netron/source/nnabla.js

var nnabla = nnabla || {};
var protobuf = protobuf || require('./protobuf');
var text = text || require('./text');

nnabla.ModelFactory = class {

    match(context) {
        const identifier = context.identifier;
        if (identifier.endsWith('.nntxt')) {
            const tags = context.tags('pbtxt');
            if (tags.has('network')) {
                return 'nnabla.pbtxt';
            }
        }
        return undefined;
    }

    open(context, match) {
        return context.require('./nnabla-proto').then(() => {
            nnabla.proto = protobuf.get('nnabla').nnabla;
            switch (match) {
                case 'nnabla.pbtxt': {
                    const stream = context.stream;
                    const reader = protobuf.TextReader.open(stream);
                    const model = nnabla.proto.NNablaProtoBuf.decodeText(reader);
                    const promises = [
                        context.request('nnp_version.txt', null),
                        context.request('parameter.protobuf', null)
                    ];
                    const open = (model, version) => {
                        return context.metadata('nnabla-metadata.json').then((metadata) => {
                            return new nnabla.Model(metadata, model, 'NNabla' + (version ? ' v' + version : ''));
                        });
                    };
                    return Promise.all(promises).then((streams) => {
                        const version = text.Reader.open(streams[0]).read();
                        const reader = protobuf.BinaryReader.open(streams[1]);
                        const params = nnabla.proto.NNablaProtoBuf.decode(reader);
                        model.parameter = params.parameter;
                        return open(model, version);
                    }).catch(() => {
                        return open(model);
                    });
                }
                default: {
                    throw new nnabla.Error("Unsupported nnabla format '" + match + "'.");
                }
            }
        });
    }
};

nnabla.Model = class {

    constructor(metadata, model, format) {
        this._format = format;
        this._graphs = [ new nnabla.Graph(metadata, model) ];
    }

    get format(){
        return this._format;
    }

    get graphs() {
        return this._graphs;
    }
};

nnabla.Graph = class {

    constructor (metadata, model) {
        const executor = model.executor[0]; // TODO: Multiple executors?
        const network_name = executor.network_name;
        const network = model.network.find((item) => item.name === network_name);
        this._dataTypes = new Map(network.variable.map((item) => {
            const shape = new nnabla.TensorShape(item.shape.dim);
            const type = new nnabla.TensorType(item.type, shape);
            return [ item.name, type ];
        }));
        this._tensors = new Map(model.parameter.map((item) => {
            const name = item.variable_name;
            return [ name, new nnabla.Tensor(name, this.dataType(name), item.data) ];
        }));
        this._arguments = new Map();

        this._inputs = executor.data_variable.map((item) => {
            const name = item.variable_name;
            return new nnabla.Parameter(name, [ this.argument(name) ]);
        });
        this._outputs = executor.output_variable.map((item) => {
            const name = item.variable_name;
            return new nnabla.Parameter(name, [ this.argument(name) ]);
        });

        const get_parameters = (func) => {
            for (const [key, value] of Object.entries(func)) {
                if (key.endsWith("_param")) {
                    return value;
                }
            }

            return undefined;
        };

        this._nodes = network.function.map((func) => {
            const parameters = get_parameters(func) || [];
            const attributes = Object.entries(parameters).map(([name, value]) => {
                return new nnabla.Attribute(metadata, func.type, name, value);
            });
            const func_type = metadata.type(func.type);
            const inputs = [];
            for (let index = 0; index < func.input.length; ) {
                const input = func_type.inputs && index < func_type.inputs.length ? func_type.inputs[index] : { name: index.toString() };
                const count = input.list ? func.input.length - index : 1;
                const args = func.input.slice(index, index + count).map((input) => this.argument(input));
                inputs.push(new nnabla.Parameter(input.name, args));
                index += count;
            }
            const outputs = [];
            for (let index = 0; index < func.output.length; ) {
                const output = func_type.outputs && index < func_type.outputs.length ? func_type.outputs[index] : { name: index.toString() };
                const count = output.list ? func.output.length - index : 1;
                const args = func.output.slice(index, index + count).map((output) => this.argument(output));
                outputs.push(new nnabla.Parameter(output.name, args));
                index += count;
            }
            return new nnabla.Node(metadata, func, attributes, inputs, outputs);
        });
    }

    get nodes() {
        return this._nodes;
    }

    get inputs() {
        return this._inputs;
    }

    get outputs() {
        return this._outputs;
    }

    dataType(name) {
        return this._dataTypes.get(name);
    }

    tensor(name) {
        return this._tensors.get(name);
    }

    argument(name) {
        if (!this._arguments.has(name)) {
            this._arguments.set(name, new nnabla.Argument(name, this.dataType(name), this.tensor(name)));
        }
        return this._arguments.get(name);
    }
};

nnabla.Parameter = class {

    constructor(name, args) {
        this._name = name;
        this._arguments = args;
    }

    get name() {
        return this._name;
    }

    get visible() {
        return true;
    }

    get arguments() {
        return this._arguments;
    }
};

nnabla.Argument = class {

    constructor(name, type, initializer) {
        this._name = name;
        this._type = type || null;
        this._initializer = initializer || null;
    }

    get name() {
        return this._name;
    }

    get type() {
        if (this._type) {
            return this._type;
        }
        if (this._initializer) {
            return this._initializer.type;
        }
        return null;
    }

    get initializer() {
        return this._initializer;
    }
};

nnabla.Node = class {

    constructor(metadata, func, attributes, inputs, outputs) {
        this._name = func.name;
        this._type = metadata.type(func.type) || { name: func.type, type: func.type };
        this._attributes = attributes || [];
        this._outputs = outputs || [];
        this._chain = [];

        // TODO: "nonlinearity" does not match metadata type
        const get_nonlinearity = (name) => {
            switch (name) {
                case "identity": return "Identity";
                case "relu": return "ReLU";
                case "sigmoid": return "Sigmoid";
                case "tanh": return "Tanh";
                case "leaky_relu": return "LeakyReLU";
                case "elu": return "ELU";
                case "relu6": return "ReLU6";
                default: return name;
            }
        };

        switch (func.type) {
            case "FusedConvolution": {
                this._inputs = inputs.slice(0, 3) || [];
                if (inputs.length > 3) {
                    this._chain.push(new nnabla.Node(metadata, { name: func.name + "/bn", type: "BatchNormalization" }, [], inputs.slice(3, 7)));
                }
                if (inputs.length > 7) {
                    this._chain.push(new nnabla.Node(metadata, { name: func.name + "/add", type: "Add2" }, [], inputs.slice(7)));
                }
                const type_a = attributes.find((item) => item.name === "nonlinearity").value;
                this._chain.push(new nnabla.Node(metadata, { name: func.name + "/act", type: get_nonlinearity(type_a) }));
                break;
            }
            case "FusedBatchNormalization": {
                this._inputs = inputs.slice(0, 5) || [];
                if (inputs.length > 4) {
                    this._chain.push(new nnabla.Node(metadata, { name: func.name + "/add", type: "Add2" }, [], inputs.slice(5)));
                }
                const type_b = attributes.find((item) => item.name === "nonlinearity").value;
                this._chain.push(new nnabla.Node(metadata, { name: func.name + "/act", type: get_nonlinearity(type_b) }));
                break;
            }
            default: {
                this._inputs = inputs || [];
                break;
            }
        }
    }

    get name() {
        return this._name;
    }

    get type() {
        return this._type;
    }

    get attributes() {
        return this._attributes;
    }

    get inputs() {
        return this._inputs;
    }

    get outputs() {
        return this._outputs;
    }

    get chain() {
        return this._chain;
    }
};

nnabla.Attribute = class {

    constructor(metadata, type, name, value) {
        this._name = name;
        const attribute = metadata.attribute(type, name);
        this._description = attribute.description;
        switch (attribute.type) {
            case "Shape":
                this._type = "int64[]";
                this._value = value.dim;
                break;
            default:
                this._type = attribute.type;
                this._value = value;
                break;
        }
        if (Object.prototype.hasOwnProperty.call(attribute, 'default') && this._value == attribute.default) {
            this._visible = false;
        }
    }

    get name() {
        return this._name;
    }

    get description() {
        return this._description;
    }

    get type() {
        return this._type;
    }

    get value() {
        return this._value;
    }

    get visible() {
        return this._visible == false ? false : true;
    }
};

nnabla.Tensor = class {

    constructor(name, type, values) {
        this._name = name;
        this._type = type;
        this._values = values;
    }

    get kind() {
        return this._kind;
    }

    get name() {
        return this._name;
    }

    get type() {
        return this._type;
    }

    get state() {
        return this._context().state;
    }

    get value() {
        const context = this._context();
        if (context.state) {
            return null;
        }
        context.limit = Number.MAX_SAFE_INTEGER;
        return this._decode(context, 0);
    }

    toString() {
        const context = this._context();
        if (context.state) {
            return '';
        }
        context.limit = 10000;
        const value = this._decode(context, 0);
        return JSON.stringify(value, null, 4);
    }

    _context() {
        const context = {};
        context.state = null;
        context.index = 0;
        context.count = 0;
        if (!this._values) {
            context.state = 'Tensor data is empty.';
            return context;
        }
        switch (this._type.dataType) {
            case 'float32':
                context.data = new Float32Array(this._values);
                break;
            default:
                context.state = 'Unknown data type.';
                return context;
        }
        context.shape = this._type.shape.dimensions;
        context.dataType = this._type.dataType;
        return context;
    }

    _decode(context, dimension) {
        const results = [];
        const size = context.shape[dimension];
        if (dimension === context.shape.length - 1) {
            for (let i = 0; i < size; i++) {
                if (context.count > context.limit) {
                    results.push('...');
                    return results;
                }
                switch (context.dataType) {
                    case 'float32':
                        results.push(context.data[context.index]);
                        break;
                    default:
                        context.state = 'Unknown data type.';
                        break;
                }
                context.index++;
                context.count++;
            }
        }
        else {
            for (let j = 0; j < size; j++) {
                if (context.count > context.limit) {
                    results.push('...');
                    return results;
                }
                results.push(this._decode(context, dimension + 1));
            }
        }
        return results;
    }
};

nnabla.TensorType = class {

    constructor(dataType, shape) {
        this._dataType = "float32";
        this._shape = shape;
        this._denotation = null; // TODO
    }

    get dataType() {
        return this._dataType;
    }

    get shape() {
        return this._shape;
    }

    get denotation() {
        return this._denotation;
    }

    toString() {
        return this._dataType + this._shape.toString();
    }
};

nnabla.TensorShape = class {

    constructor(dimensions) {
        this._dimensions = dimensions;
    }

    get dimensions() {
        return this._dimensions;
    }

    toString() {
        return (this._dimensions && this._dimensions.length) ? ('[' + this._dimensions.join(',') + ']') : '';
    }
};

nnabla.Error = class extends Error {

    constructor(message) {
        super(message);
        this.name = 'Error loading Neural Network Library model.';
    }
};

if (typeof module !== 'undefined' && typeof module.exports === 'object') {
    module.exports.ModelFactory = nnabla.ModelFactory;
}