Merge branch 'pen' of https://github.com/browngraphicslab/Dash-Web into pen

1 files changed, 137 insertions, 0 deletions

diff --git a/src/server/Recommender.ts b/src/server/Recommender.ts
new file mode 100644
index 000000000..1d2cb3858
--- /dev/null
+++ b/src/server/Recommender.ts
@@ -0,0 +1,137 @@
+//import { Doc } from "../new_fields/Doc";
+//import { StrCast } from "../new_fields/Types";
+//import { List } from "../new_fields/List";
+//import { CognitiveServices } from "../client/cognitive_services/CognitiveServices";
+
+// var w2v = require('word2vec');
+var assert = require('assert');
+var arxivapi = require('arxiv-api-node');
+import requestPromise = require("request-promise");
+import * as use from '@tensorflow-models/universal-sentence-encoder';
+import { Tensor } from "@tensorflow/tfjs-core/dist/tensor";
+require('@tensorflow/tfjs-node');
+
+//http://gnuwin32.sourceforge.net/packages/make.htm
+
+export class Recommender {
+
+    private _model: any;
+    static Instance: Recommender;
+    private dimension: number = 0;
+    private choice: string = ""; // Tensorflow or Word2Vec
+
+    constructor() {
+        console.log("creating recommender...");
+        Recommender.Instance = this;
+    }
+
+    /***
+     * Loads pre-trained model from TF
+     */
+
+    public async loadTFModel() {
+        let self = this;
+        return new Promise(res => {
+            use.load().then(model => {
+                self.choice = "TF";
+                self._model = model;
+                self.dimension = 512;
+                res(model);
+            });
+        }
+
+        );
+    }
+
+    /***
+     * Loads pre-trained model from word2vec
+     */
+
+    // private loadModel(): Promise<any> {
+    //     let self = this;
+    //     return new Promise(res => {
+    //         w2v.loadModel("./node_modules/word2vec/examples/fixtures/vectors.txt", function (err: any, model: any) {
+    //             self.choice = "WV";
+    //             self._model = model;
+    //             self.dimension = model.size;
+    //             res(model);
+    //         });
+    //     });
+    // }
+
+    /***
+     * Testing
+     */
+
+    public async testModel() {
+        if (!this._model) {
+            await this.loadTFModel();
+        }
+        if (this._model) {
+            if (this.choice === "WV") {
+                let similarity = this._model.similarity('father', 'mother');
+                console.log(similarity);
+            }
+            else if (this.choice === "TF") {
+                const model = this._model as use.UniversalSentenceEncoder;
+                // Embed an array of sentences.
+                const sentences = [
+                    'Hello.',
+                    'How are you?'
+                ];
+                const embeddings = await this.vectorize(sentences);
+                if (embeddings) embeddings.print(true /*verbose*/);
+                // model.embed(sentences).then(embeddings => {
+                //     // `embeddings` is a 2D tensor consisting of the 512-dimensional embeddings for each sentence.
+                //     // So in this example `embeddings` has the shape [2, 512].
+                //     embeddings.print(true /* verbose */);
+                // });
+            }
+        }
+        else {
+            console.log("model not found :(");
+        }
+    }
+
+    /***
+     * Uses model to convert words to vectors
+     */
+
+    public async vectorize(text: string[]): Promise<Tensor | undefined> {
+        if (!this._model) {
+            await this.loadTFModel();
+        }
+        if (this._model) {
+            if (this.choice === "WV") {
+                let word_vecs = this._model.getVectors(text);
+                return word_vecs;
+            }
+            else if (this.choice === "TF") {
+                const model = this._model as use.UniversalSentenceEncoder;
+                return new Promise<Tensor>(res => {
+                    model.embed(text).then(embeddings => {
+                        res(embeddings);
+                    });
+                });
+
+            }
+        }
+    }
+
+    // public async trainModel() {
+    //     console.log("phrasing...");
+    //     w2v.word2vec("./node_modules/word2vec/examples/eng_news-typical_2016_1M-sentences.txt", './node_modules/word2vec/examples/my_phrases.txt', {
+    //         cbow: 1,
+    //         size: 200,
+    //         window: 8,
+    //         negative: 25,
+    //         hs: 0,
+    //         sample: 1e-4,
+    //         threads: 20,
+    //         iter: 200,
+    //         minCount: 2
+    //     });
+    //     console.log("phrased!!!");
+    // }
+
+}