Merge pull request #187 from microsoft/master

merge master

docs/en_US/CustomizeTuner.md

@@ -31,7 +31,7 @@ class CustomizedTuner(Tuner):
     def __init__(self, ...):
         ...
 
-    def receive_trial_result(self, parameter_id, parameters, value):
+    def receive_trial_result(self, parameter_id, parameters, value, **kwargs):
     '''
     Receive trial's final result.
     parameter_id: int
@@ -41,7 +41,7 @@ class CustomizedTuner(Tuner):
     # your code implements here.
     ...
 
-    def generate_parameters(self, parameter_id):
+    def generate_parameters(self, parameter_id, **kwargs):
     '''
     Returns a set of trial (hyper-)parameters, as a serializable object
     parameter_id: int
@@ -51,7 +51,7 @@ class CustomizedTuner(Tuner):
     ...
 ```
 
-`receive_trial_result` will receive the `parameter_id, parameters, value` as parameters input. Also, Tuner will receive the `value` object are exactly same value that Trial send.
+`receive_trial_result` will receive the `parameter_id, parameters, value` as parameters input. Also, Tuner will receive the `value` object are exactly same value that Trial send. If `multiPhase` is set to `true` in the experiment configuration file, an additional `trial_job_id` parameter is passed to `receive_trial_result` and `generate_parameters` through the `**kwargs` parameter.
 
 The `your_parameters` return from `generate_parameters` function, will be package as json object by NNI SDK. NNI SDK will unpack json object so the Trial will receive the exact same `your_parameters` from Tuner.
 
@@ -59,7 +59,7 @@ For example:
 If the you implement the `generate_parameters` like this:
 
 ```python
-def generate_parameters(self, parameter_id):
+def generate_parameters(self, parameter_id, **kwargs):
     '''
     Returns a set of trial (hyper-)parameters, as a serializable object
     parameter_id: int

docs/en_US/MultiPhase.md

@@ -38,7 +38,15 @@ To enable multi-phase, you should also add `multiPhase: true` in your experiment
 
 ### Write a tuner that leverages multi-phase:
 
-Before writing a multi-phase tuner, we highly suggest you to go through  [Customize Tuner](https://nni.readthedocs.io/en/latest/Customize_Tuner.html). Different from writing a normal tuner, your tuner needs to inherit from `MultiPhaseTuner` (in nni.multi_phase_tuner). The key difference between `Tuner` and `MultiPhaseTuner` is that the methods in MultiPhaseTuner are aware of additional information, that is, `trial_job_id`. With this information, the tuner could know which trial is requesting a configuration, and which trial is reporting results. This information provides enough flexibility for your tuner to deal with different trials and different phases. For example, you may want to use the trial_job_id parameter of generate_parameters method to generate hyperparameters for a specific trial job.
+Before writing a multi-phase tuner, we highly suggest you to go through  [Customize Tuner](https://nni.readthedocs.io/en/latest/Customize_Tuner.html). Same as writing a normal tuner, your tuner needs to inherit from `Tuner` class. When you enable multi-phase through configuration (set `multiPhase` to true), your tuner will get an additional parameter `trial_job_id` via tuner's following methods:
+```
+generate_parameters
+generate_multiple_parameters
+receive_trial_result
+receive_customized_trial_result
+trial_end
+```
+With this information, the tuner could know which trial is requesting a configuration, and which trial is reporting results. This information provides enough flexibility for your tuner to deal with different trials and different phases. For example, you may want to use the trial_job_id parameter of generate_parameters method to generate hyperparameters for a specific trial job.
 
 Of course, to use your multi-phase tuner, __you should add `multiPhase: true` in your experiment YAML configure file__.
 

docs/en_US/WebUI.md

@@ -8,6 +8,7 @@ Click the tab "Overview".
 * Support to download the experiment result.
 * Support to export nni-manager and dispatcher log file.
 * If you have any question, you can click "Feedback" to report it.
+* If your experiment have more than 1000 trials, you can change the refresh interval on here.
 
 ![](../img/webui-img/over1.png)
 * See good performance trials.
@@ -58,6 +59,10 @@ Click the tab "Trials Detail" to see the status of the all trials. Specifically:
 
 ![](../img/webui-img/addColumn.png)
 
+* If you want to compare some trials, you can select them and then click "Compare" to see the results.
+
+![](../img/webui-img/compare.png)
+
 * You can use the button named "Copy as python" to copy trial's parameters.
 
 ![](../img/webui-img/copyParameter.png)
@@ -69,6 +74,6 @@ Click the tab "Trials Detail" to see the status of the all trials. Specifically:
 
 * Kill: you can kill a job that status is running.
 * Support to search for a specific trial.
-* Intermediate Result Graph.
+* Intermediate Result Graph: you can see default and other keys in this graph.
 
-![](../img/intermediate.png)
+![](../img/webui-img/intermediate.png)

examples/tuners/ga_customer_tuner/customer_tuner.py

@@ -79,7 +79,7 @@ class CustomerTuner(Tuner):
         logger.debug('init population done.')
         return
 
-    def generate_parameters(self, parameter_id):
+    def generate_parameters(self, parameter_id, **kwargs):
         """Returns a set of trial graph config, as a serializable object.
         parameter_id : int
         """
@@ -109,7 +109,7 @@ class CustomerTuner(Tuner):
         return temp
 
 
-    def receive_trial_result(self, parameter_id, parameters, value):
+    def receive_trial_result(self, parameter_id, parameters, value, **kwargs):
         '''
         Record an observation of the objective function
         parameter_id : int

examples/tuners/random_nas_tuner/random_nas_tuner.py

@@ -49,12 +49,12 @@ class RandomNASTuner(Tuner):
         self.searchspace_json = search_space
         self.random_state = np.random.RandomState()
 
-    def generate_parameters(self, parameter_id):
+    def generate_parameters(self, parameter_id, **kwargs):
         '''generate
         '''
         return random_archi_generator(self.searchspace_json, self.random_state)
 
-    def receive_trial_result(self, parameter_id, parameters, value):
+    def receive_trial_result(self, parameter_id, parameters, value, **kwargs):
         '''receive
         '''
         pass

examples/tuners/weight_sharing/ga_customer_tuner/customer_tuner.py

@@ -112,7 +112,7 @@ class CustomerTuner(Tuner):
             population.append(Individual(indiv_id=self.generate_new_id(), graph_cfg=graph_tmp, result=None))
         return population
 
-    def generate_parameters(self, parameter_id):
+    def generate_parameters(self, parameter_id, **kwargs):
         """Returns a set of trial graph config, as a serializable object.
         An example configuration:
         ```json
@@ -196,7 +196,7 @@ class CustomerTuner(Tuner):
         logger.debug("trial {} ready".format(indiv.indiv_id))
         return param_json
 
-    def receive_trial_result(self, parameter_id, parameters, value):
+    def receive_trial_result(self, parameter_id, parameters, value, **kwargs):
         '''
         Record an observation of the objective function
         parameter_id : int

src/nni_manager/common/utils.ts

@@ -375,7 +375,7 @@ function countFilesRecursively(directory: string, timeoutMilliSeconds?: number):
 }
 
 function validateFileName(fileName: string): boolean {
-    let pattern: string = '^[a-z0-9A-Z\.-_]+$';
+    let pattern: string = '^[a-z0-9A-Z\._-]+$';
     const validateResult = fileName.match(pattern);
     if(validateResult) {
         return true;

src/nni_manager/training_service/common/clusterJobRestServer.ts

@@ -58,6 +58,10 @@ export abstract class ClusterJobRestServer extends RestServer {
         this.port = basePort + 1;
     }
 
+    get apiRootUrl(): string {
+        return this.API_ROOT_URL;
+    }
+
     public get clusterRestServerPort(): number {
         if (this.port === undefined) {
             throw new Error('PAI Rest server port is undefined');
@@ -87,7 +91,7 @@ export abstract class ClusterJobRestServer extends RestServer {
     protected abstract handleTrialMetrics(jobId : string, trialMetrics : any[]) : void;
 
     // tslint:disable: no-unsafe-any no-any
-    private createRestHandler() : Router {
+    protected createRestHandler() : Router {
         const router: Router = Router();
 
         router.use((req: Request, res: Response, next: any) => {

src/nni_manager/training_service/local/localTrainingService.ts

@@ -355,7 +355,8 @@ class LocalTrainingService implements TrainingService {
         this.log.info('Stopping local machine training service...');
         this.stopping = true;
         for (const stream of this.jobStreamMap.values()) {
-            stream.destroy();
+            stream.end(0)
+            stream.emit('end')
         }
         if (this.gpuScheduler !== undefined) {
             await this.gpuScheduler.stop();
@@ -372,7 +373,9 @@ class LocalTrainingService implements TrainingService {
                 if (stream === undefined) {
                     throw new Error(`Could not find stream in trial ${trialJob.id}`);
                 }
-                stream.destroy();
+                //Refer https://github.com/Juul/tail-stream/issues/20
+                stream.end(0)
+                stream.emit('end')
                 this.jobStreamMap.delete(trialJob.id);
             }
         }
@@ -567,7 +570,6 @@ class LocalTrainingService implements TrainingService {
                 buffer = remain;
             }
         });
-
         this.jobStreamMap.set(trialJobDetail.id, stream);
     }
 

src/nni_manager/training_service/pai/paiData.ts

@@ -64,11 +64,11 @@ else
 fi`;
 
 export const PAI_TRIAL_COMMAND_FORMAT: string =
-`export NNI_PLATFORM=pai NNI_SYS_DIR={0} NNI_OUTPUT_DIR={1} NNI_TRIAL_JOB_ID={2} NNI_EXP_ID={3} NNI_TRIAL_SEQ_ID={4} \
+`export NNI_PLATFORM=pai NNI_SYS_DIR={0} NNI_OUTPUT_DIR={1} NNI_TRIAL_JOB_ID={2} NNI_EXP_ID={3} NNI_TRIAL_SEQ_ID={4} MULTI_PHASE={5} \
 && cd $NNI_SYS_DIR && sh install_nni.sh \
-&& python3 -m nni_trial_tool.trial_keeper --trial_command '{5}' --nnimanager_ip '{6}' --nnimanager_port '{7}' \
---pai_hdfs_output_dir '{8}' --pai_hdfs_host '{9}' --pai_user_name {10} --nni_hdfs_exp_dir '{11}' --webhdfs_path '/webhdfs/api/v1' \
---nni_manager_version '{12}' --log_collection '{13}'`;
+&& python3 -m nni_trial_tool.trial_keeper --trial_command '{6}' --nnimanager_ip '{7}' --nnimanager_port '{8}' \
+--pai_hdfs_output_dir '{9}' --pai_hdfs_host '{10}' --pai_user_name {11} --nni_hdfs_exp_dir '{12}' --webhdfs_path '/webhdfs/api/v1' \
+--nni_manager_version '{13}' --log_collection '{14}'`;
 
 export const PAI_OUTPUT_DIR_FORMAT: string =
 `hdfs://{0}:9000/`;

src/nni_manager/training_service/pai/paiJobRestServer.ts

@@ -19,17 +19,26 @@
 
 'use strict';
 
+import { Request, Response, Router } from 'express';
 import { Inject } from 'typescript-ioc';
 import * as component from '../../common/component';
 import { ClusterJobRestServer } from '../common/clusterJobRestServer';
 import { PAITrainingService } from './paiTrainingService';
 
+export interface ParameterFileMeta {
+    readonly experimentId: string;
+    readonly trialId: string;
+    readonly filePath: string;
+}
+
 /**
  * PAI Training service Rest server, provides rest API to support pai job metrics update
  *
  */
 @component.Singleton
 export class PAIJobRestServer extends ClusterJobRestServer {
+    private parameterFileMetaList: ParameterFileMeta[] = [];
+
     @Inject
     private readonly paiTrainingService : PAITrainingService;
 
@@ -52,4 +61,33 @@ export class PAIJobRestServer extends ClusterJobRestServer {
             });
         }
     }
+
+    protected createRestHandler(): Router {
+        const router: Router = super.createRestHandler();
+
+        router.post(`/parameter-file-meta`, (req: Request, res: Response) => {
+            try {
+                this.log.info(`POST /parameter-file-meta, body is ${JSON.stringify(req.body)}`);
+                this.parameterFileMetaList.push(req.body);
+                res.send();
+            } catch (err) {
+                this.log.error(`POST parameter-file-meta error: ${err}`);
+                res.status(500);
+                res.send(err.message);
+            }
+        });
+
+        router.get(`/parameter-file-meta`, (req: Request, res: Response) => {
+            try {
+                this.log.info(`GET /parameter-file-meta`);
+                res.send(this.parameterFileMetaList);
+            } catch (err) {
+                this.log.error(`GET parameter-file-meta error: ${err}`);
+                res.status(500);
+                res.send(err.message);
+            }
+        });
+
+        return router;
+    }
 }

src/nni_manager/training_service/pai/paiTrainingService.ts

@@ -33,7 +33,7 @@ import { MethodNotImplementedError } from '../../common/errors';
 import { getExperimentId, getInitTrialSequenceId } from '../../common/experimentStartupInfo';
 import { getLogger, Logger } from '../../common/log';
 import {
-    JobApplicationForm, NNIManagerIpConfig, TrainingService,
+    HyperParameters, JobApplicationForm, NNIManagerIpConfig, TrainingService,
     TrialJobApplicationForm, TrialJobDetail, TrialJobMetric
 } from '../../common/trainingService';
 import { delay, generateParamFileName,
@@ -45,7 +45,7 @@ import { HDFSClientUtility } from './hdfsClientUtility';
 import { NNIPAITrialConfig, PAIClusterConfig, PAIJobConfig, PAITaskRole } from './paiConfig';
 import { PAI_LOG_PATH_FORMAT, PAI_OUTPUT_DIR_FORMAT, PAI_TRIAL_COMMAND_FORMAT, PAITrialJobDetail } from './paiData';
 import { PAIJobInfoCollector } from './paiJobInfoCollector';
-import { PAIJobRestServer } from './paiJobRestServer';
+import { PAIJobRestServer, ParameterFileMeta } from './paiJobRestServer';
 
 import * as WebHDFS from 'webhdfs';
 
@@ -79,6 +79,7 @@ class PAITrainingService implements TrainingService {
     private copyExpCodeDirPromise?: Promise<void>;
     private versionCheck: boolean = true;
     private logCollection: string;
+    private isMultiPhase: boolean = false;
 
     constructor() {
         this.log = getLogger();
@@ -179,12 +180,22 @@ class PAITrainingService implements TrainingService {
         return deferred.promise;
     }
 
-    public updateTrialJob(trialJobId: string, form: JobApplicationForm): Promise<TrialJobDetail> {
-        throw new MethodNotImplementedError();
+    public async updateTrialJob(trialJobId: string, form: JobApplicationForm): Promise<TrialJobDetail> {
+        const trialJobDetail: undefined | TrialJobDetail = this.trialJobsMap.get(trialJobId);
+        if (trialJobDetail === undefined) {
+            throw new Error(`updateTrialJob failed: ${trialJobId} not found`);
+        }
+        if (form.jobType === 'TRIAL') {
+                await this.writeParameterFile(trialJobId, (<TrialJobApplicationForm>form).hyperParameters);
+        } else {
+            throw new Error(`updateTrialJob failed: jobType ${form.jobType} not supported.`);
+        }
+
+        return trialJobDetail;
     }
 
     public get isMultiPhaseJobSupported(): boolean {
-        return false;
+        return true;
     }
 
     // tslint:disable:no-http-string
@@ -336,6 +347,9 @@ class PAITrainingService implements TrainingService {
             case TrialConfigMetadataKey.LOG_COLLECTION:
                 this.logCollection = value;
                 break;
+            case TrialConfigMetadataKey.MULTI_PHASE:
+                this.isMultiPhase = (value === 'true' || value === 'True');
+                break;
             default:
                 //Reject for unknown keys
                 throw new Error(`Uknown key: ${key}`);
@@ -445,6 +459,7 @@ class PAITrainingService implements TrainingService {
             trialJobId,
             this.experimentId,
             trialJobDetail.sequenceId,
+            this.isMultiPhase,
             this.paiTrialConfig.command,
             nniManagerIp,
             this.paiRestServerPort,
@@ -632,7 +647,50 @@ class PAITrainingService implements TrainingService {
         return Promise.race([timeoutDelay, deferred.promise])
             .finally(() => { clearTimeout(timeoutId); });
     }
-    // tslint:enable:no-any no-unsafe-any no-http-string
+
+    private async writeParameterFile(trialJobId: string, hyperParameters: HyperParameters): Promise<void> {
+        if (this.paiClusterConfig === undefined) {
+            throw new Error('PAI Cluster config is not initialized');
+        }
+        if (this.paiTrialConfig === undefined) {
+            throw new Error('PAI trial config is not initialized');
+        }
+
+        const trialLocalTempFolder: string = path.join(getExperimentRootDir(), 'trials-local', trialJobId);
+        const hpFileName: string = generateParamFileName(hyperParameters);
+        const localFilepath: string = path.join(trialLocalTempFolder, hpFileName);
+        await fs.promises.writeFile(localFilepath, hyperParameters.value, { encoding: 'utf8' });
+        const hdfsCodeDir: string = HDFSClientUtility.getHdfsTrialWorkDir(this.paiClusterConfig.userName, trialJobId);
+        const hdfsHpFilePath: string = path.join(hdfsCodeDir, hpFileName);
+
+        await HDFSClientUtility.copyFileToHdfs(localFilepath, hdfsHpFilePath, this.hdfsClient);
+
+        await this.postParameterFileMeta({
+            experimentId: this.experimentId,
+            trialId: trialJobId,
+            filePath: hdfsHpFilePath
+        });
+    }
+
+    private postParameterFileMeta(parameterFileMeta: ParameterFileMeta): Promise<void> {
+        const deferred : Deferred<void> = new Deferred<void>();
+        const restServer: PAIJobRestServer = component.get(PAIJobRestServer);
+        const req: request.Options = {
+            uri: `${restServer.endPoint}${restServer.apiRootUrl}/parameter-file-meta`,
+            method: 'POST',
+            json: true,
+            body: parameterFileMeta
+        };
+        request(req, (err: Error, res: request.Response) => {
+            if (err) {
+                deferred.reject(err);
+            } else {
+                deferred.resolve();
+            }
+        });
+
+        return deferred.promise;
+    }
 }
 
 export { PAITrainingService };