Commit 8f06cd2d authored by Chris Coughlin's avatar Chris Coughlin

Updated to reflect new functionality

parent 3ab0175f
......@@ -319,6 +319,27 @@ The list of ROI is thinned out by removing ("suppressing") those ROI that share
A combination of non-maximum suppression and enhancement. As with conventional enhancement, tends to reduce background noise.
## Saving Results
### Individual Results
Datasets displayed in Desktop can be exported to disk as a conventional image. Double-click on an input file, ROI, or a compiled ROI result to display the data. For large datasets, Desktop will ask whether to display the dataset as-is or to attempt to subsample and scale.
Regardless of choice no permanent changes are made to the data; subsampling is used only to speed up display of the data.
To export the dataset, right-click anywhere on the dataset and select ``Save`` from the context menu. Desktop currently supports several image formats for exporting the data including BMP, JPG, GIF, TIFF, and PNG.
### Batch Results
To save all of the results of an analysis, press the ``Save`` button on the Results tab. Desktop will prompt for a destination folder and a display preference (Union, NMS, etc.). Desktop will compile the final ROI results for each input file and export as a delimited text file in the specified folder.
To save all the reported ROI, switch to the Reporting tab and press the ``Save`` button. Desktop will prompt for a destination file, and will write all the ROI results as text to this file. The text output is of the form
````
&source=E:\Big\Data\Myriad\test\190.csv&pyramid=pscale2pwsize1pstep1&op=com.emphysic.myriad.core.data.ops.SobelOperation@3cea4a82&window=xoff6yoff1w15h15&roifinder=com.emphysic.myriad.core.data.roi.PassiveAggressiveROIFinder@1ec9d294&roiorigin=0
&source=E:\Big\Data\Myriad\test\190.csv&pyramid=pscale2pwsize1pstep1&op=com.emphysic.myriad.core.data.ops.SobelOperation@3cea4a82&window=xoff6yoff2w15h15&roifinder=com.emphysic.myriad.core.data.roi.PassiveAggressiveROIFinder@1ec9d294&roiorigin=0
&source=E:\Big\Data\Myriad\test\190.csv&pyramid=pscale2pwsize1pstep1&op=com.emphysic.myriad.core.data.ops.SobelOperation@3cea4a82&window=xoff6yoff0w15h15&roifinder=com.emphysic.myriad.core.data.roi.PassiveAggressiveROIFinder@1ec9d294&roiorigin=0
&source=E:\Big\Data\Myriad\test\190.csv&pyramid=pscale2pwsize1pstep1&op=com.emphysic.myriad.core.data.ops.SobelOperation@3cea4a82&window=xoff4yoff3w15h15&roifinder=com.emphysic.myriad.core.data.roi.PassiveAggressiveROIFinder@1ec9d294&roiorigin=0
````
Where each line is a Region Of Interest, and includes information about the input file and the approximate position of the ROI within the input.
## References
1. Rauber, T. and Runger, G. [Parallel Programming for Multicore and Cluster Systems](http://www.springer.com/us/book/9783642378003) (Second Edition), Section 6.1.7. Springer: New York, 2013.
2. Iancu, Costin, et al. [Oversubscription on multicore processors](https://crd.lbl.gov/assets/pubs_presos/ovsub.pdf). Parallel & Distributed Processing (IPDPS), 2010 IEEE International Symposium on. IEEE, 2010.
......
......@@ -102,9 +102,9 @@ Each algorithm has its strengths and weaknesses; Emphysic recommends that each b
## Training
To create a new model, select an algorithm to use by choosing the appropriate tab in the Myriad Trainer interface e.g. create a new Passive Aggressive model by making the Passive Aggressive tab visible. Adjust the train:test ratios, sample balancing, and preprocessing operations (if any) as desired.
When the `Train` button is pressed, Myriad Trainer will attempt to read each of the folders specified. For each file in each of the folders, the trainer will attempt to load the contents as a Myriad dataset and if successful will assign the appropriate label to the dataset. When all the available data have been read, a random subset is set aside for testing and the remainder used to train the selected model. The trainer will produce a plot that visualizes a projection of the entire dataset in three dimensions, with red markers indicating positive samples and blue indicating negative. The projection is based on [Principal Component Analysis](https://en.wikipedia.org/wiki/Principal_component_analysis ).
When the `Train` button is pressed, Myriad Trainer will attempt to read each of the folders specified. For each file in each of the folders, the trainer will attempt to load the contents as a Myriad dataset and if successful will assign the appropriate label to the dataset. When all the available data have been read, a random subset is set aside for testing and the remainder used to train the selected model. At the same time, the trainer will produce a plot that visualizes a projection of the entire dataset in three dimensions, with red markers indicating positive samples and blue indicating negative. The projection is based on [Principal Component Analysis](https://en.wikipedia.org/wiki/Principal_component_analysis ).
For each sample in the training set, the model is sent a one-dimensional array of the sample’s data and a label indicating whether the sample is positive or negative. After the training is complete, Myriad Trainer tests the model’s accuracy by asking it to predict the category of each sample in the testing set.
For each sample in the training set, the model is sent a one-dimensional array of the sample’s data and a label indicating whether the sample is positive or negative. After the training is complete, Myriad Trainer tests the model’s accuracy by asking it to predict the category of each sample in the testing set. By default, Trainer will repeat this process 100 times to generate an average accuracy (% of correct calls) for the model.
If a model exists in memory and the `Train` button is pressed Myriad Trainer will ask whether the current model should undergo another round of training. If replacement is chosen, the previous model is scrapped and a new model is created. A model can be trained in as many or as few rounds of training as desired. During each round of training it is important that the same preprocessing and model configuration parameters are used.
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment