October 28 - November 6, 2024. Technical support via cases will continue to work as is. For any urgent requests from Students/Faculty members, please submit the form linked here
"Generate Copy buggy?"
Maven
I have a process with a training and testing part. In the testing branch, I have:
Read Database --> Set Role (for ID) --> Numerical_to_Bimodal --> Apply Normalization (from training part)
--> Set Role (for Label) --> Apply Model (SVM Classifier) --> Performance_Bimodal_Classifier
I would like to keep the original data in the output data set for reference. So, I insert a 'Generate Copy (in Testing)' before the Numerical_to_Bimodal. This completely changes the performance result.
Here is what I don't understand - are these bugs or features?
Thanks for any help! Stefan
PS: This is on RM 5.0.001
Here is the code I refer to in above text, with the exception that for sake of simplicity, I added both copy operators. In above text in each case, just one of them was present. I also removed the two Read Database operators in front of the two SetRole operators
Read Database --> Set Role (for ID) --> Numerical_to_Bimodal --> Apply Normalization (from training part)
--> Set Role (for Label) --> Apply Model (SVM Classifier) --> Performance_Bimodal_Classifier
I would like to keep the original data in the output data set for reference. So, I insert a 'Generate Copy (in Testing)' before the Numerical_to_Bimodal. This completely changes the performance result.
Here is what I don't understand - are these bugs or features?
- Obviously, my testing dataset has now one attribute more (the copied one). Is this expected to create problems?
- I find that even connecting the 'ori' output of Generate Copy results in the same unexpected performance change...
- hence, I took Generate Copy out of the testing thread: I now have an IO multiplier and just look at the two outputs (exa, ori) of Generate Copy directly: Both contain the same copied attribute.l
- In additional testing, I inserted Generate Copy in the training flow. I now have the eventual label in two columns: Once as raw data, once to be converted to a bimodal label for training purposes. So, in effect, I have a perfect predictor for the SVM. However, I'm surprised on the result I get with libSVM: I don't see the new parameter in the weight table (so, it appears that the copied parameter hasn't made it to the SVM), but the weights of the other parameters have changed (so, it must have had some impact)
Thanks for any help! Stefan
PS: This is on RM 5.0.001
Here is the code I refer to in above text, with the exception that for sake of simplicity, I added both copy operators. In above text in each case, just one of them was present. I also removed the two Read Database operators in front of the two SetRole operators
<?xml version="1.0" encoding="UTF-8" standalone="no"?>
<process version="5.0">
<context>
<input>
<location/>
</input>
<output>
<location/>
<location/>
<location/>
<location/>
<location/>
</output>
<macros/>
</context>
<operator activated="true" class="process" expanded="true" name="Process">
<process expanded="true" height="791" width="882">
<operator activated="true" class="set_role" expanded="true" height="76" name="Set Role (3)" width="90" x="179" y="30">
<parameter key="name" value="die_id"/>
<parameter key="target_role" value="id"/>
</operator>
<operator activated="true" class="set_role" expanded="true" height="76" name="Set Role (2)" width="90" x="179" y="345">
<parameter key="name" value="die_id"/>
<parameter key="target_role" value="id"/>
</operator>
<operator activated="true" class="generate_copy" expanded="true" height="76" name="Generate Copy in Training" width="90" x="246" y="165">
<parameter key="attribute_name" value="32401_IFA_IM_6dB_L1_3_24_IF_IM_IFAD_P1_1_"/>
<parameter key="new_name" value="32401_original"/>
</operator>
<operator activated="true" class="numerical_to_binominal" expanded="true" height="76" name="Numerical to Binominal" width="90" x="313" y="30">
<parameter key="attribute_filter_type" value="single"/>
<parameter key="attribute" value="32401_IFA_IM_6dB_L1_3_24_IF_IM_IFAD_P1_1_"/>
<parameter key="max" value="73.0"/>
</operator>
<operator activated="true" class="normalize" expanded="true" height="94" name="Normalize" width="90" x="447" y="30"/>
<operator activated="true" class="set_role" expanded="true" height="76" name="Set Role" width="90" x="581" y="30">
<parameter key="name" value="32401_IFA_IM_6dB_L1_3_24_IF_IM_IFAD_P1_1_"/>
<parameter key="target_role" value="label"/>
</operator>
<operator activated="true" class="support_vector_machine_libsvm" expanded="true" height="76" name="SVM" width="90" x="715" y="30">
<parameter key="gamma" value="1.0E-4"/>
<parameter key="C" value="200.0"/>
<list key="class_weights"/>
</operator>
<operator activated="true" class="generate_copy" expanded="true" height="76" name="Generate Copy in Testing" width="90" x="313" y="480">
<parameter key="attribute_name" value="32401_IFA_IM_6dB_L1_3_24_IF_IM_IFAD_P1_1_"/>
<parameter key="new_name" value="32401_original"/>
</operator>
<operator activated="true" class="numerical_to_binominal" expanded="true" height="76" name="Numerical to Binominal (2)" width="90" x="380" y="345">
<parameter key="attribute_filter_type" value="single"/>
<parameter key="attribute" value="32401_IFA_IM_6dB_L1_3_24_IF_IM_IFAD_P1_1_"/>
<parameter key="max" value="73.0"/>
</operator>
<operator activated="true" class="apply_model" expanded="true" height="76" name="Apply Model (2)" width="90" x="447" y="165">
<list key="application_parameters"/>
</operator>
<operator activated="true" class="set_role" expanded="true" height="76" name="Set Role (4)" width="90" x="581" y="300">
<parameter key="name" value="32401_IFA_IM_6dB_L1_3_24_IF_IM_IFAD_P1_1_"/>
<parameter key="target_role" value="label"/>
</operator>
<operator activated="true" class="apply_model" expanded="true" height="76" name="Apply Model" width="90" x="715" y="165">
<list key="application_parameters"/>
</operator>
<operator activated="true" class="performance_binominal_classification" expanded="true" height="76" name="Performance (2)" width="90" x="715" y="300">
<parameter key="AUC (optimistic)" value="true"/>
<parameter key="precision" value="true"/>
<parameter key="recall" value="true"/>
<parameter key="false_positive" value="true"/>
<parameter key="false_negative" value="true"/>
<parameter key="true_positive" value="true"/>
<parameter key="true_negative" value="true"/>
</operator>
<operator activated="true" class="write_csv" expanded="true" height="60" name="Write CSV" width="90" x="782" y="480">
<parameter key="csv_file" value="C:\Users\eichenbe\Documents\Backup\Laptop\LiveCopy\Software\RapidMiner_5\SVC_3Lots.csv"/>
<parameter key="column_separator" value=","/>
</operator>
<connect from_op="Set Role (3)" from_port="example set output" to_op="Generate Copy in Training" to_port="example set input"/>
<connect from_op="Set Role (2)" from_port="example set output" to_op="Generate Copy in Testing" to_port="example set input"/>
<connect from_op="Generate Copy in Training" from_port="example set output" to_op="Numerical to Binominal" to_port="example set input"/>
<connect from_op="Numerical to Binominal" from_port="example set output" to_op="Normalize" to_port="example set input"/>
<connect from_op="Normalize" from_port="example set output" to_op="Set Role" to_port="example set input"/>
<connect from_op="Normalize" from_port="preprocessing model" to_op="Apply Model (2)" to_port="model"/>
<connect from_op="Set Role" from_port="example set output" to_op="SVM" to_port="training set"/>
<connect from_op="SVM" from_port="model" to_op="Apply Model" to_port="model"/>
<connect from_op="SVM" from_port="exampleSet" to_port="result 4"/>
<connect from_op="Generate Copy in Testing" from_port="example set output" to_op="Numerical to Binominal (2)" to_port="example set input"/>
<connect from_op="Numerical to Binominal (2)" from_port="example set output" to_op="Apply Model (2)" to_port="unlabelled data"/>
<connect from_op="Apply Model (2)" from_port="labelled data" to_op="Set Role (4)" to_port="example set input"/>
<connect from_op="Set Role (4)" from_port="example set output" to_op="Apply Model" to_port="unlabelled data"/>
<connect from_op="Apply Model" from_port="labelled data" to_op="Performance (2)" to_port="labelled data"/>
<connect from_op="Apply Model" from_port="model" to_port="result 1"/>
<connect from_op="Performance (2)" from_port="performance" to_port="result 2"/>
<connect from_op="Performance (2)" from_port="example set" to_op="Write CSV" to_port="input"/>
<connect from_op="Write CSV" from_port="through" 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"/>
<portSpacing port="sink_result 5" spacing="0"/>
</process>
</operator>
</process>
Tagged:
0
Answers
thank you for reporting that. That the original example set has been changed was an error and I corrected this behavior.
If you have one attribute twice in a data set and they are not both special attributes, they will be included in the analysis. Since a new attribute will change the distance between the examples, results might change. If you insert an already present regular attribute a second time, this attribute will receive a much higher importance in the analysis, because it influences the distance between objects twice as strong as other attributes.
I did not understand your last point completely. Would it be possible to separate this description and melt it down to the point? Your general description was very detailed and that's very good, but I think separating the effects would help me a lot in reproducing the problems
One final request: Please insert data generators instead of simply removing the input operators. This way I save the time to remove unnecessary operators, change parameter values where attribute names were mentioned. This time I will then be able to spend with fixing the bug...
Greetings,
Sebastian
thanks for your reply - clarifies things; so the only point open is the last one.
Looking at below code (which refers to your sample data sets), you'll see that the Kernel model refers to [a1] whereas the the copier creates [a1_numeric] - [a1] is actually the label.
So - the model is correct except for the attribute name (my decription from yesterday was wrong; since I gave the copied attribute a short name I missed to figure that the long name was actually present in the model parameter list, it just didn't stick out as it should have...) Regards, Stefan
PS: I removed the data generators because it was from a database anyway. Sorry for the inconvenience caused!
thank you for modifying the process, so that it run without problems.
I noticed now what you were referring to. It was not a bug, it was intended behavior. Instead of showing simply the attribute names, it showed the construction description of that attribute. And a copy is simply = the original attribute. Since this is a very misleading and confusing behavior, I changed it, so that the real name is show with the additional construction if different from name.
The change will come with the next update.
Greetings,
Sebastian