feat(webpage): Add IR test page (#293)

* feat(webpage): Add IR test page * Fix errors and add QA to leaderboard
CLARIN-PL · Sep 20, 2023 · 9ce347e · 9ce347e
1 parent a65e90d
commit 9ce347e
Show file tree

Hide file tree

Showing 11 changed files with 255 additions and 115 deletions.
diff --git a/.gitignore b/.gitignore
@@ -152,4 +152,6 @@ _proc
 _docs
 
 lepiszcze-submissions/
-checkpoints/
+checkpoints/
+# Local Netlify folder
+.netlify
diff --git a/webpage/.netlify/edge-functions-import-map.json b/webpage/.netlify/edge-functions-import-map.json
@@ -0,0 +1 @@
+{"imports":{"@netlify/edge-functions":"https://edge.netlify.com/v1/index.ts","netlify:edge":"https://edge.netlify.com/v1/index.ts?v=legacy"},"scopes":{}}
diff --git a/webpage/content/tasks/taskIR.md b/webpage/content/tasks/taskIR.md
@@ -0,0 +1,11 @@
+---
+url: "/tasks/InformationRetrieval"
+type: docs
+geekdocNav: false
+geekdocBreadcrumb: false
+---
+
+{{< pageHeader >}}
+{{< info taskname="Information Retrieval" taskdesc="Information Retrieval (IR) is the process of locating and retrieving relevant data or documents from a large dataset based on user queries, often used in search engines">}}
+{{< averageResults tasktype="Information Retrieval" >}}
+{{< results type="Information Retrieval" >}}
diff --git a/webpage/content/tasks/taskQA.md b/webpage/content/tasks/taskQA.md
@@ -0,0 +1,11 @@
+---
+url: "/tasks/QuestionAnswering"
+type: docs
+geekdocNav: false
+geekdocBreadcrumb: false
+---
+
+{{< pageHeader >}}
+{{< info taskname="Question Answering" taskdesc="Extractive Question Answering identifies precise spans within a document to answer a query, without generating new text or rephrasing">}}
+{{< averageResults tasktype="Question Answering">}}
+{{< results type="Question Answering" >}}
diff --git a/webpage/data/results/msmarco_bm_25.json b/webpage/data/results/msmarco_bm_25.json
@@ -0,0 +1,25 @@
+{
+  "submission_name": "msmarco_bm_25",
+  "dataset_name": "MSMARCO",
+  "dataset_version": "0.0.0",
+  "embedding_name": "BM25",
+  "hparams": {},
+  "packages": [],
+  "config": null,
+  "leaderboard_task_name": "Information Retrieval",
+  "metrics": [
+    {
+      "NDCG@10": 31.50,
+      "MRR@10": 56.36
+    }
+  ],
+  "metrics_avg": {
+      "NDCG@10": 31.50,
+      "MRR@10": 56.36
+    },
+  "metrics_std": {
+      "NDCG@10": 0.0,
+      "MRR@10": 0.0
+    },
+  "averaged_over": 1
+}
diff --git a/webpage/data/results/msmarco_ict.json b/webpage/data/results/msmarco_ict.json
@@ -0,0 +1,25 @@
+{
+  "submission_name": "msmarco_ict",
+  "dataset_name": "MSMARCO",
+  "dataset_version": "0.0.0",
+  "embedding_name": "ICT",
+  "hparams": {},
+  "packages": [],
+  "config": null,
+  "leaderboard_task_name": "Information Retrieval",
+  "metrics": [
+    {
+      "NDCG@10": 29.02,
+      "MRR@10": 43.91
+    }
+  ],
+  "metrics_avg": {
+      "NDCG@10": 29.02,
+      "MRR@10": 43.91
+    },
+  "metrics_std": {
+      "NDCG@10": 0.0,
+      "MRR@10": 0.0
+    },
+  "averaged_over": 1
+}
diff --git a/webpage/data/results/qa_all_allegro__herbert-base-cased.json b/webpage/data/results/qa_all_allegro__herbert-base-cased.json
@@ -0,0 +1,34 @@
+{
+  "submission_name": "qa_all_allegro__herbert-base-cased",
+  "dataset_name": "qa_all",
+  "dataset_version": "0.0.0",
+  "embedding_name": "allegro/herbert-base-cased",
+  "hparams": {},
+  "packages": [],
+  "config": null,
+  "leaderboard_task_name": "Question Answering",
+  "metrics": [
+    {
+      "f1": 81.06677705918307,
+      "exact": 72.24858757062147,
+      "HasAns_f1": 72.2968454891892,
+      "HasAns_exact": 56.738437001594896,
+      "NoAns_f1": 92.54042775169536
+    }
+  ],
+  "metrics_avg": {
+    "f1": 81.06677705918307,
+    "exact": 72.24858757062147,
+    "HasAns_f1": 72.2968454891892,
+    "HasAns_exact": 56.738437001594896,
+    "NoAns_f1": 92.54042775169536
+  },
+  "metrics_std": {
+    "f1": 0,
+    "exact": 0,
+    "HasAns_f1": 0,
+    "HasAns_exact": 0,
+    "NoAns_f1": 0
+  },
+  "averaged_over": 1
+}
diff --git a/webpage/layouts/shortcodes/averageResults.html b/webpage/layouts/shortcodes/averageResults.html
@@ -7,12 +7,16 @@
 
 {{ $metrics := slice }}
 {{ $metricsToIgnore := slice "classes" "f1_weighted" "recall_weighted" "precision_weighted" }}
-{{ range $.Site.Data.results }}
-   {{ range $k, $v := .metrics_avg }}
-      {{if (not (in $metricsToIgnore $k)) }}
-         {{ $metrics = uniq ($metrics | append $k) }}
-      {{end}}
-   {{ end }}
+{{ range $file, $result := $.Site.Data.results }}
+   {{if or (eq $tasktype "Tasks") (eq (index $result "leaderboard_task_name") $tasktype)}}
+        {{if (not (or (eq (index $result "leaderboard_task_name") "Information Retrieval") (eq (index $result "leaderboard_task_name") "Question Answering")))}}
+            {{ range $k, $v := .metrics_avg }}
+               {{if (not (in $metricsToIgnore $k)) }}
+                  {{ $metrics = uniq ($metrics | append $k) }}
+               {{end}}
+            {{ end }}
+        {{end}}
+     {{end}}
 {{ end }}
 {{ $sum := slice }}
 <style scoped>
@@ -27,121 +31,127 @@
 
 </style>
 <div style="margin: 0 2em;">
-    <div class="container title-container">    
-        {{ if eq $tasktype "Tasks"}} 
+    <div class="container title-container">
+        {{ if eq $tasktype "Tasks"}}
        <div class="title">Model ranking</div>
        <p style="text-align: center;">Average model performance across all tasks.</p>
-       {{else}}
-       <div class="title">Model's task ranking</div>
-       <p style="text-align: center;">Average model performance across task datasets.</p>
+        {{else}}
+           {{if (not (or (eq $tasktype "Information Retrieval") (eq $tasktype "Question Answering")))}}
+            <div class="title">Model's task ranking</div>
+            <p style="text-align: center;">Average model performance across task datasets.</p>
+            {{end}}
+
        {{end}}
 
 </div>
+    {{if (not (or (eq $tasktype "Information Retrieval") (eq $tasktype "Question Answering")))}}
 
-<div id="data-tasktype" style="visibility: hidden; height: 0px; overflow: hidden;">{{ $tasktype }}</div>
-<div id="data-average" style="visibility: hidden; height: 0px; overflow: hidden;">
-{{ range $model := $models }}
-    {{ range $metric := $metrics }}
-        {{ range $file, $result := $.Site.Data.results }}
-            {{if eq (index $result "embedding_name") $model}}
-                {{ $sum := ($sum | append (index $result "metrics_avg" $metric))}}
-                <div> {{ $model }};;{{ $metric }};;{{(index $result "leaderboard_task_name")}};;{{(index $result "dataset_name")}};;{{index $result "metrics_avg" $metric}}</div>
+    <div id="data-tasktype" style="visibility: hidden; height: 0px; overflow: hidden;">{{ $tasktype }}</div>
+    <div id="data-average" style="visibility: hidden; height: 0px; overflow: hidden;">
+    {{ range $model := $models }}
+        {{ range $metric := $metrics }}
+            {{ range $file, $result := $.Site.Data.results }}
+                {{if eq (index $result "embedding_name") $model}}
+                    {{ $sum := ($sum | append (index $result "metrics_avg" $metric))}}
+                    <div> {{ $model }};;{{ $metric }};;{{(index $result "leaderboard_task_name")}};;{{(index $result "dataset_name")}};;{{index $result "metrics_avg" $metric}}</div>
+                {{end}}
             {{end}}
         {{end}}
     {{end}}
-{{end}}
-</div>
-<div style="display: flex; justify-content: center;"><div class="lds-ring" id="average-spinner"><div></div><div></div><div></div><div></div></div></div>
-<table id="data-table" style="display: none;">
-    <thead>
-        <tr>
-            <th>
-                Model
-            </th>
-            {{ range $metric := $metrics }}
+    </div>
+    <div style="display: flex; justify-content: center;"><div class="lds-ring" id="average-spinner"><div></div><div></div><div></div><div></div></div></div>
+    <table id="data-table" style="display: none;">
+        <thead>
+            <tr>
                 <th>
-                    {{ $metric  |  humanize  }} 
+                    Model
                 </th>
-            {{end}}
-       </tr>
-    </thead>
-    <tr id="table-rows">
+                {{ range $metric := $metrics }}
+                    <th>
+                        {{ $metric  |  humanize  }}
+                    </th>
+                {{end}}
+           </tr>
+        </thead>
+        <tr id="table-rows">
 
-    </tr>
- </table>
+        </tr>
+     </table>
+    {{end}}
 
  <script>
-    let table = new Tabulator("#data-table", {
-        layout:"fitColumns",
-        movableColumns: true
-    });
-
-    let formattedColumns = table.getColumnDefinitions().map(c => { 
-        c.formatter = "html"; 
-        return c
-    })       
-    formattedColumns.forEach((col,i) => {
-        if(i==0){
-            col.headerFilter = true;
-            col.widthGrow = 3;
-            col.minWidth= 180;
-        }else{
-            col.formatter = "money";
-            col.hozAlign = "center";
-            col.formatterParams = {
-                precision:3,
-            };
-            col.headerFilter = false;
-        }
-    })
-    table.setColumns(formattedColumns);
+     if (document.querySelector('#data-table')) {
+        let table = new Tabulator("#data-table", {
+            layout:"fitColumns",
+            movableColumns: true
+        });
 
+        let formattedColumns = table.getColumnDefinitions().map(c => {
+            c.formatter = "html";
+            return c
+        })
+        formattedColumns.forEach((col,i) => {
+            if(i==0){
+                col.headerFilter = true;
+                col.widthGrow = 3;
+                col.minWidth= 180;
+            }else{
+                col.formatter = "money";
+                col.hozAlign = "center";
+                col.formatterParams = {
+                    precision:3,
+                };
+                col.headerFilter = false;
+            }
+        })
+        table.setColumns(formattedColumns);
+
+
+        table.setSort("f1_macro", "desc")
+        const dataDiv = document.getElementById("data-average");
+        let array = [...dataDiv.children];
+        let dataArray = []
+        array.forEach((el)=>{
+            const temp = el.innerHTML.split(";;");
+            dataArray.push({model: temp[0], metric: temp[1], task: temp[2], dataset: temp[3], value: temp[4]})
+        })
 
-    table.setSort("f1_macro", "desc")
-    const dataDiv = document.getElementById("data-average");
-    let array = [...dataDiv.children];
-    let dataArray = []
-    array.forEach((el)=>{
-        const temp = el.innerHTML.split(";;");
-        dataArray.push({model: temp[0], metric: temp[1], task: temp[2], dataset: temp[3], value: temp[4]})
-    })
+        const tasktype = document.getElementById("data-tasktype").innerHTML;
+        const uniqueModels = [...new Set(dataArray.map(item => item.model))];
+        const uniqueMetric = [...new Set(dataArray.map(item => item.metric))];
 
-    const tasktype = document.getElementById("data-tasktype").innerHTML;
-    const uniqueModels = [...new Set(dataArray.map(item => item.model))];
-    const uniqueMetric = [...new Set(dataArray.map(item => item.metric))];
-
-    let values = new Array(0);
-    uniqueModels.forEach(model => {
-        let modelObj = {model: model};
-        dataArray.forEach(el =>{
-            if( (tasktype == "Tasks" || el.task == tasktype) && el.model == model){
-                uniqueMetric.forEach( m =>{
-                    if(el.metric == m){
-                        if(modelObj[m+"_sum"] == undefined){
-                            modelObj[m+"_sum"] = +el.value;
-                            modelObj[m+"_count"] = 1;
-                            modelObj[m] = modelObj[m+"_sum"] / modelObj[m+"_count"];
-                        }else{
-                            modelObj[m+"_sum"] += +el.value;
-                            modelObj[m+"_count"] += 1;
-                            modelObj[m] = modelObj[m+"_sum"] / modelObj[m+"_count"];
+        let values = new Array(0);
+        uniqueModels.forEach(model => {
+            let modelObj = {model: model};
+            dataArray.forEach(el =>{
+                if( (tasktype == "Tasks" || el.task == tasktype) && el.model == model){
+                    uniqueMetric.forEach( m =>{
+                        if(el.metric == m){
+                            if(modelObj[m+"_sum"] == undefined){
+                                modelObj[m+"_sum"] = +el.value;
+                                modelObj[m+"_count"] = 1;
+                                modelObj[m] = modelObj[m+"_sum"] / modelObj[m+"_count"];
+                            }else{
+                                modelObj[m+"_sum"] += +el.value;
+                                modelObj[m+"_count"] += 1;
+                                modelObj[m] = modelObj[m+"_sum"] / modelObj[m+"_count"];
+                            }
                         }
-                    }
+                    })
+                }
+            })
+            if(Object.keys(modelObj).length>1){
+                const deleteKeys = Object.keys(modelObj).filter( item => item.includes("_sum") || item.includes("_count"));
+                deleteKeys.forEach(d => {
+                    delete modelObj[d];
                 })
+                values.push(modelObj);
             }
         })
-        if(Object.keys(modelObj).length>1){
-            const deleteKeys = Object.keys(modelObj).filter( item => item.includes("_sum") || item.includes("_count"));
-            deleteKeys.forEach(d => {
-                delete modelObj[d];
-            })
-            values.push(modelObj);
-        }
-    })
-    table.replaceData(values);
-    const content = document.getElementById("data-table");
-    const spinner = document.getElementById("average-spinner");
-    content.style.display = "block";
-    spinner.style.display = "none";
-
+        table.replaceData(values);
+        const content = document.getElementById("data-table");
+        const spinner = document.getElementById("average-spinner");
+        content.style.display = "block";
+        spinner.style.display = "none";
+    }
  </script>
diff --git a/webpage/layouts/shortcodes/homepage.html b/webpage/layouts/shortcodes/homepage.html
@@ -121,7 +121,9 @@ <h1 class="title">LEPISZCZE</h1>
           "Entailment Classification",
           "Aspect-based Sentiment Analysis",
           "Abusive Clauses Detection",
-          "Dialogue Acts Classification"
+          "Dialogue Acts Classification",
+          "Information Retrieval",
+          "Question Answering"
         ];
         const container = document.getElementById("tasks-container");
         taskTypes.forEach((type) => {