Delphi modeling 1 #499
Merged
Delphi modeling 1 #499
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
The external function should provide the derivative at each time point.
This is the dumbest model. Each root node predict close to the average of the observation series.
or any time series data.
…ation fault happening on old test code
1. All the programs compile. 2. All the tests pass. However all the executions are not correct. The head node modeling components are not properly integrated with CauseMos.
1. Need to do more: a. Cleaning b. Testing
1. Remember the sequence of log likelihoods throughout the sampling process including burning. 2. Complete state now includes the log likelihood sequence. 3. Plotting function now plots the log likelihoods.
1. Added code to identify and retain the MAP sample even if we encounter during the burning period. 2. If we encounter the MAP sample within the burning period it is always put at res - 1 position of the retained samples and we reduce the number of samples retained during the sample resolution period by 1. 3. Finally delphi remembers the index of the MAP sample withing the retained samples. Note: There are some (or more) repeated code in this push and previous ones that needs to be cleaned up when there is more time.
* Accumulating log likelihoods 1. Remember the sequence of log likelihoods throughout the sampling process including burning. 2. Complete state now includes the log likelihood sequence. 3. Plotting function now plots the log likelihoods. * MAP Sample 1. Added code to identify and retain the MAP sample even if we encounter during the burning period. 2. If we encounter the MAP sample within the burning period it is always put at res - 1 position of the retained samples and we reduce the number of samples retained during the sample resolution period by 1. 3. Finally delphi remembers the index of the MAP sample withing the retained samples. Note: There are some (or more) repeated code in this push and previous ones that needs to be cleaned up when there is more time. * Fixed a place that gave a segmentation fault due to faulty logic * Fixed an initialization bug Co-authored-by: Manujinda Wathugala <manujinda@email.arizona.edu>
The seasonality was not visible when Delphi was invoked through the CauseMos interface. When a model is created, each concept’d period for seasonality is inferred from data. However, it was not saved into the serialized model. When creating an experiment, an AnalysisGraph object is created by deserializing the saved model. Since concept periods were not saved all the concepts ended up having the default period of 1.
* Accumulating log likelihoods 1. Remember the sequence of log likelihoods throughout the sampling process including burning. 2. Complete state now includes the log likelihood sequence. 3. Plotting function now plots the log likelihoods. * MAP Sample 1. Added code to identify and retain the MAP sample even if we encounter during the burning period. 2. If we encounter the MAP sample within the burning period it is always put at res - 1 position of the retained samples and we reduce the number of samples retained during the sample resolution period by 1. 3. Finally delphi remembers the index of the MAP sample withing the retained samples. Note: There are some (or more) repeated code in this push and previous ones that needs to be cleaned up when there is more time. * Fixed a place that gave a segmentation fault due to faulty logic * Fixed an initialization bug * Fixed a bug The seasonality was not visible when Delphi was invoked through the CauseMos interface. When a model is created, each concept’d period for seasonality is inferred from data. However, it was not saved into the serialized model. When creating an experiment, an AnalysisGraph object is created by deserializing the saved model. Since concept periods were not saved all the concepts ended up having the default period of 1. Co-authored-by: Manujinda Wathugala <manujinda@email.arizona.edu>
Added some methods to generate a random CAG and populate it with synthetic observations.
1. Removed commented code 2. Moved a block of code to make some repeated code identical in two places. I did not refactor that into a separate method since it is only a few lines of code and those few lines alone does not create a coherent functionality.
This caused a chunk of repeated code. I feel that using a separate class to handle the delphi.db is a better approach.
The initial implementation of generating a cyclic curve using a LDS based on Fourier analysis.
Updated the timing program a bit.
* Accumulating log likelihoods 1. Remember the sequence of log likelihoods throughout the sampling process including burning. 2. Complete state now includes the log likelihood sequence. 3. Plotting function now plots the log likelihoods. * MAP Sample 1. Added code to identify and retain the MAP sample even if we encounter during the burning period. 2. If we encounter the MAP sample within the burning period it is always put at res - 1 position of the retained samples and we reduce the number of samples retained during the sample resolution period by 1. 3. Finally delphi remembers the index of the MAP sample withing the retained samples. Note: There are some (or more) repeated code in this push and previous ones that needs to be cleaned up when there is more time. * Fixed a place that gave a segmentation fault due to faulty logic * Fixed an initialization bug * Fixed a bug The seasonality was not visible when Delphi was invoked through the CauseMos interface. When a model is created, each concept’d period for seasonality is inferred from data. However, it was not saved into the serialized model. When creating an experiment, an AnalysisGraph object is created by deserializing the saved model. Since concept periods were not saved all the concepts ended up having the default period of 1. * Generating synthetic CAGs Added some methods to generate a random CAG and populate it with synthetic observations. * Cleanup 1. Made the code more modularized 2. Removed some debug print lines * More cleaning 1. Removed commented code 2. Moved a block of code to make some repeated code identical in two places. I did not refactor that into a separate method since it is only a few lines of code and those few lines alone does not create a coherent functionality. * Created a class to perform timing * Fixed CMakeLists.txt * Changed a file name * A basic macro profiler completed. * Added some progress messages to the timer. * Added proper error checking for opening delphi.db This caused a chunk of repeated code. I feel that using a separate class to handle the delphi.db is a better approach. * Improved the newly created delphi.db open method to accept the database open mode. * Improved the Delphi macro timing program to accept command line options. * Removed inclusion of dbg.h in main source files. * Fixed the help option in timer. * The june embed jupytre notebook got opened and some version number got chaned * Seasonal curve from a Fourier series The initial implementation of generating a cyclic curve using a LDS based on Fourier analysis. * Commented out some unused synthetic data generating methods. Updated the timing program a bit. * Deleted commented code related to synthetic data generation. Co-authored-by: Manujinda Wathugala <manujinda@email.arizona.edu>
This should output a create model json file that would create the untrained model of the object the method export_create_model_json_string() is called upon. This method would be mostly used to output synthetic CAGs with synthetic data into create model json files so that we could run the training pipeline on those models in a freshly created object other than the object that was used to generate the synthetic CAG to prevent any ground truth parameter values leaking from the synthetic CAG generation process into the training process. To aid assess the goodness of training, it would be beneficial to store the ground truth parameters of the CAG that generated the data in a separate additional section of the create model json produced.
1. Removed some commented code. 2. Updated transition matrix and initial state to match the math I wrote in the WorldModelers document. 3. Tried to make the transition matrix matrix exponential compatible. It did not work.
Changed "independent node" to "head node"
Moved the timer.cpp app that performs macro timing of Delphi into apps folder from scripts folder. Adding the timing_analysis.py script that plots the timing results collected by timer.cpp app.
…delphi_modeling_1 Had to resolve some conflicts.
take_model_snapshot.py Takes a snapshot of a how Delphi works at a particular point of its implementation A snapshot taken before and a snapshot taken after a code change such as an optimization or refactoring could be used to verify whether the accuracy of the code got changed as a byproduct of the changes made to the code. If there is no change at all, the before and after snaps should be identical. take_model_snapshot.py Compares two snapshots of how Delphi worked at two points of its implementation A snapshot taken before and a snapshot taken after a code change such as an optimization or refactoring could be compared to verify whether the accuracy of the code got changed as a byproduct of the changes made to the code. If there is no change at all, the before and after snaps should be identical and the plots produced by this script should be a single bar at zero.
…delphi_modeling_1
plot produced by snap_diff.py script
…delphi_modeling_1 Added mean and standard deviation of the snapshot differences to the plot produced by snap_diff.py script
* Fully matrix exponential compatible periodic LDS. Implemented and debugged. Based on the Fourier decomposition. As per my tests, it is working. * Deleted a lot of commented code and print statements. * Did some more cleaning. Co-authored-by: Manujinda Wathugala <manujinda@email.arizona.edu>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Initial head node seasonal model implementation.
Initial prototype of the fully matrix exponential compatible periodic curve generation.
Delphi macro timing.
Regression testing.
Synthetic CAG and data generation.
Other code improvements.