Fix last minute bugs (#595)

- Compute concept scores for all datasets - Show the max score in the concept preview - pin "ML" and "OpenOrca" so it's the first dataset. https://huggingface.co/spaces/lilacai/lilac Meta tags fixed: https://lilacml.com/ https://metatags.io/?url=https%3A%2F%2Flilacml.com%2F
databricks · Aug 24, 2023 · 910e9e2 · 910e9e2
1 parent 0c88092
commit 910e9e2
Show file tree

Hide file tree

Showing 21 changed files with 131 additions and 35 deletions.
diff --git a/README.md b/README.md
@@ -4,7 +4,7 @@
 [![Downloads](https://static.pepy.tech/badge/lilac/month)](https://pepy.tech/project/lilac)
 [![License](https://img.shields.io/badge/License-Apache_2.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)
 [![Twitter](https://img.shields.io/twitter/follow/lilac_ai)](https://twitter.com/lilac_ai)
-[![](https://dcbadge.vercel.app/api/server/YpGxQMyk?compact=true&style=flat)](https://discord.gg/YpGxQMyk)
+[![](https://dcbadge.vercel.app/api/server/jNzw9mC8pp?compact=true&style=flat)](https://discord.gg/jNzw9mC8pp)
 
 > **NEW: Try the [Lilac hosted demo with pre-loaded datasets](https://lilacai-lilac.hf.space/)**
 

diff --git a/docs/_static/logo.png b/docs/_static/logo.png
diff --git a/docs/_static/logo_wide.png b/docs/_static/logo_wide.png
diff --git a/docs/blog/introducing-lilac.md b/docs/blog/introducing-lilac.md
@@ -1,5 +1,5 @@
 ```{tip}
-Try the Lilac hosted **[demo on HuggingFace](https://huggingface.co/spaces/lilacai/lilac)** or find
+Try the Lilac hosted **[demo on HuggingFace](https://lilacai-lilac.hf.space/)** or find
 us on GitHub: **[github.com/lilacai/lilac](https://github.com/lilacai/lilac)**
 ```
 

diff --git a/docs/concepts/concept_metrics.md b/docs/concepts/concept_metrics.md
@@ -1,7 +1,7 @@
 # Concept metrics
 
 ```{tip}
-[Try Lilac concepts on HuggingFace](https://huggingface.co/spaces/lilacai/lilac)
+[Try Lilac concepts on HuggingFace](https://lilacai-lilac.hf.space/)
 ```
 
 We can quantify the quality of a concept using an [F1 score](https://en.wikipedia.org/wiki/F-score)

diff --git a/docs/concepts/concept_tuning.md b/docs/concepts/concept_tuning.md
@@ -1,7 +1,7 @@
 # Tuning a concept
 
 ```{tip}
-[Try Lilac concepts on HuggingFace](https://huggingface.co/spaces/lilacai/lilac)
+[Try Lilac concepts on HuggingFace](https://lilacai-lilac.hf.space/)
 ```
 
 Often times, after creating a concept or using an off-the-shelf-concept, the concept needs to be

diff --git a/docs/concepts/concept_use.md b/docs/concepts/concept_use.md
@@ -17,7 +17,7 @@ for details on understanding the quality of a concept with an embedding.
 ## From the UI
 
 ```{tip}
-[Try Lilac concepts on HuggingFace](https://huggingface.co/spaces/lilacai/lilac)
+[Try Lilac concepts on HuggingFace](https://lilacai-lilac.hf.space/)
 ```
 
 To use a concept from the UI, click the concept from the Navigation panel, which will open the

diff --git a/docs/concepts/concepts.md b/docs/concepts/concepts.md
@@ -1,7 +1,7 @@
 # Concepts
 
 ```{tip}
-[Try Lilac concepts on HuggingFace](https://huggingface.co/spaces/lilacai/lilac)
+[Try Lilac concepts on HuggingFace](https://lilacai-lilac.hf.space/)
 ```
 
 ## What is a concept?

diff --git a/docs/datasets/dataset_export.md b/docs/datasets/dataset_export.md
@@ -1,7 +1,7 @@
 # Export data
 
 ```{tip}
-[Download enrichments for popular datasets on HuggingFace](https://huggingface.co/spaces/lilacai/lilac)
+[Download enrichments for popular datasets on HuggingFace](https://lilacai-lilac.hf.space/)
 ```
 
 Once we've computed signals and concepts over a dataset, it can be very useful to download the

diff --git a/docs/huggingface/huggingface_spaces.md b/docs/huggingface/huggingface_spaces.md
@@ -1,7 +1,7 @@
 # Duplicate the HuggingFace demo
 
-Lilac hosts a [HuggingFace spaces demo](https://huggingface.co/spaces/lilacai/lilac) so you can try
-Lilac before installing it.
+Lilac hosts a [HuggingFace spaces demo](https://lilacai-lilac.hf.space/) so you can try Lilac before
+installing it.
 
 Thanks to HuggingFace, this space can be duplicated and customized with your own data. You can
 decide to make your duplicated private for use with private or sensitive data.

diff --git a/docs/index.rst b/docs/index.rst
@@ -1,5 +1,12 @@
 🌸 Lilac
 =================
+.. meta::
+   :description: Analyze, structure and clean unstructured data with AI
+   :keywords: datasets, AI, machine learning, unstructured, lilac
+
+.. raw:: html
+
+    <meta property="og:image" content="https://lilacml.com/_static/logo_wide.png" />
 
 .. include:: welcome.md
    :parser: myst_parser.sphinx_

diff --git a/docs/signals/signals.md b/docs/signals/signals.md
@@ -1,7 +1,7 @@
 # Signals
 
 ```{tip}
-[Try Lilac signals on HuggingFace](https://huggingface.co/spaces/lilacai/lilac)
+[Try Lilac signals on HuggingFace](https://lilacai-lilac.hf.space/)
 ```
 
 There are two types of [](#Signal) base classes based on the input:

diff --git a/docs/welcome.md b/docs/welcome.md
@@ -1,21 +1,22 @@
 ```{tip}
-Try the Lilac hosted **[demo on HuggingFace](https://huggingface.co/spaces/lilacai/lilac)** or find us on GitHub: **[github.com/lilacai/lilac](https://github.com/lilacai/lilac)**
+Try the Lilac hosted **[demo on HuggingFace](https://lilacai-lilac.hf.space/)** or find us on GitHub: **[github.com/lilacai/lilac](https://github.com/lilacai/lilac)**
 ```
 
 [![GitHub Repo stars](https://img.shields.io/github/stars/lilacai/lilac?logo=github&label=lilacai%2Flilac)](https://github.com/lilacai/lilac)
 [![Downloads](https://static.pepy.tech/badge/lilac/month)](https://pepy.tech/project/lilac)
 [![License](https://img.shields.io/badge/License-Apache_2.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)
 [![Twitter](https://img.shields.io/twitter/follow/lilac_ai)](https://twitter.com/lilac_ai)
-[![](https://dcbadge.vercel.app/api/server/YpGxQMyk?compact=true&style=flat)](https://discord.gg/YpGxQMyk)
+[![](https://dcbadge.vercel.app/api/server/jNzw9mC8pp?compact=true&style=flat)](https://discord.gg/jNzw9mC8pp)
 
 ## 👋 Welcome
 
 [Lilac](http://lilacml.com) is an open-source product that helps you **analyze**, **structure**, and
 **clean** unstructured data with AI.
 
 See the [Installation](./getting_started/installation.md) and
-[Quick Start](./getting_started/quickstart.md) guides to get started. Read the
-[Announcement Blog](./blog/introducing-lilac.md) for more details.
+[Quick Start](./getting_started/quickstart.md) guides to get started.
+
+Read the [Announcement Blog](./blog/introducing-lilac.md) for more details.
 
 <video loop muted autoplay controls src="https://github-production-user-asset-6210df.s3.amazonaws.com/2294279/260771834-cb1378f8-92c1-4f2a-9524-ce5ddd8e0c53.mp4"></video>
 

diff --git a/lilac/signals/text_statistics.py b/lilac/signals/text_statistics.py
@@ -9,6 +9,7 @@
 
 SPACY_LANG_MODEL = 'en_core_web_sm'
 SPACY_BATCH_SIZE = 128
+SPACY_MAX_LENGTH = 2_000_000
 
 NUM_CHARS = 'num_characters'
 READABILITY = 'readability'
@@ -55,6 +56,7 @@ def setup(self) -> None:
       disable=[
         'parser', 'tagger', 'ner', 'lemmatizer', 'textcat', 'custom', 'tok2vec', 'attribute_ruler'
       ])
+    self._lang.max_length = SPACY_MAX_LENGTH
 
   @override
   def compute(self, data: Iterable[RichData]) -> Iterable[Optional[Item]]:

diff --git a/lilac_hf_space.yml b/lilac_hf_space.yml
@@ -203,6 +203,38 @@ signals:
   - signal_name: text_statistics
   - signal_name: near_dup
   - signal_name: lang_detection
+  - signal_name: concept_score
+    namespace: lilac
+    concept_name: legal-termination
+    embedding: gte-small
+  - signal_name: concept_score
+    namespace: lilac
+    concept_name: negative-sentiment
+    embedding: gte-small
+  - signal_name: concept_score
+    namespace: lilac
+    concept_name: non-english
+    embedding: gte-small
+  - signal_name: concept_score
+    namespace: lilac
+    concept_name: positive-sentiment
+    embedding: gte-small
+  - signal_name: concept_score
+    namespace: lilac
+    concept_name: profanity
+    embedding: gte-small
+  - signal_name: concept_score
+    namespace: lilac
+    concept_name: question
+    embedding: gte-small
+  - signal_name: concept_score
+    namespace: lilac
+    concept_name: source-code
+    embedding: gte-small
+  - signal_name: concept_score
+    namespace: lilac
+    concept_name: toxicity
+    embedding: gte-small
 
 concept_model_cache_embeddings:
   - gte-small

diff --git a/web/blueprint/src/lib/components/HuggingFaceSpaceWelcome.svelte b/web/blueprint/src/lib/components/HuggingFaceSpaceWelcome.svelte
@@ -27,17 +27,24 @@
 <div class="mx-32 flex w-full flex-col items-center gap-y-6 px-8">
   <div class="mt-8 w-full text-center">
     <h2>Welcome to Lilac</h2>
-    <div class="mt-2 text-base text-gray-700">
-      analyze, structure and clean unstructured data with AI
-    </div>
+    <div class="mt-2 text-base text-gray-700">analyze, structure and clean data with AI</div>
     <div class="mt-2 text-sm text-gray-700">
-      <a href="https://lilacml.com">visit our website</a>
+      <a target="_blank" href="https://lilacml.com">visit our website</a>
     </div>
     <div class="duplicate mt-6 flex flex-row items-center justify-center gap-x-4 text-gray-700">
-      <Button
-        href={`https://huggingface.co/spaces/${huggingFaceSpaceId}?duplicate=true`}
-        kind="tertiary">duplicate</Button
+      <div
+        use:hoverTooltip={{
+          text:
+            'Duplicate the HuggingFace space to manage your own instance. ' +
+            'You must be logged into HuggingFace for the duplicate modal to appear.'
+        }}
       >
+        <Button
+          href={`https://huggingface.co/spaces/${huggingFaceSpaceId}?duplicate=true`}
+          target="_blank"
+          kind="tertiary">duplicate</Button
+        >
+      </div>
       <a
         class="-ml-2"
         use:hoverTooltip={{
@@ -58,8 +65,9 @@
       title={`Browse the ${tryDataset.name} dataset`}
     >
       <p class="text-sm">
-        Try the Lilac dataset viewer on the the pre-loaded <a href={tryDataset.originalLink}
-          >{tryDataset.displayName}</a
+        Try the Lilac dataset viewer on the the pre-loaded <a
+          target="_blank"
+          href={tryDataset.originalLink}>{tryDataset.displayName}</a
         > dataset.
       </p>
     </WelcomeBanner>

diff --git a/web/blueprint/src/lib/components/WelcomeBanner.svelte b/web/blueprint/src/lib/components/WelcomeBanner.svelte
@@ -7,7 +7,7 @@
   export let backgroundColorClass: string;
 </script>
 
-<div class="welcome-item mt-8 w-full rounded shadow-md">
+<div class="my-4 w-full rounded shadow-md">
   <div
     class={`flex cursor-pointer flex-row justify-between ${backgroundColorClass} rounded-t px-4 py-4 font-semibold`}
     on:click={() => goto(link)}

diff --git a/web/blueprint/src/lib/components/concepts/ConceptPreview.svelte b/web/blueprint/src/lib/components/concepts/ConceptPreview.svelte
@@ -14,6 +14,7 @@
   import {Button, Select, SelectItem, SkeletonText, TextArea} from 'carbon-components-svelte';
   import {onMount} from 'svelte';
   import StringSpanHighlight from '../datasetView/StringSpanHighlight.svelte';
+  import {colorFromScore} from '../datasetView/colors';
   import type {SpanValueInfo} from '../datasetView/spanHighlight';
 
   export let concept: Concept;
@@ -82,6 +83,15 @@
       valuePaths = spanValuePaths.valuePaths;
     }
   }
+  let maxConceptScore: number;
+  // Compute the max score and show it separately.
+  $: if ($conceptScore?.data != null) {
+    maxConceptScore = 0;
+    for (const row of $conceptScore.data[0]) {
+      const score = row['score'] as number;
+      maxConceptScore = Math.max(score, maxConceptScore);
+    }
+  }
 </script>
 
 <div class="flex flex-col gap-x-8">
@@ -111,6 +121,12 @@
     {#if conceptScore && $conceptScore?.isFetching}
       <SkeletonText />
     {:else if previewResultItem != null && previewText != null}
+      <div class="my-2">
+        <span class="font-medium">Max score:</span>
+        <span class="rounded p-0.5" style:background-color={colorFromScore(maxConceptScore)}
+          >{maxConceptScore.toFixed(3)}</span
+        >
+      </div>
       <StringSpanHighlight
         text={previewText}
         row={previewResultItem}

diff --git a/web/blueprint/src/lib/components/schemaView/SchemaField.svelte b/web/blueprint/src/lib/components/schemaView/SchemaField.svelte
@@ -125,9 +125,24 @@
         checked={isVisible}
         on:change={() => {
           if (!isVisible) {
+            // For signals, when the root of a signal is checked, enable all the children.
+            if (field.signal != null) {
+              const children = childFields(field);
+              children.forEach(f => {
+                datasetViewStore.addSelectedColumn(f.path);
+              });
+            }
             datasetViewStore.addSelectedColumn(path);
+            // Repeated fields are collapsed. When clicked, we need to also make them visible.
+            if (field.repeated_field != null) {
+              datasetViewStore.addSelectedColumn([...path, PATH_WILDCARD]);
+            }
           } else {
             datasetViewStore.removeSelectedColumn(path);
+            // Repeated fields are collapsed. When clicked, we need to also make them visible.
+            if (field.repeated_field != null) {
+              datasetViewStore.removeSelectedColumn([...path, PATH_WILDCARD]);
+            }
           }
         }}
       />

diff --git a/web/blueprint/src/lib/view_utils.ts b/web/blueprint/src/lib/view_utils.ts
@@ -167,9 +167,10 @@ export function isPathVisible(
     return true;
   }
 
-  if (selectedColumns[path] != null)
+  if (selectedColumns[path] != null) {
     // If a user has explicitly selected a column, return the value of the selection.
     return selectedColumns[path];
+  }
 
   const pathArray = deserializePath(path);
 
@@ -328,21 +329,32 @@ export function getTaggedDatasets(
       tagDatasets[tag][dataset.namespace].push(dataset);
     }
   }
+  const tagSortPriorities = ['machine-learning'];
+  const sortedTags = Object.keys(tagDatasets).sort(
+    (a, b) => tagSortPriorities.indexOf(b) - tagSortPriorities.indexOf(a) || a.localeCompare(b)
+  );
+
   const namespaceSortPriorities = ['lilac'];
+  // TODO(nsthorat): Don't hard-code this. Let's make this a config.
+  const pinnedDatasets = ['OpenOrca-100k'];
 
   // Sort each tag by namespace and then dataset name.
   const taggedDatasetGroups: NavigationTagGroup[] = [];
-  for (const tag of Object.keys(tagDatasets).sort()) {
+  for (const tag of sortedTags) {
     const sortedNamespaceDatasets: NavigationGroupItem[] = Object.keys(tagDatasets[tag])
       .sort(
         (a, b) =>
-          namespaceSortPriorities.indexOf(a) - namespaceSortPriorities.indexOf(b) ||
+          namespaceSortPriorities.indexOf(b) - namespaceSortPriorities.indexOf(a) ||
           a.localeCompare(b)
       )
       .map(namespace => ({
         group: namespace,
         items: tagDatasets[tag][namespace]
-          .sort((a, b) => a.dataset_name.localeCompare(b.dataset_name))
+          .sort(
+            (a, b) =>
+              pinnedDatasets.indexOf(b.dataset_name) - pinnedDatasets.indexOf(a.dataset_name) ||
+              a.dataset_name.localeCompare(b.dataset_name)
+          )
           .map(d => ({
             name: d.dataset_name,
             link: datasetLink(d.namespace, d.dataset_name),
@@ -378,6 +390,7 @@ export function getTaggedConcepts(
       tagConcepts[tag][concept.namespace].push(concept);
     }
   }
+
   const namespaceSortPriorities = ['lilac'];
 
   // Sort each tag by namespace and then dataset name.

diff --git a/web/blueprint/src/routes/+page.svelte b/web/blueprint/src/routes/+page.svelte
@@ -11,11 +11,13 @@
 </script>
 
 <Page>
-  {#if $authInfo.isFetching}
-    <SkeletonText />
-  {:else if huggingFaceSpaceId != null && !canCreateDataset}
-    <HuggingFaceSpaceWelcome />
-  {:else}
-    <GettingStarted />
-  {/if}
+  <div class="flex h-full w-full gap-y-4 overflow-y-scroll p-4">
+    {#if $authInfo.isFetching}
+      <SkeletonText />
+    {:else if huggingFaceSpaceId != null && !canCreateDataset}
+      <HuggingFaceSpaceWelcome />
+    {:else}
+      <GettingStarted />
+    {/if}
+  </div>
 </Page>