knowpro: Image Collection persistance (#777)

Image Collection Persistence:
- Moved to using 2 files; JSON for data, binary for embeddings. 
- Breaking change: old saved indexes will no longer work.

ts/examples/chat/src/memory/knowproMemory.ts

@@ -25,13 +25,7 @@ import {
     parseFreeAndNamedArguments,
     keyValuesFromNamedArgs,
 } from "./common.js";
-import {
-    dateTime,
-    ensureDir,
-    getFileName,
-    readJsonFile,
-    writeJsonFile,
-} from "typeagent";
+import { dateTime, ensureDir, getFileName } from "typeagent";
 import path from "path";
 import chalk from "chalk";
 import { KnowProPrinter } from "./knowproPrinter.js";
@@ -310,9 +304,12 @@ export async function createKnowproCommands(
         context.printer.writeLine("Saving index");
         context.printer.writeLine(namedArgs.filePath);
         if (context.images) {
-            const cData = await context.images.serialize();
-            await ensureDir(path.dirname(namedArgs.filePath));
-            await writeJsonFile(namedArgs.filePath, cData);
+            const dirName = path.dirname(namedArgs.filePath);
+            await ensureDir(dirName);
+            await context.images.writeToFile(
+                dirName,
+                getFileName(namedArgs.filePath),
+            );
         }
     }
 
@@ -337,24 +334,15 @@ export async function createKnowproCommands(
             context.printer.writeError("No filepath or name provided");
             return;
         }
-        if (!fs.existsSync(imagesFilePath)) {
-            context.printer.writeError(`${imagesFilePath} not found`);
-            return;
-        }
-
-        const data = await readJsonFile<im.ImageCollectionData>(imagesFilePath);
-        if (!data) {
-            context.printer.writeError("Could not load image collection data");
+        context.images = await im.ImageCollection.readFromFile(
+            path.dirname(imagesFilePath),
+            getFileName(imagesFilePath),
+        );
+        if (!context.images) {
+            context.printer.writeLine("ImageCollection not found");
             return;
         }
-        context.images = new im.ImageCollection(
-            data.nameTag,
-            data.messages,
-            data.tags,
-            data.semanticRefs,
-        );
-        await context.images.deserialize(data);
-        context.conversation = context.podcast;
+        context.conversation = context.images;
         context.printer.conversation = context.conversation;
         context.printer.writeImageCollectionInfo(context.images);
     }

ts/packages/memory/image/src/importImages.ts

@@ -18,6 +18,9 @@ import {
     buildSecondaryIndexes,
     ConversationSecondaryIndexes,
     IndexingEventHandlers,
+    IPersistedConversationData,
+    writeConversationToFile,
+    readConversationFromFile,
 } from "knowpro";
 import { conversation as kpLib, image } from "knowledge-processor";
 import fs from "node:fs";
@@ -356,8 +359,8 @@ export class ImageCollection implements IConversation<ImageMeta> {
     public semanticRefIndex: ConversationIndex;
     public secondaryIndexes: ConversationSecondaryIndexes;
     constructor(
-        public nameTag: string,
-        public messages: Image[],
+        public nameTag: string = "",
+        public messages: Image[] = [],
         public tags: string[] = [],
         public semanticRefs: SemanticRef[] = [],
     ) {
@@ -397,16 +400,18 @@ export class ImageCollection implements IConversation<ImageMeta> {
         }
 
         this.addMetadataToIndex();
-        await buildSecondaryIndexes(this, true);
+        await buildSecondaryIndexes(this, true, eventHandler);
 
         let indexingResult: IndexingResults = {
             chunksIndexedUpto: { messageIndex: this.messages.length - 1 },
         };
         return indexingResult;
     }
 
-    public async serialize(): Promise<ImageCollectionData> {
-        return {
+    public async serialize(): Promise<
+        IPersistedConversationData<ImageCollectionData>
+    > {
+        const conversationData: ImageCollectionData = {
             nameTag: this.nameTag,
             messages: this.messages,
             tags: this.tags,
@@ -415,9 +420,33 @@ export class ImageCollection implements IConversation<ImageMeta> {
             relatedTermsIndexData:
                 this.secondaryIndexes.termToRelatedTermsIndex.serialize(),
         };
+        let persistentData: IPersistedConversationData<ImageCollectionData> = {
+            conversationData,
+        };
+        const embeddingData =
+            conversationData.relatedTermsIndexData?.textEmbeddingData;
+        if (embeddingData) {
+            persistentData.embeddings = embeddingData.embeddings;
+            embeddingData.embeddings = [];
+        }
+        return persistentData;
     }
 
-    public async deserialize(data: ImageCollectionData): Promise<void> {
+    public async deserialize(
+        persistentData: IPersistedConversationData<ImageCollectionData>,
+    ): Promise<void> {
+        const data = persistentData.conversationData;
+        const embeddingData =
+            persistentData.conversationData.relatedTermsIndexData
+                ?.textEmbeddingData;
+        if (persistentData.embeddings && embeddingData) {
+            embeddingData.embeddings = persistentData.embeddings;
+        }
+
+        this.nameTag = data.nameTag;
+        this.messages = data.messages;
+        this.semanticRefs = data.semanticRefs;
+        this.tags = data.tags;
         if (data.semanticIndexData) {
             this.semanticRefIndex = new ConversationIndex(
                 data.semanticIndexData,
@@ -430,6 +459,36 @@ export class ImageCollection implements IConversation<ImageMeta> {
         }
         await buildSecondaryIndexes(this, false);
     }
+
+    public async writeToFile(
+        dirPath: string,
+        baseFileName: string,
+    ): Promise<void> {
+        await writeConversationToFile(
+            this,
+            dirPath,
+            baseFileName,
+            async (conversation) => this.serialize(),
+        );
+    }
+
+    public static async readFromFile(
+        dirPath: string,
+        baseFileName: string,
+    ): Promise<ImageCollection | undefined> {
+        const imageCollection = new ImageCollection();
+        await readConversationFromFile<ImageCollectionData>(
+            dirPath,
+            baseFileName,
+            imageCollection.settings.relatedTermIndexSettings
+                .embeddingIndexSettings?.embeddingSize,
+            async (persistentData) => {
+                await imageCollection.deserialize(persistentData);
+                return imageCollection;
+            },
+        );
+        return imageCollection;
+    }
 }
 
 /**
@@ -540,7 +599,7 @@ async function indexImage(
         console.log(`Could not find part of the file path '${fileName}'`);
         return;
     } else if (!isImageFileType(path.extname(fileName))) {
-        console.log(`Skipping '${fileName}', not a known image file.`);
+        //console.log(`Skipping '${fileName}', not a known image file.`);
         return;
     }