Skip to content

Enhance NeuralNetworkBuilder #998

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
ishii-norimi merged 1 commit into from
Sep 13, 2025
Merged

Enhance NeuralNetworkBuilder #998

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
188 changes: 99 additions & 89 deletions js/neuralnetwork_builder.js
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,31 +2,89 @@ import * as opt from '../../lib/model/nns/optimizer.js'

const layerTypes = {
abs: {},
clip: { min: 0, max: 1 },
conv: { kernel: 5, channel: 16 },
dropout: { drop_rate: 0.5 },
acos: {},
acoh: {},
asin: {},
asinh: {},
atan: {},
atanh: {},
bdaa: { alpha: { type: 'number', default: 1, multipleOf: 0.1 } },
bent_identity: {},
blu: { beta: { type: 'number', default: 0.1, multipleOf: 0.1 } },
brelu: { a: { type: 'number', default: 1, multipleOf: 0.1 } },
ceil: {},
celu: { a: { type: 'number', default: 1, multipleOf: 0.1 } },
clip: {
min: { type: 'number', default: 0, multipleOf: 0.1 },
max: { type: 'number', default: 1, multipleOf: 0.1 },
},
cloglog: {},
cloglogm: {},
conv: { kernel: { type: 'number', default: 5 }, channel: { type: 'number', default: 16 } },
cos: {},
cosh: {},
crelu: {},
dropout: {
drop_rate: { type: 'number', label: 'Drop rate', default: 0.5, multipleOf: 0.1, minimum: 0, maximum: 1 },
},
eelu: {
k: { type: 'number', default: 1, multipleOf: 0.1 },
alpha: { type: 'number', default: 1, multipleOf: 0.1 },
beta: { type: 'number', default: 1, multipleOf: 0.1 },
},
elish: {},
elliott: {},
elu: { a: { type: 'number', default: 1, multipleOf: 0.1 } },
erelu: {},
erf: {},
eswish: { beta: { type: 'number', default: 1, multipleOf: 0.1 } },
exp: {},
felu: { alpha: { type: 'number', default: 1, multipleOf: 0.1 } },
flatten: {},
full: { size: 10, a: 'sigmoid' },
floor: {},
frelu: { b: { type: 'number', default: 0, multipleOf: 0.1 } },
full: {
out_size: { type: 'number', label: 'Output size', default: 10, minimum: 1, maximum: 100 },
activation: {
type: 'string',
label: 'Activation',
default: 'sigmoid',
enum: [
'sigmoid',
'tanh',
'relu',
'leaky_relu',
'softsign',
'softplus',
'identity',
'polynomial',
'abs',
'gaussian',
'softmax',
],
},
},
function: { func: { type: 'string', default: '2*x' } },
gaussian: {},
leaky_relu: { a: 0.1 },
gelu: {},
leaky_relu: { a: { type: 'number', default: 0.1, multipleOf: 0.1, minimum: 0, maximum: 1 } },
identity: {},
log: {},
mean: { axis: 0 },
mean: { axis: { type: 'number', default: 0, minimum: 0, maximum: 10 } },
negative: {},
relu: {},
reshape: { size: [1, 1] },
reshape: { size: { type: 'array', default: [1, 1] } },
sigmoid: {},
softmax: {},
softplus: {},
softsign: {},
sparsity: { rho: 0.02 },
sparsity: { rho: { type: 'number', default: 0.02, multipleOf: 0.01 } },
square: {},
sqrt: {},
sum: { axis: 0 },
sum: { axis: { type: 'number', default: 0, minimum: 0, maximum: 10 } },
tanh: {},
transpose: { axis: [1, 0] },
variance: { axis: 0 },
transpose: { axis: { type: 'array', default: [1, 0] } },
variance: { axis: { type: 'number', default: 0, minimum: 0, maximum: 10 } },
}

const arrayAttrDefinition = {
Expand All @@ -49,22 +107,23 @@ const nnModelDefinition = {
const layers = Vue.ref([
{
type: 'full',
size: 10,
a: 'sigmoid',
poly_pow: 2,
out_size: 10,
activation: 'sigmoid',
},
])

const changeType = function (idx) {
const layer = { type: layers.value[idx].type, ...layerTypes[layers.value[idx].type] }
const layer = { type: layers.value[idx].type }
for (const [k, v] of Object.entries(layerTypes[layers.value[idx].type])) {
layer[k] = v.default
}
layers.value.splice(idx, 1, layer)
}
const addLayer = function () {
layers.value.push({
type: 'full',
size: 10,
a: 'sigmoid',
poly_pow: 2,
out_size: 10,
activation: 'sigmoid',
})
}

Expand All @@ -77,19 +136,7 @@ const nnModelDefinition = {
data: function () {
return {
layerTypeNames: Object.keys(layerTypes),
activations: [
'sigmoid',
'tanh',
'relu',
'leaky_relu',
'softsign',
'softplus',
'identity',
'polynomial',
'abs',
'gaussian',
'softmax',
],
layerTypes: layerTypes,
}
},
template: `
Expand All @@ -101,47 +148,24 @@ const nnModelDefinition = {
<select v-model="layer.type" v-on:change="changeType(i)">
<option v-for="type in layerTypeNames" :value="type">{{ type }}</option>
</select>
<template v-if="layer.type === 'clip'">
Min: <input v-model.number="layer.min" type="number" step="0.1">
Max: <input v-model.number="layer.max" type="number" step="0.1">
</template>
<template v-if="layer.type === 'conv'">
Kernel: <input v-model.number="layer.kernel" type="number">
Channel: <input v-model.number="layer.channel" type="number">
</template>
<template v-if="layer.type === 'dropout'">
Drop Rate: <input v-model.number="layer.drop_rate" type="number" min="0" max="1" step="0.1">
</template>
<template v-if="layer.type === 'full'">
Size: <input v-model.number="layer.size" type="number" min="1" max="100">
Activation: <select v-model="layer.a" v-on:change="$forceUpdate()">
<option v-for="a in activations" :value="a">{{ a }}</option>
</select>
<input v-if="layer.a === 'polynomial'" v-model.number="layer.poly_pow" type="number" min="1" max="10">
</template>
<template v-if="layer.type === 'leaky_relu'">
Alpha: <input v-model.number="layer.a" type="number" min="0" max="1" step="0.1">
</template>
<template v-if="layer.type === 'mean'">
Axis: <input v-model.number="layer.axis" type="number" min="0" max="10">
</template>
<template v-if="layer.type === 'polynomial'">
n: <input v-model.number="layer.n" type="number" min="0" max="10">
</template>
<template v-if="layer.type === 'reshape'">
Sizes: <array_attr v-model="layer.size" />
</template>
<template v-if="layer.type === 'sparsity'">
Rho: <input v-model.number="layer.rho" type="number" />
</template>
<template v-if="layer.type === 'sum'">
Axis: <input v-model.number="layer.axis" type="number" min="0" max="10">
</template>
<template v-if="layer.type === 'transpose'">
Axis: <array_attr v-model="layer.axis" />
</template>
<template v-if="layer.type === 'variance'">
Axis: <input v-model.number="layer.axis" type="number" min="0" max="10">
<template v-for="(aobj, attr) in layerTypes[layer.type]" :key="attr">
{{ aobj.label ?? attr }}
<template v-if="aobj.type === 'number'">
<input v-model.number="layer[attr]" type="number" :step="aobj.multipleOf" :min="aobj.minimum" :max="aobj.maximum">
</template>
<template v-if="aobj.type === 'string'">
<template v-if="aobj.enum">
<select v-model="layer[attr]">
<option v-for="a in aobj.enum" :value="a">{{ a }}</option>
</select>
</template>
<template v-if="!aobj.enum">
<input :value="layer[attr]" type="text">
</template>
</template>
<template v-if="aobj.type === 'array'">
<array_attr v-model="layer[attr]" />
</template>
</template>
<input type="button" value="x" v-on:click="layers.splice(i, 1)">
</div>
Expand All @@ -160,26 +184,12 @@ export default class NeuralNetworkBuilder {
}

get layers() {
const l = this._vue ? this._vue.$refs.layerselm.layers : [{ type: 'full', size: 10, a: 'sigmoid' }]
const r = []
for (let i = 0; i < l.length; i++) {
if (l[i].type === 'full') {
r.push({ type: 'full', out_size: l[i].size })
r.push({ type: l[i].a, n: l[i].poly_pow })
} else {
r.push(l[i])
}
}
return r
const l = this._vue ? this._vue.$refs.layerselm.layers : [{ type: 'full', out_size: 10, a: 'sigmoid' }]
return l.map(v => ({ ...v }))
}

get invlayers() {
const l = this.layers
const r = []
for (let i = l.length - 1; i >= 0; i -= 2) {
r.push(l[i - 1], l[i])
}
return r
return this.layers.concat().reverse()
}

get optimizer() {
Expand Down