model.md

ReImproveJS

Globals / Model /

Class: Model

The Model class is handling everything concerning the neural network

Hierarchy

• Model

Index

Constructors

• constructor

Properties

• fitConfig
• model

Accessors

• FitConfig
• InputSize
• OutputSize

Methods

• addLayer
• compile
• export
• fit
• predict
• randomOutput
• FromNetwork
• loadFromFile

Constructors

constructor

+ new Model(config?: SequentialArgs, fitConfig?: ModelFitArgs): Model

Defined in reimprove/model.ts:48

The sequential config is truly optional and is to use only if you want to provide a complete tf.layers implementation of your model. Currently only dense layers are supported but convolutions etc will be implemented quickly. The [[ModelFitConfig]] is concerning the steps, steps per epoch etc ... which is how is the model going to train itself, which is handled by TensorFlowJS.

Parameters:

Name	Type	Description
config?	SequentialArgs	-
fitConfig?	ModelFitArgs

Returns: Model

---

Properties

fitConfig

● fitConfig: ModelFitArgs

Defined in reimprove/model.ts:48

---

model

● model: LayersModel

Defined in reimprove/model.ts:47

---

Accessors

FitConfig

● set FitConfig(fitConfig: ModelFitArgs): void

Defined in reimprove/model.ts:147

Parameters:

Name	Type
fitConfig	ModelFitArgs

Returns: void

---

InputSize

● get InputSize(): number

Defined in reimprove/model.ts:143

Returns: number

---

OutputSize

● get OutputSize(): number

Defined in reimprove/model.ts:139

Returns: number

---

Methods

addLayer

▸ addLayer(type: LayerType, config: LayerConfig): void

Defined in reimprove/model.ts:87

Method to just add a layer to the model, concatenating it with the previous ones.

deprecated Please now use NeuralNetwork

Parameters:

Name	Type	Description
type	LayerType	a type among DENSE, FLATTEN or CONV2D
config	LayerConfig	-

Returns: void

---

compile

▸ compile(config: ModelCompileArgs): Model

Defined in reimprove/model.ts:121

To compile the model, refer to [[ModelCompileConfig]] to know exactly what to use, but essentially, give the optimizer ('sgd', 'crossEntropy' , ...) and the loss function ('meanSquaredError', ...), see TFJS's documentation for the exhaustive list.

Parameters:

Name	Type
config	ModelCompileArgs

Returns: Model

---

export

▸ export(destination: string, place: string): Promise<void>

Defined in reimprove/model.ts:77

Export model to as destination.

Parameters:

Name	Type	Default	Description
destination	string	-	Can be one of 'downloads' (triggers browser download) [default], 'localstorage', 'indexeddb' or in http request 'http', 'https'.
place	string	"downloads"	-

Returns: Promise<void>

---

fit

▸ fit(x: Tensor, y: Tensor): Promise<any>

Defined in reimprove/model.ts:130

Parameters:

Name	Type
x	Tensor
y	Tensor

Returns: Promise<any>

---

predict

▸ predict(x: Tensor, config?: ModelPredictConfig): Result

Defined in reimprove/model.ts:126

Parameters:

Name	Type
x	Tensor
config?	ModelPredictConfig

Returns: Result

---

randomOutput

▸ randomOutput(): number

Defined in reimprove/model.ts:134

Returns: number

---

Static FromNetwork

▸ FromNetwork(network: NeuralNetwork, fitConfig?: ModelFitArgs, name: string): Model

Defined in reimprove/model.ts:160

Static method to create a Model from a NeuralNetwork. The fit config is optional as well as the name. It returns a prepared model, but not compiled.

constructor

Parameters:

Name	Type	Default
network	NeuralNetwork	-
fitConfig?	ModelFitArgs	-
name	string	v4()

Returns: Model

---

Static loadFromFile

▸ loadFromFile(file: string | object): Promise<Model>

Defined in reimprove/model.ts:62

Parameters:

Name	Type
file	string | object

Returns: Promise<Model>

---