Releases: marcellejs/marcelle
Releases · marcellejs/marcelle
v0.5.0
modelParameters
was improved to support other typesgenericChart
was improved to support lazy iterators (e.g. for datasets)- New component
onnxModel
using onnruntime-web with a wasm backend by default - New widgets (components):
number
andnumberArray
textField
was updated totextInput
and its API was updated ($value
stream instead of$text
)button
,text
,toggle
,select
: minor API changes (simplified constructors)slider
was improved with better defaults for pips and a continuous mode
v0.4.0
Breaking Changes ⚠️
- The dataset interface has changed significantly. Refer to the documentation for upgrading:
- The interface for models'
.train()
method has changed.train
now accepts either a Dataset or an iterable for training, where instances are objects withx
andy
properties for input and output, respectively. BatchPrediction.predict()
now accepts datasets or iterables.- data stores now take a location as argument rather than a configuration object
Many components and methods have been renamed for consistency. Any application built with a previous version of Marcelle is expected to break. The online documentation is up to date with the current naming. The list of name changes is as follows:
Previous name | New Name |
---|---|
Module | Component |
ModuleBase | ViewContainer |
Dashboard.start() | Dashboard.show() |
Dashboard.destroy() | Dashboard.hide() |
DashboardPage.useLeft() | DashboardPage.sidebar() |
Wizard.start() | Wizard.show() |
Wizard.destroy() | Wizard.hide() |
Wizard.step() | Wizard.page() |
Button.$down | Button.$pressed |
textfield | textField |
chart | genericChart |
sketchpad | sketchPad |
mlp | mlpClassifier |
knn | knnClassifier |
onnx | onnxModel |
tfGenericModel | tfjsModel |
mobilenet | mobileNet |
parameters | modelParameters |
classificationPlot | confidencePlot |
visObjectDetection | detectionBoxes |
New Features
- Lazy iterables have been added to process service data, in particular dataset data (TODO: add link to documentation)
- Services now support a
$distinct
query parameter to get the list of distinct values for a particular field. - K-Means clustering
- Styling no longer pollutes global CSS. Tailwind's Preflight was removed so that Marcelle does not interact with top-level styles when integrated in another application.
Internal Changes
- The core UI components have been updated
- Storybook has been added to support the development of core UI components
- The performance of KNN and DatasetBrowser has been improved
- [style] ESLint configuration has been improved
- [style] For-of loops are used instead of forEach
v0.2.0
Breaking changes
tfImageClassifier
has been replaced bytfGenericModel
- Module
name
property has been replaced bytitle
- Modules' mount method now accepts an HTML Element rather than a selector
imageDrop
has been renamed toimageUpload
browser
has been renamed todatasetBrowser
progress
has been renamed totrainingProgress
confusion
has been renamed toconfusionMatrix
- Dashboards now have a
closable
option, which is disabled by default. - Dataset's
$created
stream has been replaced by a more generic$changes
stream
New Features
- Mobilenet and COCO-SSD are now cached in the browser's storage for improved network performance
- Model Storage: models can be saved and loaded from datastores. They can be automatically synchronized with a datastore. File import/export is also possible
- Dataset upload has been added
- Dashboard now expose $active and $page streams to monitor its state
- Dashboard settings have been improved (TODO: give details + breaking changes)
trainingPlot
: The logs to display can be be configured in the module's options. It is possible to pass either a scalar (incremental mode) an array for each log.imageUpload
now supports target image dimensions and automatically crops and resizes imagesdatasetBrowser
now allows the selection, deletion or class change of instances
New Modules
tfGenericModel
fileUpload
imageUpload
onnxImageClassifier
Deprecated Modules
tfImageClassifier
imageDrop
Performance Improvements
- Datasets have been updated for improved performance
- Memory management related to TFJS has been improved
Bug Fixes
webcam
now supports multiple video devices- Quite a few others...
Internal Changes
- Build system update
- Examples now use vite
- Svelte components now use Typescript
- Module's
description
was unused and has been removed - Data stores have a
$services
stream and have been simplified - The architecture for models has been improved