Embedding visualization connect (#356)

* Make the frontend to fetch the data from the flask server.

* Provide loading animation.

* Fixed format style
Fix the issue where the display label not working.

* Provide a 2D 3D switcher.
Display the chart in 3D.

* Include PCA method to do dimension reduction

* Fix the style.

* Fix typo
Use === to compare type as well.

frontend/mock/data/plugin/embeddings/embeddings.js

@@ -6,31 +6,86 @@
  * @param {Object} postParam post params
  * @return {Object}
  */
-module.exports = function (path, queryParam, postParam) {
-    return {
-        // moock delay
-        _timeout: 0,
-        // mock http status
-        _status: 200,
-        // mock response data
-        _data: {
-            status: 0,
-            msg: 'SUCCESS',
-            data: {
-                "embedding": [
-                    [10.0, 8.04, "yellow"],
-                    [8.0, 6.95, "blue"],
-                    [13.0, 7.58, "red"],
-                    [9.0, 8.81, "king"],
-                    [11.0, 8.33, "queen"],
-                    [14.0, 9.96, "man"],
-                    [6.0, 7.24, "women"],
-                    [4.0, 4.26, "kid"],
-                    [12.0, 10.84, "adult"],
-                    [7.0, 4.82, "light"],
-                    [5.0, 5.68, "dark"]
-                ]
+module.exports = function(path, queryParam, postParam) {
+    if (queryParam.dimension === '3') {
+        return {
+            // moock delay
+            _timeout: 0,
+            // mock http status
+            _status: 200,
+            // mock response data
+            _data: {
+                status: 0,
+                msg: 'SUCCESS',
+                data: {
+                    "embedding": [
+                        [10.0, 8.04, 3],
+                        [8.0, 6.95, 4],
+                        [13.0, 7.58, 1],
+                        [9.0, 8.81, 3],
+                        [11.0, 8.33, 5],
+                        [14.0, 9.96, 6],
+                        [6.0, 7.24, 1],
+                        [4.0, 4.26, 2],
+                        [12.0, 10.84, 6],
+                        [7.0, 4.8, 3],
+                        [5.0, 5.68, 3]
+                    ],
+                    "labels": [
+                        "yellow",
+                        "blue",
+                        "red",
+                        "king",
+                        "queen",
+                        "man",
+                        "women",
+                        "kid",
+                        "adult",
+                        "light",
+                        "dark"
+                    ]
+                }
             }
-        }
-    };
+        };
+    } else {
+        return {
+            // moock delay
+            _timeout: 0,
+            // mock http status
+            _status: 200,
+            // mock response data
+            _data: {
+                status: 0,
+                msg: 'SUCCESS',
+                data: {
+                    "embedding": [
+                        [10.0, 8.04],
+                        [8.0, 6.95],
+                        [13.0, 7.58],
+                        [9.0, 8.81],
+                        [11.0, 8.33],
+                        [14.0, 9.96],
+                        [6.0, 7.24],
+                        [4.0, 4.26],
+                        [12.0, 10.84],
+                        [7.0, 4.8],
+                        [5.0, 5.68]
+                    ],
+                    "labels": [
+                        "yellow",
+                        "blue",
+                        "red",
+                        "king",
+                        "queen",
+                        "man",
+                        "women",
+                        "kid",
+                        "adult",
+                        "light",
+                        "dark"
+                    ]
+                }
+            }
+        };
+    }
 };

frontend/package.json

@@ -20,7 +20,8 @@
     "d3-format": "^1.2.1",
     "dagre": "^0.8.2",
     "dagre-d3": "^0.6.1",
-    "echarts": "^3.8.5",
+    "echarts": "^4.0.0",
+    "echarts-gl": "^1.1.0",
     "file-saver": "^1.3.3",
     "graphlib": "^1.0.5",
     "htmlcs": "^0.4.1",

frontend/src/high-dimensional/HighDimensional.vue

@@ -5,6 +5,7 @@
                     :config="config"
                     :displayWordLabel="config.displayWordLabel"
                     :searchText="config.searchText"
+                    :dimension="config.dimension"
                     :embedding_data="embedding_data"
             ></ui-chart>
         </div>
@@ -35,20 +36,45 @@ export default {
             config: {
                 searchText: '',
                 displayWordLabel: true,
+                dimension: "2",
+                reduction: "tsne",
                 running: true
             },
             embedding_data: []
         }
     },
     created() {
-    	    getHighDimensionalDatasets().then(({errno, data}) => {
-            this.embedding_data = data.embedding;
-        });
+        this.fetchDatasets()
+    },
+    watch: {
+        'config.dimension': function(val) {
+            this.fetchDatasets()
+        },
+        'config.reduction': function(val) {
+            this.fetchDatasets()
+        }
     },
     mounted() {
         autoAdjustHeight();
     },
     methods: {
+        fetchDatasets() {
+            // Fetch the data from the server. Passing dimension and reduction method
+            let params = {
+                dimension: this.config.dimension,
+                reduction: this.config.reduction
+            };
+            getHighDimensionalDatasets(params).then(({errno, data}) => {
+                var vector_data = data.embedding;
+                var labels = data.labels;
+
+                for ( var i = 0; i < vector_data.length; i ++) {
+                  vector_data[i].push(labels[i])
+                }
+
+                this.embedding_data = vector_data
+            });
+        },
     }
 };
 

frontend/src/high-dimensional/ui/Chart.vue

@@ -7,9 +7,10 @@
 
 <script>
 import echarts from 'echarts';
+import 'echarts-gl';
 
 export default {
-    props: ['config', 'displayWordLabel', 'searchText', 'embedding_data'],
+    props: ['config', 'displayWordLabel', 'searchText', 'embedding_data', 'dimension'],
     data() {
         return {
             width: 900,
@@ -18,20 +19,21 @@ export default {
     },
     computed: {
         computedStyle() {
-            return 'height:' + this.height + 'px;'
-                + 'width:' + this.width + 'px;';
+            return 'height:' + this.height + 'px;' +
+                'width:' + this.width + 'px;';
         }
     },
-    created() {
-
-    },
+    created() {},
     mounted() {
         this.createChart();
-        this.setChartsOptions();
+        this.myChart.showLoading()
+
+        this.set2DChartOptions();
         this.setDisplayWordLabel();
     },
     watch: {
         embedding_data: function(val) {
+            this.myChart.hideLoading();
             this.myChart.setOption({
                 series: [{
                     // Grab the 'matched' series data
@@ -43,13 +45,25 @@ export default {
         displayWordLabel: function(val) {
             this.setDisplayWordLabel()
         },
+        dimension: function(val) {
+            this.myChart.clear()
+            this.myChart.showLoading()
+            if (val === "2") {
+                this.set2DChartOptions();
+                this.setDisplayWordLabel();
+            } else {
+                this.set3DChartOptions();
+                this.setDisplayWordLabel();
+            }
+        },
         searchText: function(val) {
+            // Filter the data that has the hasPrefix
             var matched_words = []
             if (val != '') {
                 val = val.toLowerCase()
 
                 function hasPrefix(value) {
-                    var word = value[2]
+                    var word = value[value.length - 1]
                     return (typeof word == "string" && word.toLowerCase().startsWith(val))
                 }
 
@@ -72,13 +86,12 @@ export default {
             let el = this.$refs.chartBox;
             this.myChart = echarts.init(el);
         },
-        setChartsOptions(){
+        set2DChartOptions() {
             var typeD = "normal";
             var option = {
                 xAxis: {},
                 yAxis: {},
-                series: [
-                    {
+                series: [{
                         name: "all",
                         symbolSize: 10,
                         data: this.embedding_data,
@@ -87,7 +100,7 @@ export default {
                     {
                         name: "matched",
                         animation: false,
-                        symbolSize:10,
+                        symbolSize: 10,
                         data: [],
                         itemStyle: {
                             normal: {
@@ -95,37 +108,80 @@ export default {
                             }
                         },
                         label: {
+                            normal: {
+                                show: true,
+                                formatter: function(param) {
+                                    return param.data[param.data.length - 1];
+                                },
+                                position: 'top'
+                            }
+                        },
+                        type: 'scatter'
+                    }
+                ]
+            };
+            this.myChart.setOption(option);
+        },
+        set3DChartOptions() {
+          var symbolSize = 2.5;
+          var option3d = {
+            grid3D: {},
+                    xAxis3D: {
+                        type: 'category'
+                    },
+                    yAxis3D: {},
+                    xAxis3D: {},
+                    zAxis3D: {},
+                    dataset: {
+                        source: this.embedding_data
+                    },
+                    series: [
+                        {
+                            name: "all",
+                            type: 'scatter3D',
+                            symbolSize: symbolSize,
+                            data: []
+                        },
+                        {
+                            name: "matched",
+                            animation: false,
+                            symbolSize: symbolSize,
+                            data: [],
+                            label: {
                                 normal: {
                                     show: true,
-                                    formatter: function (param) {
-                                        return param.data[2];
+                                    formatter: function(param) {
+                                        return param.data[param.data.length - 1];
                                     },
                                     position: 'top'
                                 }
                             },
-                        type: 'scatter'
-                    }
-                ]
-            };
-            this.myChart.setOption(option);
+                            type: 'scatter3D'
+                        }
+                    ]
+          }
+          this.myChart.setOption(option3d);
         },
         setDisplayWordLabel() {
-                this.myChart.setOption({
+            this.myChart.setOption({
+                series: [{
+                    // Grab the 'all' series data
+                    name: 'all',
                     label: {
                         normal: {
                             show: this.displayWordLabel,
-                            formatter: function (param) {
-                                return param.data[2];
-                                },
-                                position: 'top'
+                            formatter: function(param) {
+                                return param.data[param.data.length - 1];
                             },
-                            emphasis: {
-                                show: true,
-                            }
+                            position: 'top'
+                        },
+                        emphasis: {
+                            show: true,
                         }
-                })
+                    }
+                }]
+            });
         },
-
     }
 };
 
@@ -136,7 +192,4 @@ export default {
         float left
         padding 10px
         position relative
-
-
-
 </style>

frontend/src/high-dimensional/ui/Config.vue

@@ -13,6 +13,16 @@
                     label="Display All Labels"
                     v-model="config.displayWordLabel" dark></v-checkbox>
 
+        <v-radio-group label="Dimension" v-model="config.dimension" dark>
+            <v-radio label="2D" value="2"></v-radio>
+            <v-radio label="3D" value="3"></v-radio>
+        </v-radio-group>
+
+        <v-radio-group label="Reduction Method" v-model="config.reduction" dark>
+            <v-radio label="T-SNE" value="tsne"></v-radio>
+            <v-radio label="PCA" value="pca"></v-radio>
+        </v-radio-group>
+
         <v-btn :color="config.running ? 'primary' : 'error'"
                   v-model="config.running"
                   @click="toggleAllRuns"
@@ -64,5 +74,3 @@ export default {
             margin-bottom 10px
 
 </style>
-
-

visualdl/server/lib.py

@@ -196,6 +196,32 @@ def get_texts(storage, mode, tag, num_records=100):
         return res
 
 
+def get_embeddings(storage,
+                   mode,
+                   tag,
+                   reduction,
+                   dimension=2,
+                   num_records=5000):
+    with storage.mode(mode) as reader:
+        embedding = reader.embedding(tag)
+        labels = embedding.get_all_labels()
+        high_dimensional_vectors = embedding.get_all_embeddings()
+
+        # TODO: Move away from sklearn
+        if reduction == 'tsne':
+            from sklearn.manifold import TSNE
+            tsne = TSNE(
+                perplexity=30, n_components=dimension, init='pca', n_iter=5000)
+            low_dim_embs = tsne.fit_transform(high_dimensional_vectors)
+
+        elif reduction == 'pca':
+            from sklearn.decomposition import PCA
+            pca = PCA(n_components=3)
+            low_dim_embs = pca.fit_transform(high_dimensional_vectors)
+
+        return {"embedding": low_dim_embs.tolist(), "labels": labels}
+
+
 def get_histogram(storage, mode, tag):
     with storage.mode(mode) as reader:
         histogram = reader.histogram(tag)

visualdl/server/visualDL

@@ -236,6 +236,14 @@ def texts():
     result = gen_result(0, "", data)
     return Response(json.dumps(result), mimetype='application/json')
 
+@app.route('/data/plugin/embeddings/embeddings')
+def embeddings():
+    dimension = request.args.get('dimension')
+    reduction = request.args.get('reduction')
+    key = os.path.join('/data/plugin/embeddings/embeddings', dimension, reduction)
+    data = cache_get(key, try_call, lib.get_embeddings, log_reader, 'train', 'scratch/embedding', reduction, int(dimension))
+    result = gen_result(0, "", data)
+    return Response(json.dumps(result), mimetype='application/json')
 
 @app.route('/data/plugin/graphs/graph')
 def graph():