You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
* skeleton of generic Neural Network
* proof of concept for training and prediction
* use ml5 callcallbac to handle callback or promisek
* added activation to options
* load save model
* adds placeholder to test and adds defaults
* DIY Neural Net Updates (#568)
* begins restructuring
* adds tfjs-vis and adds generic createModel() function
* adds tf viz and adds loadData function
* adds normalization function
* adds train function
* adds predict function
* work on DIY Neural Network (#569)
* fixing output and hidden activations
* moving data fields config to loadData()
Eventually this should be ml5.utils() or ml5.data()
* trimming whitespace
* rename labels to targets for training output
* using camelCase
* moves shuffle out of normalize
* [WIP] DIY Neural Net with focus on handling data internally (#571)
* adds data obj in class
* adds initialization step
* creates nn data class
* moves functions to data class
* working with titanic
* adds tf vis for training
* changes tf viz show on training is debug is true
* changes optimizer for regression
* adds unNormalize function
* fix normalization
* rm log
* refactors out to sep modules
* adds options to train
* adds options to train
* adds comments
* adds input/output array function
* removes i/o label check
* fixes optionOrCallback
* fix test
* adds comments
* updates data handling
* updates normalization data handling
* DIY NN data handling updates (#572)
* set up scaffold for creating inputs
* added output scaffolding - turn into functions todo
* adds function to encode data with ontHot or not
* ensure inputTypes/outputTypes exist
* [In progress] DIY NN - handles json and csv loading (#573)
* adds placeholders for json and csv loader'
* add json parsing
* adds true flag to normalization to fix broken function
* Fixes normalization function for NN (#575)
* adds possible fix - moves norm to external function
* adds correctly ordered data
* DIY Neural Network: Handles onehot() encoding for inputs and outputs in normalization (#576)
* adds correct input structure - accounts for onehot inputs
* code formatting and cleaning
* fixes need for onehot in classification for numeric output
* adds whileTraining cb support to train (#577)
* regression needs to pickup learning rate
* Adds temporary fixes for i/o values given to/output by .predict() (#578)
* adds min and max to meta and checks predict sample and forces array
* adds onehot encoding legend
* updating input handling for predict()
* adds label,confidence output for classification
* fix merge conflicts
* rm log
* need to use inputUnits and not input length
* [WIP] Fixes for before class (#581)
* changes for class
* values not value, but this does not match feature extractor regression
* [Diy nn] Fixes output reversal and gives unnormalized data as outputs of predict() (#582)
* unnormalize outputs for regression
* use .reverse() - note: we should use .unshift() where order matters
* code cleanup
* updates config inputs learning rate and code cleanup
* changed value to return result.outputs
* adds blob reader for json
* [diy nn] moves tf-vis visualizations to helper class (#583)
* adds tfvis to neuralNetworkVis.js
* adds tfvis to class
* handle text and convert to json (#584)
* allow user more control in the case of classification
* allow user more control in the case of classification (#585)
* Refactoring DIY Neural Network (#591)
* [wip] refactoring data handling and logic separation
* wip adding convertRawToTensor
* adds conversion to tensor
* adds normalization step
* adds train()
* adds predictInternal()
* fixes unique value mapping issue
* comment out .print()
* return all regression results
* adds experimental normalizationOptions object
* quick fix for checking if normalizationOptions exists
* commenting out auto tensor printing
* adds checks for normalizationOptions
* fix obj ref in conditional
* adds .normalizeData() - keeps .normalize() for now
* changes whileTraining function when debug is true
* adds layer options for adding more than 2 layers - experimentalgit add .
* remove .normalize() in favor of .normalizeData()
* adds save() and load()
* adds data.warmUp() to allow training without normalization - handled internally on train
* updates outputs of .predict() and .classify() to be array
* Divide by 0 issue when data is not normalized (#596)
* skipping calculating inputMin and inputMax when data not normalized
* adds check for this.data.meta.isNormalized in predictInternal
* Diy nn code cleanup (#606)
* adds input/output checks to specify number or array of input/output names
* moves model creation to training to compile after input and output units have been calcd
* cleanup code in .trainInternal()
* move vars up to top of function in .trainInternal()
* rm model creation from .createModelFromDataInternal()
* code cleanup and adds comments to .initializeIOUnits()
* code cleanup
* [DIY NN] Adds .saveData() and .loadData() (#607)
* adds saveData function
* updates saveData() with input checks
* adds loadData function
* [Diy NN] Adds .predictMultiple() for batch predictions/classifications (#609)
* adds predictMultiple()
* rm ys.print()
* moves bodypix and faceapi to preload support to match base branch
* [DIY nn] Neural Net Fixes loading pre-trained model (#613)
* adds model_meta.json to savedfiles
* adds loading of model_meta.json - temp fix
* use substring method for url checking
* adds check for if input is JSON
* loadData and text variables mixed up (#615)
* updates docs on loading model and adds contributors
* updates neural network docs with .load docs
* fix paths in examples
* adds nn src and adds placeholder
* If a string path to the `model.json` data object is given, then the `model.json`, `model_meta.json` file and its accompanying `model.weights.bin` file will be loaded. Note that the names must match.
410
410
* If InputFiles from html input `type="file"`. Then make sure to select ALL THREE of the `model.json`, `model_meta.json` and the `model.weights.bin` file together to upload otherwise the load will throw an error.
411
+
* Method 1: using a json object. In this case, the paths to the specific files are set directly.
* Method 2: specifying only the path to th model.json. In this case, the `model_meta.json` and the `model.weights.bin` are assumed to be in the same directory, named exactly like `model_meta.json` and `model.weights.bin`.
0 commit comments