0.4.12 released
ADAMS 0.4.12 has been released. More details and download links are available from here.
ADAMS 0.4.12 has been released. More details and download links are available from here.
Here is the final update for this year. Later on today, I will prepare a new release.
'Tis the season! ;-)
Well, actually mainly because of all the GUI improvements and the new adams-nlp module.
Fixes
Renaming variables, callable actors, events and undo/redo no longer result in the flow wobbling around (better queuing in Swing thread)
Improved the layout of the search panels, allowing for proper resizing and avoiding accidental hiding of search box
Changes
Plots now offer a Copy plot menu item from the right-click menu, to copy a screenshot of the plot to the clipboard
Count control actor now offers a monitor variable which allows resetting the internal count whenever the variable changes value
The Create callable actor menu item from the Flow editor's right-click menu now allows the creation callable actors under CallableActors, GridView and TabView.
Additions
Copy to clipboard target got added to the Send to functionality, which allows you to copy text, tables and graphical components (= screenshots) to the system's clipboard
TextOverlay added for simple text overlays in plots (eg SequencePlotter/SimplePlot); gets used by the Weka plugin for classifier errors to display key on lines drawn on screen
Stay tuned!
Last week, I gave a data mining workshop using ADAMS in Kuching, Malaysia. The material of the workshop is available from my github account.
I noticed that Windows is not particular great in regards to displaying unicode characters (e.g., emoticons). During my workshop we were also covering twitter sentiment analysis and showing all emoticons as little squares is not a very useful. My Linux Mint displayed them without any issues, on the other hand.
Thanks to these issues and the workshop, there have been a number of changes:
Fixes
Flow editor: annotations/quick info/input+output menu items in the View menu now get correctly updated when switching tabs
run/run.bat start scripts now specify the correct java-cup jar library in the CLASSPATH
Flow editor: Parameters tab now displays inline scripts in a more sensible format
Changes
added support for reading/writing flow files (.flow, .flow.gz format) with different file encodings (eg UTF-8/16)
Additions
adams-weka: added more string tokenizers and new token cleaner class hierarchy
adams-core: SwitchedSource actor allows you to output tokens based on boolean conditions, similar to the Switch control actor
adams-video: simple VLC-based video player got added by my student Steven Brown - requires VLC to be installed locally
added dialog for customizing fonts of widgets, e.g., to select more suitable font for displaying various unicode characters
the module adams-nlp has left the incubating stage and got moved to adams-addons
Have a good weekend!
Workshop in Kuching, Malaysia, on using ADAMS for big data analysis. More details here.
ADAMS 0.4.11 has been released. More details and download links are available from here.
Just a heads-up that I'll be making a new release tomorrow.
Since last week, only a few things changed:
fixed a flicker issue when processing videos in the flow and displaying the individual frames with a custom zoom in the ImageViewer (zoom got reset to 100% before changed to custom zoom)
adams-weka: added WekaGetCapabilities transformer to retrieve capabilities from classes like Filter, Classifier and Clusterer. The WekaCapabilitiesToInstances and WekaCapabilitiesToSpreadSheet conversion use this information. The former to generate random data, the latter to output the capabilities in tabular format.
the Flow editor now allows you to define keyboard actions (eg adding or enclosing actors using a specific actor) to speed up flow development (see manual "Keyboard actions")
Fingers crossed that tomorrow's release goes smoothly. :-)
It's this time of the week again and there have been a number of updates again.
Fixes
adams-core: the Inspect control actor no longer interprets skip as cancel
adams-core: the BreakPoint actor no longer locks the application when attempting to stop the flow inside a block that uses atomic execution (= finishedBeforeStopping option in Tee/Trigger/etc)
Changes
adams-core: the ImagePanel now allows undo/redo through the right-click popup menu as well
adams-imaging: Crop paint selection plugin offers undo/redo and also stores the x/y and width/height of the crop (in relation to the original image); PaintSelection supports undo/redo now too
option parsing of integer types now handles float-type strings (eg 11.0) as well, by parsing them as float and the retrieving the correct value - this is useful when dealing with numeric values stored in reports, as the default type there is float (even if you store an integer)
adams-meka: MEKA has been upgraded to 1.9.0
Index/Range (and spreadsheet/WEKA related classes) now allow enforcing of numeric indices by preceding them with "#" (eg "#13"); all names (columns/attributes/labels) can be enclosed with double quotes now
tool tips in the GenericObjectEditor have been improved; e.g., classes that implement ExampleProvider (eg Index/Range) now automatically append the example text
Additions
adams-core: new actor processors ListAllVariables and ListAllStorageNames
adams-weka: LeanMultiScheme is an extended version of Weka's MultiScheme meta-classifier with better memory management
adams-core: ReportToVariables and ReportToStorage transformers allow the transfer for multiple report values into variables/storage items in one go
adams-core: ModuleInfo source actor outputs information on the ADAMS module that make up the ADAMS application
Have a good weekend!
I'm running out of space on my piece of paper here, so I think it's time for an update. ;-)
Fixes
fixed some regressions in the flow editor due to changes relating to better Swing thread queuing
when copy/pasting actor with variable, variable disappeared
creating a callable actor didn't no update the tree immediately
the showProperties option of the ImageViewer sink now works inside the DisplayPanelManager as well
Changes
added the DATETIMEMSEC type to the DateTimeType enum
added DATETIMEMSEC support to the following conversions
ConvertDateTimeType
DateTimeTypeToString
ExtractDateTimeField
StringToDateTimeType
added DATETIMEMSEC support to the following actors
DateTimeTypeDifference
SpreadSheetConvertCells
SpreadSheetConvertHeaderCells
Flow editor: long files now get shortened for the tab; hovering over the tab reveals the full name
Cast control actor no longer throws a cast exception when using Unknown
data containers now list Notes and Reports in the debug inspection as well
adams-spreadsheet: SpreadSheetSetCell and SpreadSheetGetCell now support Row objects as well apart from SpreadSheet ones
adams-heatmap: the report table in the HeatmapDisplay can be hidden now
native ADAMS file choosers now offer a Filter text field which allows the display of files/dirs to be restricted to the ones that contain this substring (case-sensitive) - useful when looking for a single file or group of files in directories with 1000s of files
adams-weka: the WekaModelReader now has an option makeThreadSafe which attempts to make the loaded classifier thread-safe, if necessary (using the ThreadSafeClassifierWrapper mentioned below)
Additions
adams-heatmap: added the CountValues meta-generator
adams-heatmap: added TextOverlay for HeatmapDisplay sink
adams-core: added support for named counters, i.e., counters that count how often a string passes through
CounterInit (standalone/transformer) - initializes/resets a named counter in internal storage
CounterAdd (transformer) - increments the counter for the string representation of the token passing through
Counter (source) - outputs the names with their associated counts as spreadsheet
adams-weka: added an interface ThreadSafeClassifier that indicates that a classifier is thread-safe (Weka classifiers/filters aren't by design) and a generic ThreadSafeClassifierWrapper meta-classifier that attempts to make the wrapped classifier safer (there are no guarantees, though; if the code is bad by design, this won't help either)
Just a heads-up... There will be a new release coming up in the next couple of weeks (as part of my preparations for a data mining workshop end of November).
It's been a while since the last update on ADAMS, as I've been busy with our commercial projects and development mainly occurring in these frameworks. However, there were some changes that also affected the public ADAMS code:
Fixes
ConsoleWindow standalone now correctly displays the window
fixed launching Terminal on Windows (for Windows users: you know terminals as command-prompts)
StorageForLoop source now implements the VariableUpdater interface to make the variable usage check pass
Changes
ConsoleWindow now allows saving all the messages coming through to a file
Additions
adams-weka: the Veto meta-classifier favors the specified label if it is predicted by at least the specified number/percentage of classifiers in the ensemble. If not, then Vote with majority rule is used.
adams-weka: the SuppressModelOutput meta-classifier is useful if you want to suppress any large model output, e.g., from large ensembles
adams-weka: with the ThresholdedBinaryClassification meta-classifier you can move the default probability threshold of 0.5 to decide which label is selected to another value
adams-weka: the RemoveOutliers filter uses the outlier detection framework introduced here, removing all instances that get marked as outliers
Have a good weekend!
It's time to bring you up-to-date what's been happening under the hood of ADAMS lately. Well, for one thing, I got invited to give a demo at the BigMine 2015 workshop as part of KDD 2015 in Sydney. Eventually, a paper will be out.
The other major thing is that we definitely stick now with Java 8, mainly due to the better support of Java 8 in regards to file handling (java.nio.Files namely) and atomic moves of files/directories. I had the problem with a commercial product (and, of course, Windows) that files got move, but Windows still thought that they might be in use. Resulted in failures to process files. Rather annoying. However, moving to atomic moves, this problem has gone away now.
Fixes
AbstractChooserPanel and its derived component no longer doubles up the entry when pasting text into it.
PropertySheetPanelPage (used in wizards) now correctly updates the buttons when values change.
Fixed a performance issue in the Instance Explorer when selecting the color for many selected rows.
The Clear graphical output menu item in the Flow editor was a bit to aggressive in the cleanup and left the editor in a broken state (NullPointerExceptions galore).
Updates
The message option of the EnterManyValues source handles variables now as well, automatically expanding them
The storage panel, as used in the Breakpoint's user interface, now allows the export of selected objects, just like the object tree.
SpreadSheetTable/Model now allow you to toggle between cell value and cell type view. Sometimes it's easier to spot a string value where a number should be with this functionality (via popup menu).
Instance Explorer now allows you to use the value of an attribute as ID rather than the row number.
Tables now offer copy row/column as well through their popup menus.
HashSetAdd transformer now accepts spreadsheets and arrays of objects as well, for speeding up the add operation.
The CsvSpreadSheetReader received a performance boost option, which can be used for spreadsheets that have uniform column types: it allows you to specify a number of rows to use for identifying the column type automatically. The speedup on my laptop for a 2400 column by 2000 row spreadsheet with mainly floating values was 50% (30s instead of 60s).
Additions
WekaGeneticAlgorithm transformer added, which allows you to use the DarkLord genetic algorithm now also in the flow, not just through the wizard.
String conversion dialog added to the maintenance section (requires developer user mode), allows you to backquote/unbackquote etc strings.
RemoveTestInstances filter (package weka.filters.unsupervised.instance) allows you to remove rows from the data passing through (only first batch), by matching the IDs from a selected test set against the current data. Allows you to avoid manually removal of instances from a data set (hold out dataset) when training a classifier on the full dataset.
Have a great weekend!