Log (AI Studio Core)

Synopsis

This operator stores information into the log table. This information can be almost anything including parameter values of operators, apply-count of operators, execution time etc. The stored information can be plotted by the GUI when the process execution is complete. Moreover, the information can also be written into a file.

Description

The Log operator is mostly used when you want to see the values calculated during the execution of the process that are otherwise not visible. For example you want to see values of different parameters in all iterations of any Loop operator. In such scenarios the ideal operator is the Log operator. A large variety of information can be stored using this operator. Values of all parameters of all operators in the current process can be stored. Other information like apply-count, cpu-time, execution-time, loop-time etc can also be stored. The information stored in the log table can be viewed in the Results View. The information can also be analyzed in form of various graphs using the Plot View in the Results Workspace. The information can also be written directly into a file using the filename parameter.

The log parameter is used for specifying the information to be stored. The column name option specifies the name of the column in the log table (and/or file). Then you can select any operator from the drop down menu. Once you have selected an operator, you have two choices. You can either store a parameter value or store other values. If you opt for the parameter value, you can choose any parameter of the selected operator through the drop down menu. If you opt for other values, you can choose any value like apply-count, cpu-time etc from the last drop down menu.

Each time the Log operator is applied, all the values and parameters specified by the log parameter are collected and stored in a data row. When the process finishes, the operator writes the collected data rows into a file (if the filename parameter has a valid path). In GUI mode, 2D or 3D plots are automatically generated and displayed in the Results Workspace. Please study the attached Example Processes to understand working of this operator.

Input

through (IOObject)
It is not compulsory to connect any object with this port. Any object connected at this port is delivered without any modifications to the output port. This operator can have multiple inputs. When one input is connected, another through input port becomes available which is ready to accept another input (if any). The order of inputs remains the same. The object supplied at the first through input port of the Log operator is available at the first through output port.

Output

through (IOObject)
The objects that were given as input are passed without changing to the output through this port. It is not compulsory to attach this port to any other port. The Log operator can have multiple outputs. When one output is connected, another through output port becomes available which is ready to deliver another output (if any). The order of outputs remains the same. The object delivered at the first through input port of the Log operator is delivered at the first through output port

Parameters

filenameThis parameter is used if you want to write the stored values into a file. The path of the file is specified here.
logThis is the most important parameter of this operator. It is used for specifying the values that should be stored by the Log operator. The log parameter is used for specifying the information to be stored. The column name option specifies the name of the column in the log table (and/or file). Then you can select any operator from the drop down menu. Once you have selected an operator, you have two choices. You can either store a parameter value or store other values. If you opt for the parameter value, you can choose any parameter of the selected operator through the drop down menu. If you opt for other values, you can choose any value like apply-count, cpu-time etc from the last drop down menu.
sorting typeThis parameter indicates if the logged values should be sorted according to the specified dimension.
sorting dimensionThis parameter is only available when the sorting type parameter is set to 'top-k' or 'bottom-k'. This parameter is used for specifying the dimension that is to be used for sorting.
sorting kThis parameter is only available when the sorting type parameter is set to 'top-k' or 'bottom-k'. Only k results will be kept.
persistentThis is an expert parameter. This parameter indicates if the results should be written to the specified file immediately.

Tutorial Processes

Introduction to the Log operator

This Example Process shows usage of the Log and Extract Macro operator in complex preprocessing. Other than concepts related to the Log operator, this process will cover a number of concepts of macros including redefining macros, macro of Loop Values operator and use of the Extract Macro operator. This process starts with a subprocess which is used to generate data. What is happening inside this subprocess is not relevant to the use of the Log operator, so it is not discussed here. A breakpoint is inserted after this subprocess so that you can view the ExampleSet. You can see that the ExampleSet has 12 examples and 2 attributes: 'att1' and 'att2'. 'att1' is nominal and has 3 possible values: 'range1', 'range2' and 'range3'. 'att2' has real values.

The Loop Values operator is applied on the ExampleSet. It iterates over the values of the specified attribute (i.e. att1) and applies the inner operators on the given ExampleSet while the current value can be accessed via the macro defined by the iteration macro parameter. The iteration macro parameter is set to 'loop_value', thus the current value can be accessed by specifying %{loop_value} in the parameter values. As att1 has 3 possible values, the Loop Values operator will iterate 3 times, once for each possible value of att1.

Here is an explanation of what happens inside the Loop Values operator. The Loop Values operator is provided with an ExampleSet as input. The Filter Examples operator is applied on it. The condition class parameter is set to 'attribute value filter' and the parameter string is set to 'att1 = %{loop_value}'. Note the use of the loop_value macro here. Only those examples are selected where the value of att1 is equal to the value of the loop_value macro. A breakpoint is inserted here so that you can view the selected examples. Then Aggregation operator is applied on the selected examples. It is configured to take the average of the att2 values of the selected examples. This average value is stored in a new ExampleSet in an attribute named 'average(att2)'. A breakpoint is inserted here so that you can see the average of the att2 values of the selected examples. The Extract Macro operator is applied on this new ExampleSet to store this average value in a macro named 'current_average'. The originally selected examples are passed to the Generate Attributes operator that generates a new attribute named 'att2_abs_avg'. This attribute is defined by the expression 'abs(att2 - eval(%{current_average}))'. Note the use of the current_average macro here. Value of the current_average macro is subtracted from all values of att2 and stored in a new attribute named 'att2_abs_avg'. The Resultant ExampleSet is delivered at the output of the Loop Values operator. A breakpoint is inserted here so that you can see the ExampleSet with the 'att2_abs_avg' attribute. This output is fed to the Append operator in the main process. The Append operator merges the results of all the iterations into a single ExampleSet which is visible at the end of this process in the Results Workspace.

Note the Log operator in the subprocess of the Loop Values operator. Three columns are created using the log parameter. The 'Average att2' column stores the value of the macro of the Extract Macro operator. The 'Iteration' column stores the apply-count of the Aggregate operator which is the same as the number of iterations of the Loop Values operator. The 'att1 value' column stores the value of att1 in the current iteration. At the end of the process, you will see that the Log operator stores a lot of information that was not directly accessible. Moreover, it displays all the required information at the end of the process, thus breakpoints are not required.

Optionally, you can use the filename parameter of the Log operator to store a text file holding the information the Log operator collected.

Here is what you see when you run the process:

The ExampleSet generated by the Generate Data subprocess. Then the process enters the Loop Value operator and iterates 3 times.

Iteration 1:

The ExampleSet where the 'att1' value is equal to the current value of the loop_value macro i.e. 'range1'
The average of the 'att2' values for the selected examples. The average is -1.161.
The ExampleSet with the 'att2_abs_avg' attribute for iteration 1.

Iteration 2:

The ExampleSet where the 'att1' value is equal to the current value of loop_value macro i.e. 'range2'
The Average of the 'att2' values for the selected examples. The average is -1.656.
The ExampleSet with the 'att2_abs_avg' attribute for iteration 2.

Iteration 3:

The ExampleSet where the 'att1' value is equal to the current value of loop_value macro i.e. 'range3'
The Average of the 'att2' values for the selected examples. The average is 1.340.
The ExampleSet with the 'att2_abs_avg attribute' for iteration 3.

Now the process comes out of the Loop Values operator and the Append operator merges the final ExampleSets of all three iterations into a single ExampleSet that you can see in the Results Workspace.

Now have a look at the results of the Log operator. You can see all the required values in tabular form using the Table View. You can see that all the values that were viewed using breakpoints are available in a single table. You can see the results in the Plot View as well.

Viewing Training vs Testing error using the Log operator

The 'Weighting' is loaded using the Retrieve operator. The Loop Parameters operator is applied on it. The parameters of the Loop Parameters operator are set such that this operator loops 25 times. Thus its subprocess is executed 25 times. In every iteration, the value of the C parameter of the SVM(LibSVM) operator is changed. The value of the C parameter is 0.001 in the first iteration. The value is increased logarithmically until it reaches 100000 in the last iteration.

Have a look at the subprocess of the Loop Parameters operator. First the data is split into two equal partitions using the Split Data operator. The SVM (LibSVM) operator is applied on one partition. The resultant classification model is applied using two Apply Model operators on both the partitions. The statistical performance of the SVM model on both testing and training partitions is measured using the Performance (Classification) operators. At the end the Log operator is used to store the required results.

The log parameter of the Log operator stores four things. The iterations of the Loop Parameter operator are counted by apply-count of the SVM operator. This is stored in a column named 'Count'. The value of the classification error parameter of the Performance (Classification) operator that was applied on the Training partition is stored in a column named 'Training Error'. The value of the classification error parameter of the Performance (Classification) operator that was applied on the Testing partition is stored in a column named 'Testing Error'. The value of the C parameter of the SVM (LibSVM) operator is stored in a column named 'SVM C'. Also note that the stored information will be written into a file as specified in the filename parameter.

Run the process and turn to the Results View. You can see all the values in the Table View. This table can be used to study how classification errors in training and testing partitions behave with the increase in the value of the C parameter of the SVM(LibSVM) operator. To view these results in graphical form, switch to the Plot View. Select an appropriate plotter. You can use 'Series Multiple' plotter with 'SVM-C' as the 'Index Dimension'. Select 'Training Error' and 'Testing Error' in the 'Plot Series'. The 'scatter multiple' plotter can also be used. Now you can analyze how the training and testing error behaved with the increase in the parameter C.

Please note that since RapidMiner version 8.0, the Loop Parameters Operator has been updated to a) be parallel and b) log the parameter set and performance automatically. Please see the help of that operator for more information.

Categories

Versions