Support save and open experiments

docs/en_US/Tutorial/Nnictl.md

@@ -444,9 +444,6 @@ Debug mode will disable version check function in Trialkeeper.
   |--all|  False| |delete all of experiments|
 
 
-
-<a name="export"></a>
-
 * __nnictl experiment export__
   * Description
 
@@ -531,6 +528,62 @@ Debug mode will disable version check function in Trialkeeper.
     nnictl experiment import [experiment_id] -f experiment_data.json
     ```
 
+* __nnictl experiment save__
+  * Description
+
+    Save nni experiment metadata and code data.
+
+  * Usage
+
+    ```bash
+    nnictl experiment save [OPTIONS]
+    ```
+
+  * Options
+
+  |Name, shorthand|Required|Default|Description|
+  |------|------|------ |------|
+  |id|  True| |The id of the experiment you want to save|
+  |--path, -p|  False| |the folder path to store nni experiment data, default current working directory|
+  |--saveCodeDir, -s| False| |save codeDir data of the experiment, default False|
+
+  * Examples
+
+  > save an expeirment
+
+  ```bash
+  nnictl experiment save [experiment_id] --saveCodeDir
+  ```
+
+* __nnictl experiment load__
+  * Description
+
+    Load an nni experiment.
+
+  * Usage
+
+    ```bash
+    nnictl experiment load [OPTIONS]
+    ```
+
+  * Options
+
+  |Name, shorthand|Required|Default|Description|
+  |------|------|------ |------|
+  |--path, -p|  True| |the file path of nni package|
+  |--codeDir, -c| True| |the path of codeDir for loaded experiment, this path will also put the code in the loaded experiment package|
+  |--logDir, -l| False| |the path of logDir for loaded experiment|
+
+  * Examples
+
+  > load an expeirment
+
+  ```bash
+  nnictl experiment load --path [path] --codeDir [codeDir]
+  ```
+
+
+
 <a name="platform"></a>
 ### Manage platform information
 

tools/nni_cmd/common_utils.py

@@ -4,7 +4,10 @@
 import os
 import sys
 import json
+import tempfile
 import socket
+import string
+import random
 import ruamel.yaml as yaml
 import psutil
 from colorama import Fore
@@ -83,3 +86,12 @@ def check_tensorboard_version():
         print_error('import tensorboard error!')
         exit(1)
 
+def generate_temp_dir():
+    '''generate a temp folder'''
+    def generate_folder_name():
+        return os.path.join(tempfile.gettempdir(), 'nni', ''.join(random.sample(string.ascii_letters + string.digits, 8)))
+    temp_dir = generate_folder_name()
+    while os.path.exists(temp_dir):
+        temp_dir = generate_folder_name()
+    os.makedirs(temp_dir)
+    return temp_dir

tools/nni_cmd/config_utils.py

@@ -54,13 +54,13 @@ def __init__(self):
         self.experiment_file = os.path.join(NNICTL_HOME_DIR, '.experiment')
         self.experiments = self.read_file()
 
-    def add_experiment(self, expId, port, time, file_name, platform, experiment_name):
+    def add_experiment(self, expId, port, startTime, file_name, platform, experiment_name, endTime='N/A', status='INITIALIZED'):
         '''set {key:value} paris to self.experiment'''
         self.experiments[expId] = {}
         self.experiments[expId]['port'] = port
-        self.experiments[expId]['startTime'] = time
-        self.experiments[expId]['endTime'] = 'N/A'
-        self.experiments[expId]['status'] = 'INITIALIZED'
+        self.experiments[expId]['startTime'] = startTime
+        self.experiments[expId]['endTime'] = endTime
+        self.experiments[expId]['status'] = status
         self.experiments[expId]['fileName'] = file_name
         self.experiments[expId]['platform'] = platform
         self.experiments[expId]['experimentName'] = experiment_name

tools/nni_cmd/constants.py

@@ -6,6 +6,8 @@
 
 NNICTL_HOME_DIR = os.path.join(os.path.expanduser('~'), '.local', 'nnictl')
 
+NNI_HOME_DIR = os.path.join(os.path.expanduser('~'), 'nni-experiments')
+
 ERROR_INFO = 'ERROR: '
 NORMAL_INFO = 'INFO: '
 WARNING_INFO = 'WARNING: '

tools/nni_cmd/nnictl.py

@@ -11,7 +11,8 @@
 from .nnictl_utils import stop_experiment, trial_ls, trial_kill, list_experiment, experiment_status,\
                           log_trial, experiment_clean, platform_clean, experiment_list, \
                           monitor_experiment, export_trials_data, trial_codegen, webui_url, \
-                          get_config, log_stdout, log_stderr, search_space_auto_gen, webui_nas
+                          get_config, log_stdout, log_stderr, search_space_auto_gen, webui_nas, \
+                          save_experiment, load_experiment
 from .package_management import package_install, package_uninstall, package_show, package_list
 from .constants import DEFAULT_REST_PORT
 from .tensorboard_utils import start_tensorboard, stop_tensorboard
@@ -129,15 +130,6 @@ def parse_args():
     parser_experiment_clean.add_argument('id', nargs='?', help='the id of experiment')
     parser_experiment_clean.add_argument('--all', action='store_true', default=False, help='delete all of experiments')
     parser_experiment_clean.set_defaults(func=experiment_clean)
-
-    #parse experiment command
-    parser_platform = subparsers.add_parser('platform', help='get platform information')
-    #add subparsers for parser_experiment
-    parser_platform_subparsers = parser_platform.add_subparsers()
-    parser_platform_clean = parser_platform_subparsers.add_parser('clean', help='clean up the platform data')
-    parser_platform_clean.add_argument('--config', '-c', required=True, dest='config', help='the path of yaml config file')
-    parser_platform_clean.set_defaults(func=platform_clean)
-
     #import tuning data
     parser_import_data = parser_experiment_subparsers.add_parser('import', help='import additional data')
     parser_import_data.add_argument('id', nargs='?', help='the id of experiment')
@@ -149,6 +141,29 @@ def parse_args():
     parser_trial_export.add_argument('--type', '-t', choices=['json', 'csv'], required=True, dest='type', help='target file type')
     parser_trial_export.add_argument('--filename', '-f', required=True, dest='path', help='target file path')
     parser_trial_export.set_defaults(func=export_trials_data)
+    #save an NNI experiment
+    parser_save_experiment = parser_experiment_subparsers.add_parser('save', help='save an experiment')
+    parser_save_experiment.add_argument('id', nargs='?', help='the id of experiment')
+    parser_save_experiment.add_argument('--path', '-p', required=False, help='the folder path to store nni experiment data, \
+                                   default current working directory')
+    parser_save_experiment.add_argument('--saveCodeDir', '-s', action='store_true', default=False, help='save codeDir data \
+                                   of the experiment')
+    parser_save_experiment.set_defaults(func=save_experiment)
+    #load an NNI experiment
+    parser_load_experiment = parser_experiment_subparsers.add_parser('load', help='load an experiment')
+    parser_load_experiment.add_argument('--path', '-p', required=True, help='the path of nni package file')
+    parser_load_experiment.add_argument('--codeDir', '-c', required=True, help='the path of codeDir for loaded experiment, \
+                                   this path will also put the code in the loaded experiment package')
+    parser_load_experiment.add_argument('--logDir', '-l', required=False, help='the path of logDir for loaded experiment')
+    parser_load_experiment.set_defaults(func=load_experiment)
+
+    #parse platform command
+    parser_platform = subparsers.add_parser('platform', help='get platform information')
+    #add subparsers for parser_platform
+    parser_platform_subparsers = parser_platform.add_subparsers()
+    parser_platform_clean = parser_platform_subparsers.add_parser('clean', help='clean up the platform data')
+    parser_platform_clean.add_argument('--config', '-c', required=True, dest='config', help='the path of yaml config file')
+    parser_platform_clean.set_defaults(func=platform_clean)
 
     #TODO:finish webui function
     #parse board command

tools/nni_cmd/nnictl_utils.py

@@ -18,9 +18,9 @@
 from .rest_utils import rest_get, rest_delete, check_rest_server_quick, check_response
 from .url_utils import trial_jobs_url, experiment_url, trial_job_id_url, export_data_url
 from .config_utils import Config, Experiments
-from .constants import NNICTL_HOME_DIR, EXPERIMENT_INFORMATION_FORMAT, EXPERIMENT_DETAIL_FORMAT, \
+from .constants import NNICTL_HOME_DIR, NNI_HOME_DIR, EXPERIMENT_INFORMATION_FORMAT, EXPERIMENT_DETAIL_FORMAT, \
      EXPERIMENT_MONITOR_INFO, TRIAL_MONITOR_HEAD, TRIAL_MONITOR_CONTENT, TRIAL_MONITOR_TAIL, REST_TIME_OUT
-from .common_utils import print_normal, print_error, print_warning, detect_process, get_yml_content
+from .common_utils import print_normal, print_error, print_warning, detect_process, get_yml_content, generate_temp_dir
 from .command_utils import check_output_command, kill_command
 from .ssh_utils import create_ssh_sftp_client, remove_remote_directory
 
@@ -736,3 +736,165 @@ def search_space_auto_gen(args):
         print_warning('Expected search space file \'{}\' generated, but not found.'.format(file_path))
     else:
         print_normal('Generate search space done: \'{}\'.'.format(file_path))
+
+def save_experiment(args):
+    '''save experiment data to a zip file'''
+    experiment_config = Experiments()
+    experiment_dict = experiment_config.get_all_experiments()
+    if args.id is None:
+        print_error('Please set experiment id.')
+        exit(1)
+    if args.id not in experiment_dict:
+        print_error('Cannot find experiment {0}.'.format(args.id))
+        exit(1)
+    if experiment_dict[args.id].get('status') != 'STOPPED':
+        print_error('Can only save stopped experiment!')
+        exit(1)
+    print_normal('Saving...')
+    nni_config = Config(experiment_dict[args.id]['fileName'])
+    logDir = os.path.join(NNI_HOME_DIR, args.id)
+    if nni_config.get_config('logDir'):
+        logDir = os.path.join(nni_config.get_config('logDir'), args.id)
+    temp_root_dir = generate_temp_dir()
+
+    # Step1. Copy logDir to temp folder
+    if not os.path.exists(logDir):
+        print_error('logDir: %s does not exist!' % logDir)
+        exit(1)
+    temp_experiment_dir = os.path.join(temp_root_dir, 'experiment')
+    shutil.copytree(logDir, temp_experiment_dir)
+
+    # Step2. Copy nnictl metadata to temp folder
+    temp_nnictl_dir = os.path.join(temp_root_dir, 'nnictl')
+    os.makedirs(temp_nnictl_dir, exist_ok=True)
+    try:
+        with open(os.path.join(temp_nnictl_dir, '.experiment'), 'w') as file:
+            experiment_dict[args.id]['id'] = args.id
+            json.dump(experiment_dict[args.id], file)
+    except IOError:
+        print_error('Write file to %s failed!' % os.path.join(temp_nnictl_dir, '.experiment'))
+        exit(1)
+    nnictl_config_dir = os.path.join(NNICTL_HOME_DIR, experiment_dict[args.id]['fileName'])
+    shutil.copytree(nnictl_config_dir, os.path.join(temp_nnictl_dir, experiment_dict[args.id]['fileName']))
+
+    # Step3. Copy code dir
+    if args.saveCodeDir:
+        temp_code_dir = os.path.join(temp_root_dir, 'code')
+        shutil.copytree(nni_config.get_config('experimentConfig')['trial']['codeDir'], temp_code_dir)
+
+    # Step4. Archive folder
+    zip_package_name = 'nni_experiment_%s' % args.id
+    if args.path:
+        os.makedirs(args.path, exist_ok=True)
+        zip_package_name = os.path.join(args.path, zip_package_name)
+    shutil.make_archive(zip_package_name, 'zip', temp_root_dir)
+    print_normal('Save to %s.zip success!' % zip_package_name)
+
+    # Step5. Cleanup temp data
+    shutil.rmtree(temp_root_dir)
+
+def load_experiment(args):
+    '''load experiment data'''
+    package_path = os.path.expanduser(args.path)
+    if not os.path.exists(args.path):
+        print_error('file path %s does not exist!' % args.path)
+        exit(1)
+    temp_root_dir = generate_temp_dir()
+    shutil.unpack_archive(package_path, temp_root_dir)
+    print_normal('Loading...')
+    # Step1. Validation
+    if not os.path.exists(args.codeDir):
+        print_error('Invalid: codeDir path does not exist!')
+        exit(1)
+    if args.logDir:
+        if not os.path.exists(args.logDir):
+            print_error('Invalid: logDir path does not exist!')
+            exit(1)
+    experiment_temp_dir = os.path.join(temp_root_dir, 'experiment')
+    if not os.path.exists(os.path.join(experiment_temp_dir, 'db')):
+        print_error('Invalid archive file: db file does not exist!')
+        shutil.rmtree(temp_root_dir)
+        exit(1)
+    nnictl_temp_dir = os.path.join(temp_root_dir, 'nnictl')
+    if not os.path.exists(os.path.join(nnictl_temp_dir, '.experiment')):
+        print_error('Invalid archive file: nnictl metadata file does not exist!')
+        shutil.rmtree(temp_root_dir)
+        exit(1)
+    try:
+        with open(os.path.join(nnictl_temp_dir, '.experiment'), 'r') as file:
+            experiment_metadata = json.load(file)
+    except ValueError as err:
+        print_error('Invalid nnictl metadata file: %s' % err)
+        shutil.rmtree(temp_root_dir)
+        exit(1)
+    experiment_config = Experiments()
+    experiment_dict = experiment_config.get_all_experiments()
+    experiment_id = experiment_metadata.get('id')
+    if experiment_id in experiment_dict:
+        print_error('Invalid: experiment id already exist!')
+        shutil.rmtree(temp_root_dir)
+        exit(1)
+    if not os.path.exists(os.path.join(nnictl_temp_dir, experiment_metadata.get('fileName'))):
+        print_error('Invalid: experiment metadata does not exist!')
+        shutil.rmtree(temp_root_dir)
+        exit(1)
+
+    # Step2. Copy nnictl metadata
+    src_path = os.path.join(nnictl_temp_dir, experiment_metadata.get('fileName'))
+    dest_path = os.path.join(NNICTL_HOME_DIR, experiment_metadata.get('fileName'))
+    if os.path.exists(dest_path):
+        shutil.rmtree(dest_path)
+    shutil.copytree(src_path, dest_path)
+
+    # Step3. Copy experiment data
+    nni_config = Config(experiment_metadata.get('fileName'))
+    nnictl_exp_config = nni_config.get_config('experimentConfig')
+    if args.logDir:
+        logDir = args.logDir
+        nnictl_exp_config['logDir'] = logDir
+    else:
+        if nnictl_exp_config.get('logDir'):
+            logDir = nnictl_exp_config['logDir']
+        else:
+            logDir = NNI_HOME_DIR
+    os.rename(os.path.join(temp_root_dir, 'experiment'), os.path.join(temp_root_dir, experiment_id))
+    src_path = os.path.join(os.path.join(temp_root_dir, experiment_id))
+    dest_path = os.path.join(os.path.join(logDir, experiment_id))
+    if os.path.exists(dest_path):
+        shutil.rmtree(dest_path)
+    shutil.copytree(src_path, dest_path)
+
+    # Step4. Copy code dir
+    codeDir = os.path.expanduser(args.codeDir)
+    if not os.path.isabs(codeDir):
+        codeDir = os.path.join(os.getcwd(), codeDir)
+        print_normal('Expand codeDir to %s' % codeDir)
+    nnictl_exp_config['trial']['codeDir'] = codeDir
+    archive_code_dir = os.path.join(temp_root_dir, 'code')
+    if os.path.exists(archive_code_dir):
+        file_list = os.listdir(archive_code_dir)
+        for file_name in file_list:
+            src_path = os.path.join(archive_code_dir, file_name)
+            target_path = os.path.join(codeDir, file_name)
+            if os.path.exists(target_path):
+                print_error('Copy %s failed, %s exist!' % (file_name, target_path))
+                continue
+            if os.path.isdir(src_path):
+                shutil.copytree(src_path, target_path)
+            else:
+                shutil.copy(src_path, target_path)
+
+    # Step5. Create experiment metadata
+    nni_config.set_config('experimentConfig', nnictl_exp_config)
+    experiment_config.add_experiment(experiment_id,
+                                     experiment_metadata.get('port'),
+                                     experiment_metadata.get('startTime'),
+                                     experiment_metadata.get('fileName'),
+                                     experiment_metadata.get('platform'),
+                                     experiment_metadata.get('experimentName'),
+                                     experiment_metadata.get('endTime'),
+                                     experiment_metadata.get('status'))
+    print_normal('Load experiment %s succsss!' % experiment_id)
+
+    # Step6. Cleanup temp data
+    shutil.rmtree(temp_root_dir)