Support for DAGRUN and DAGRUN_RO

README.md

@@ -125,6 +125,73 @@ Example of AI.SCRIPTSET and AI.SCRIPTRUN
 })();
 ```
 
+Example of AI.DAGRUN enqueuing multiple SCRIPTRUN and MODELRUN commands
+
+A common pattern is enqueuing multiple SCRIPTRUN and MODELRUN commands within a DAG. The following example uses ResNet-50,to classify images into 1000 object categories. 
+
+Given that our input tensor contains each color represented as a 8-bit integer and that neural networks usually work with floating-point tensors as their input we need to cast a tensor to floating-point and normalize the values of the pixels - for that we will use `pre_process_4ch` function. 
+
+To optimize the classification process we can use a post process script to return only the category position with the maximum classification - for that we will use `post_process` script. 
+
+Using the DAG capabilities we've removed the necessity of storing the intermediate tensors in the keyspace. You can even run the entire process without storing the output tensor, as follows:
+
+
+```javascript
+var redis = require('redis');
+var redisai = require('redisai-js');
+var fs = require("fs");
+
+(async () => {
+    const nativeClient = redis.createClient();
+    const aiclient = new redisai.Client(nativeClient);
+    const scriptFileStr = fs.readFileSync('./tests/test_data/imagenet/data_processing_script.txt').toString();
+    const jsonLabels = fs.readFileSync('./tests/test_data/imagenet/imagenet_class_index.json');
+    const labels = JSON.parse(jsonLabels);
+
+    const dataProcessingScript = new redisai.Script('CPU', scriptFileStr);
+    const resultScriptSet = await aiclient.scriptset('data_processing_script', dataProcessingScript);
+    // AI.SCRIPTSET result: OK
+    console.log(`AI.SCRIPTSET result: ${resultScriptSet}`)
+
+    const modelBlob = fs.readFileSync('./tests/test_data/imagenet/resnet50.pb');
+    const imagenetModel = new redisai.Model(Backend.TF, 'CPU', ['images'], ['output'], modelBlob);
+    const resultModelSet = await aiclient.modelset('imagenet_model', imagenetModel);
+
+    // AI.MODELSET result: OK
+    console.log(`AI.MODELSET result: ${resultModelSet}`)
+
+    const inputImage = await Jimp.read('./tests/test_data/imagenet/cat.jpg');
+    const imageWidth = 224;
+    const imageHeight = 224;
+    const image = inputImage.cover(imageWidth, imageHeight);
+    const tensor = new redisai.Tensor(Dtype.uint8, [imageWidth, imageHeight, 4], Buffer.from(image.bitmap.data));
+
+    ///
+    // Prepare the DAG enqueuing multiple SCRIPTRUN and MODELRUN commands
+    const dag = new redisai.Dag();
+
+    dag.tensorset('tensor-image', tensor);
+    dag.scriptrun('data_processing_script', 'pre_process_4ch', ['tensor-image'], ['temp_key1']);
+    dag.modelrun('imagenet_model', ['temp_key1'], ['temp_key2']);
+    dag.scriptrun('data_processing_script', 'post_process', ['temp_key2'], ['classification']);
+    dag.tensorget('classification');
+
+    // Send the AI.DAGRUN command to RedisAI server
+    const resultDagRun = await aiclient.dagrun_ro(null, dag);
+
+    // The 5th element of the reply will be the `classification` tensor 
+    const classTensor = resultDagRun[4];
+
+    // Print the category in the position with the max classification
+    const idx = classTensor.data[0];
+
+    // 281 [ 'n02123045', 'tabby' ]
+    console.log(idx, labels[idx.toString()]);
+
+    await aiclient.end();
+})();
+```
+
 ### Further examples
 
 The [RedisAI examples repo](https://github.com/RedisAI/redisai-examples) shows more advanced examples
@@ -147,8 +214,8 @@ AI.SCRIPTGET | scriptget
 AI.SCRIPTDEL | scriptdel
 AI.SCRIPTRUN | scriptrun
 AI._SCRIPTSCAN | N/A  
-AI.DAGRUN | N/A
-AI.DAGRUN_RO | N/A
+AI.DAGRUN | dagrun
+AI.DAGRUN_RO | dagrun_ro
 AI.INFO | info and infoResetStat (for resetting stats)
 AI.CONFIG * | configLoadBackend and configBackendsPath
 

src/client.ts

@@ -4,16 +4,17 @@ import { Model } from './model';
 import * as util from 'util';
 import { Script } from './script';
 import { Stats } from './stats';
+import { Dag } from './dag';
 
 export class Client {
-  private _sendCommand: any;
+  private readonly _sendCommand: any;
 
   constructor(client: RedisClient) {
     this._client = client;
     this._sendCommand = util.promisify(this._client.send_command).bind(this._client);
   }
 
-  private _client: RedisClient;
+  private readonly _client: RedisClient;
 
   get client(): RedisClient {
     return this._client;
@@ -23,23 +24,13 @@ export class Client {
     this._client.end(flush);
   }
 
-  public tensorset(keName: string, t: Tensor): Promise<any> {
-    const args: any[] = [keName, t.dtype];
-    t.shape.forEach((value) => args.push(value.toString()));
-    if (t.data != null) {
-      if (t.data instanceof Buffer) {
-        args.push('BLOB');
-        args.push(t.data);
-      } else {
-        args.push('VALUES');
-        t.data.forEach((value) => args.push(value.toString()));
-      }
-    }
+  public tensorset(keyName: string, t: Tensor): Promise<any> {
+    const args: any[] = t.tensorSetFlatArgs(keyName);
     return this._sendCommand('ai.tensorset', args);
   }
 
-  public tensorget(keName: string): Promise<any> {
-    const args: any[] = [keName, 'META', 'VALUES'];
+  public tensorget(keyName: string): Promise<any> {
+    const args: any[] = Tensor.tensorGetFlatArgs(keyName);
     return this._sendCommand('ai.tensorget', args)
       .then((reply: any[]) => {
         return Tensor.NewTensorFromTensorGetReply(reply);
@@ -55,10 +46,7 @@ export class Client {
   }
 
   public modelrun(modelName: string, inputs: string[], outputs: string[]): Promise<any> {
-    const args: any[] = [modelName, 'INPUTS'];
-    inputs.forEach((value) => args.push(value));
-    args.push('OUTPUTS');
-    outputs.forEach((value) => args.push(value));
+    const args: any[] = Model.modelRunFlatArgs(modelName, inputs, outputs);
     return this._sendCommand('ai.modelrun', args);
   }
 
@@ -68,7 +56,7 @@ export class Client {
   }
 
   public modelget(modelName: string): Promise<any> {
-    const args: any[] = [modelName, 'META', 'BLOB'];
+    const args: any[] = Model.modelGetFlatArgs(modelName);
     return this._sendCommand('ai.modelget', args)
       .then((reply: any[]) => {
         return Model.NewModelFromModelGetReply(reply);
@@ -78,22 +66,13 @@ export class Client {
       });
   }
 
-  public scriptset(keName: string, s: Script): Promise<any> {
-    const args: any[] = [keName, s.device];
-    if (s.tag !== undefined) {
-      args.push('TAG');
-      args.push(s.tag);
-    }
-    args.push('SOURCE');
-    args.push(s.script);
+  public scriptset(keyName: string, s: Script): Promise<any> {
+    const args: any[] = s.scriptSetFlatArgs(keyName);
     return this._sendCommand('ai.scriptset', args);
   }
 
   public scriptrun(scriptName: string, functionName: string, inputs: string[], outputs: string[]): Promise<any> {
-    const args: any[] = [scriptName, functionName, 'INPUTS'];
-    inputs.forEach((value) => args.push(value));
-    args.push('OUTPUTS');
-    outputs.forEach((value) => args.push(value));
+    const args: any[] = Script.scriptRunFlatArgs(scriptName, functionName, inputs, outputs);
     return this._sendCommand('ai.scriptrun', args);
   }
 
@@ -103,7 +82,7 @@ export class Client {
   }
 
   public scriptget(scriptName: string): Promise<any> {
-    const args: any[] = [scriptName, 'META', 'SOURCE'];
+    const args: any[] = Script.scriptGetFlatArgs(scriptName);
     return this._sendCommand('ai.scriptget', args)
       .then((reply: any[]) => {
         return Script.NewScriptFromScriptGetReply(reply);
@@ -137,6 +116,40 @@ export class Client {
       });
   }
 
+  /**
+   * specifies a direct acyclic graph of operations to run within RedisAI
+   *
+   * @param loadKeys
+   * @param persistKeys
+   * @param dag
+   */
+  public dagrun(loadKeys: string[] | null, persistKeys: string[] | null, dag: Dag): Promise<any> {
+    const args: any[] = dag.dagRunFlatArgs(loadKeys, persistKeys);
+    return this._sendCommand('ai.dagrun', args)
+      .then((reply: any[]) => {
+        return dag.ProcessDagReply(reply);
+      })
+      .catch((error: any) => {
+        throw error;
+      });
+  }
+
+  /**
+   * specifies a Read Only direct acyclic graph of operations to run within RedisAI
+   *
+   * @param loadKeys
+   * @param dag
+   */
+  public dagrun_ro(loadKeys: string[] | null, dag: Dag): Promise<any> {
+    const args: any[] = dag.dagRunFlatArgs(loadKeys, null);
+    return this._sendCommand('ai.dagrun_ro', args)
+      .then((reply: any[]) => {
+        return dag.ProcessDagReply(reply);
+      })
+      .catch((error: any) => {
+        throw error;
+      });
+  }
   /**
    * Loads the DL/ML backend specified by the backend identifier from path.
    *

src/dag.ts

@@ -0,0 +1,88 @@
+import { Model } from './model';
+import { Script } from './script';
+import { Tensor } from './tensor';
+
+export interface DagCommandInterface {
+  tensorset(keyName: string, t: Tensor): DagCommandInterface;
+
+  tensorget(keyName: string): DagCommandInterface;
+
+  tensorget(keyName: string): DagCommandInterface;
+
+  modelrun(modelName: string, inputs: string[], outputs: string[]): DagCommandInterface;
+
+  scriptrun(scriptName: string, functionName: string, inputs: string[], outputs: string[]): DagCommandInterface;
+}
+
+/**
+ * Direct mapping to RedisAI DAGs
+ */
+export class Dag implements DagCommandInterface {
+  private _commands: any[][];
+  private readonly _tensorgetflag: boolean[];
+
+  constructor() {
+    this._commands = [];
+    this._tensorgetflag = [];
+  }
+
+  public tensorset(keyName: string, t: Tensor): Dag {
+    const args: any[] = ['AI.TENSORSET'];
+    t.tensorSetFlatArgs(keyName).forEach((arg) => args.push(arg));
+    this._commands.push(args);
+    this._tensorgetflag.push(false);
+    return this;
+  }
+
+  public tensorget(keyName: string): Dag {
+    const args: any[] = ['AI.TENSORGET'];
+    Tensor.tensorGetFlatArgs(keyName).forEach((arg) => args.push(arg));
+    this._commands.push(args);
+    this._tensorgetflag.push(true);
+    return this;
+  }
+
+  public modelrun(modelName: string, inputs: string[], outputs: string[]): Dag {
+    const args: any[] = ['AI.MODELRUN'];
+    Model.modelRunFlatArgs(modelName, inputs, outputs).forEach((arg) => args.push(arg));
+    this._commands.push(args);
+    this._tensorgetflag.push(false);
+    return this;
+  }
+
+  public scriptrun(scriptName: string, functionName: string, inputs: string[], outputs: string[]): Dag {
+    const args: any[] = ['AI.SCRIPTRUN'];
+    Script.scriptRunFlatArgs(scriptName, functionName, inputs, outputs).forEach((arg) => args.push(arg));
+    this._commands.push(args);
+    this._tensorgetflag.push(false);
+    return this;
+  }
+
+  public dagRunFlatArgs(loadKeys: string[] | null, persistKeys: string[] | null): string[] {
+    const args: any[] = [];
+    if (loadKeys != null && loadKeys.length > 0) {
+      args.push('LOAD');
+      args.push(loadKeys.length);
+      loadKeys.forEach((value) => args.push(value));
+    }
+    if (persistKeys != null && persistKeys.length > 0) {
+      args.push('PERSIST');
+      args.push(persistKeys.length);
+      persistKeys.forEach((value) => args.push(value));
+    }
+    this._commands.forEach((value) => {
+      args.push('|>');
+      value.forEach((arg) => args.push(arg));
+    });
+    return args;
+  }
+
+  public ProcessDagReply(reply: any[]): any[] {
+    for (let i = 0; i < reply.length; i++) {
+      if (this._tensorgetflag[i] === true) {
+        reply[i] = Tensor.NewTensorFromTensorGetReply(reply[i]);
+      }
+    }
+    return reply;
+  }
+}

src/index.ts

@@ -3,8 +3,9 @@ import { Backend, BackendMap } from './backend';
 import { Tensor } from './tensor';
 import { Model } from './model';
 import { Script } from './script';
+import { Dag } from './dag';
 import { Client } from './client';
 import { Stats } from './stats';
 import { Helpers } from './helpers';
 
-export { DTypeMap, Dtype, BackendMap, Backend, Model, Script, Tensor, Client, Stats, Helpers };
+export { DTypeMap, Dtype, BackendMap, Backend, Model, Script, Tensor, Dag, Client, Stats, Helpers };

src/model.ts

@@ -187,7 +187,19 @@ export class Model {
     return model;
   }
 
-  modelSetFlatArgs(keyName: string) {
+  static modelGetFlatArgs(keyName: string): string[] {
+    return [keyName, 'META', 'BLOB'];
+  }
+
+  static modelRunFlatArgs(modelName: string, inputs: string[], outputs: string[]): string[] {
+    const args: string[] = [modelName, 'INPUTS'];
+    inputs.forEach((value) => args.push(value));
+    args.push('OUTPUTS');
+    outputs.forEach((value) => args.push(value));
+    return args;
+  }
+
+  modelSetFlatArgs(keyName: string): any[] {
     const args: any[] = [keyName, this.backend.toString(), this.device];
     if (this.tag !== undefined) {
       args.push('TAG');

src/script.ts

@@ -87,4 +87,28 @@ export class Script {
     }
     return script;
   }
+
+  scriptSetFlatArgs(keyName: string): string[] {
+    const args: string[] = [keyName, this.device];
+    if (this.tag !== undefined) {
+      args.push('TAG');
+      args.push(this.tag);
+    }
+    args.push('SOURCE');
+    args.push(this.script);
+    return args;
+  }
+
+  static scriptRunFlatArgs(scriptName: string, functionName: string, inputs: string[], outputs: string[]): string[] {
+    const args: string[] = [scriptName, functionName, 'INPUTS'];
+    inputs.forEach((value) => args.push(value));
+    args.push('OUTPUTS');
+    outputs.forEach((value) => args.push(value));
+    return args;
+  }
+
+  static scriptGetFlatArgs(scriptName: string): string[] {
+    const args: string[] = [scriptName, 'META', 'SOURCE'];
+    return args;
+  }
 }