Turi Create 5.7

New Features

• Python 3.7 support on macOS (#788)

Feature Enhancements

• Increased performance for XGBoost (#2033)
• Added an option to inline render the SFrame .explore() method for greater Jupyter Notebook compatibility (#929)

Bug Fixes

• Reduced disk footprint of One-Shot Object Detector (#2169)
• Reduced memory usage of One-Shot Object Detector (#2191)
• Fixed broken test CoreMLExportTest on macOS 10.13 (#1332)
• Increased default max_iterations value for Drawing Classifier toolkit (#2092)
• Fixed broken links in API docs (#2112)
• Improved error message for One-Shot Object Detector when using empty arrays (#2002)
• Toolkits now better handle missing values in SFrames (#1994)
• Fixed how unpack and dropna lazily compute (#2170)
• Fixed type checking for max_iterations in Style Transfer toolkit (#2013)

Notes

• One-Shot Object Detector is now listed as a beta toolkit.

Turi Create 5.6

New Features

• New Toolkit: One Shot Object Detection (#1790)
• New UI: Image Classifier & Drawing Classifier Evaluation (#1335)

Feature Enhancements

• Annotation: similar images are automatically suggested for annotation (#1868)
• Drawing Classifier: added annotate method (#2028)

Bug Fixes

• Annotation: Ignore existing undefined labels when annotating (#2008)
• Sound Classifier - consistent logging and bug fix (#1966)
• Sound Classifier - put sample rate in Core ML metadata (#1908)
• Drawing Classifier - better logging and error message. Bug fix: reported training time (#1967)
• Drawing Classifier - better swift code in user guide for using Core ML model. (#1969)
• Drawing Classifier - API doc fixes (#1988)
• Visualization: Fix null value support in .explore (#1923)
• Visualization: Fix key press handler in tcviz app (#1910)
• Faster buffering for Activity Classifier (#1819)
• Better recommender updates (#1844)
• Fix CoreML export for tree classifiers (#1832)
• Add confusion and precision recall matrix to training/validationMetrics (stats after training) for C-API activity classifier (#1807)

Turi Create 5.5.1

Bug Fixes

• Fixes error image_classifier #1866
• Fixes regression in TC image classifier evaluate #1889
• Fixes import broken in 5.4 on linux GPUs for OD #1890

Note: the API surface is not impacted by this release, so for documentation, please see release 5.5.

Turi Create 5.5

New Features

• Added Ubuntu 14.04 compatibility #1441
• Refined output images from Style Transfer model #666

Feature Enhancements

• Object Detector will now compute training and validation metrics automatically #1726
• Added an Advanced Usage page to documentation #1639
• Added ability to navigate to arbitrary image during Image Annotation #1521
• Added ability to tune neural network structure for Sound Classifier #1477
• Now targeting iOS 10 as minimum deployment target #1654
• Sound classifier now saves deep features #1601

Bug Fixes

• Fixed crash related to Sound Classifier analyzing clips less than 975 milliseconds in length #1702
• Fixed issue with Drawing Classifier warm start options #1808
• Fixed compiler warning causing failing ninja builds #1667
• Fixed issue with histogram axis labels #1664
• Fixed issue with documentation formatting #1648
• Added missing documentation for .annotate method #1645
• Fix issue with Drawing Classifier infinite loop #1620
• Clarified Drawing Classifier error message #1617
• Fixed issue with UI selection for model evaluation #1564
• Fixed crash with Int labels when using model evaluation #1561
• Fixed issue with model evaluation window focus #1548
• Fixed issue with Image Classifier image column name expectations #1535
• Fixed issue related to Image Classifier model evaluation #1510
• Added ability for SFrame filter_by to work with map data type #1140
• Image deep feature extractor model downloads now check return code #1033
• Histogram of ints now has integer-aligned bin boundaries #167

Turi Create 5.4

New Features

• New Toolkit: Drawing Classifier #1436
• New Toolkit: Sound Classifier #1451
• New Annotation Tool #1482

Feature Enhancements

• CMake simplification & build speedup #1413
• Improvements to LBFGS Solver. #1372
• Replace Armadillo with Eigen #1402

Bug Fixes

• SFrame.show crashes if there are 0 effective columns #1505
• Exception when importing TuriCreate on a machine with no GPUs #1405
• Handle exception thrown when calling nvidia-smi when no GPU is installed #1406
• Killed upon reaching iteration limit #1075
• Linear Regression Gives Incorrect Error Message with Image Column - #1298
• Per-class accuracy doesn't make any sense #1407
• Fix issue with documentation failing to display generated images #1595
• SFrame.explore() throws AssertionError on a column with a list of list of dictionaries. #1503
• Scatter plot and heat map start axes at 0 #1194
• six dependency is too rigid #1417
• Logistic Classifier: disable probability output type for multiclass predict #1414
• Recommender Exclude Parameter Not Working #1164
• make_wheel.sh creates wheels that are not pip installable #927
• gl_sframe cache invalidation bug #1422
• Fixes S3 write path for SFrames #1416

Turi Create 5.3.1

Installation

To install Turi Create 5.3.1 into your virtual environment, use:

(venv) pip install turicreate==5.3.1

Fixes

• Write support for S3 #57

Documentation and other downloads are not affected by this release; only .whl files have changed.

Turi Create 5.3

Installation

To install Turi Create 5.3 into your virtual environment, use:

(venv) pip install turicreate==5.3

Enhancements

• Handle null & missing values in categorical heatmap (#1331)
• Handle null values in numeric histogram (#1326)
• Handle null values in categorical histogram (#1325)

Fixes

• Support empty labels in plots (#1211)
• More robust HDFS search paths (#1234)
• Integrate progress printing into os_log on Apple platforms (#1240)
• Corrected precision-recall computation in recommender.util.precision_recall_by_user (#1262)
• CSV parsing improvements, including support for fields that contain newlines (#1263, #1266)
• Improved error reporting when loading JPEG files with unsupported color spaces (#1279)
• Support logistic classification training options in image classification (#1299)
• Support custom IOU threshold for non-maximum suppression in object detector prediction (#1333)
• Fix axis title bug in Box & Whisker plot (#1363)

Turi Create 5.2.1

Installation

To install Turi Create 5.2.1 into your virtual environment, use:

(venv) pip install turicreate==5.2.1

Fixes

• Fix plot get_data for SFrame summary view #1252

Note: artifacts other than Python wheels are omitted from this release, since only the Python wheel content has changed. For all other artifacts, use release 5.2.

Turi Create 5.2

Installation

To install Turi Create 5.2 into your virtual environment, use:

(venv) pip install turicreate==5.2

Fixes

• Clarify sample code to avoid erroneous bounding boxes when deploying object detectors (#1016)
• Restore object detector support for RGBA .PNG files; throw an error for unsupported color spaces (#1046)
• Add SArray.filter_by (#1061)
• Support dictionaries with keys that are tuples (#1096)
• Support for nd-array features in supervised learning (#1127)
• Fix heat map rendering (#1129)
• Resolve JSON parsing bug for JSON input of exactly 128 characters (#1134)
• Fix box plot rendering (#1179)
• Allow object detector export to set confidence threshold to 0 (#1208)
• No longer require npm when building from source (#1213)

Turi Create 5.1

Installation

To install Turi Create 5.1 into your virtual environment, use:

(venv) pip install turicreate==5.1

Fixes

• Add auto-detection of JSON file format for SFrame.save (#802)
• Allow SArray initializer to convert an existing SArray to a different type (#1070)
• Allow SFrame.add_column to accept iterables and to fill with a constant value (#1072)