support classic nas mode: each chosen arch as a separate trial job (#1775)

docs/en_US/Tutorial/Nnictl.md

@@ -22,6 +22,7 @@ nnictl support commands:
 * [nnictl webui](#webui)
 * [nnictl tensorboard](#tensorboard)
 * [nnictl package](#package)
+* [nnictl ss_gen](#ss_gen)
 * [nnictl --version](#version)
 
 ### Manage an experiment
@@ -733,6 +734,37 @@ Debug mode will disable version check function in Trialkeeper.
     nnictl package show
     ```
 
+<a name="ss_gen"></a>
+
+![](https://placehold.it/15/1589F0/000000?text=+) `Generate search space`
+
+* __nnictl ss_gen__
+  * Description
+
+    Generate search space from user trial code which uses NNI NAS APIs.
+
+  * Usage
+
+    ```bash
+    nnictl ss_gen [OPTIONS]
+    ```
+
+  * Options
+
+  |Name, shorthand|Required|Default|Description|
+  |------|------|------ |------|
+  |--trial_command|  True| |The command of the trial code|
+  |--trial_dir|  False| ./ |The directory of the trial code|
+  |--file|  False| nni_auto_gen_search_space.json |The file for storing generated search space|
+
+  * Example
+
+    > Generate a search space
+
+    ```bash
+    nnictl ss_gen --trial_command="python3 mnist.py" --trial_dir=./ --file=ss.json
+    ```
+
 <a name="version"></a>
 
 ![](https://placehold.it/15/1589F0/000000?text=+) `Check NNI version`

examples/nas/classic_nas/config_nas.yml

+authorName: default
+experimentName: example_mnist
+trialConcurrency: 1
+maxExecDuration: 1h
+maxTrialNum: 10
+#choice: local, remote, pai
+trainingServicePlatform: local
+#please use `nnictl ss_gen` to generate search space file first
+searchSpacePath: <the_generated_search_space_path>
+useAnnotation: False
+tuner:
+  codeDir: ../../tuners/random_nas_tuner
+  classFileName: random_nas_tuner.py
+  className: RandomNASTuner
+trial:
+  command: python3 mnist.py
+  codeDir: .
+  gpuNum: 0

examples/nas/classic_nas/config_ppo.yml

+authorName: default
+experimentName: example_mnist
+trialConcurrency: 1
+maxExecDuration: 100h
+maxTrialNum: 1000
+#choice: local, remote, pai
+trainingServicePlatform: local
+#please use `nnictl ss_gen` to generate search space file first
+searchSpacePath: <the_generated_search_space_path>
+useAnnotation: False
+tuner:
+  builtinTunerName: PPOTuner
+  classArgs:
+    optimize_mode: maximize
+trial:
+  command: python3 mnist.py
+  codeDir: .
+  gpuNum: 0

examples/nas/classic_nas/mnist.py

+"""
+A deep MNIST classifier using convolutional layers.
+
+This file is a modification of the official pytorch mnist example:
+https://github.com/pytorch/examples/blob/master/mnist/main.py
+"""
+
+import os
+import argparse
+import logging
+import nni
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import torch.optim as optim
+from torchvision import datasets, transforms
+
+from nni.nas.pytorch.mutables import LayerChoice, InputChoice
+from nni.nas.pytorch.classic_nas import get_and_apply_next_architecture
+
+
+logger = logging.getLogger('mnist_AutoML')
+
+
+class Net(nn.Module):
+    def __init__(self, hidden_size):
+        super(Net, self).__init__()
+        # two options of conv1
+        self.conv1 = LayerChoice([nn.Conv2d(1, 20, 5, 1),
+                                  nn.Conv2d(1, 20, 3, 1)],
+                                 key='first_conv')
+        # two options of mid_conv
+        self.mid_conv = LayerChoice([nn.Conv2d(20, 20, 3, 1, padding=1),
+                                     nn.Conv2d(20, 20, 5, 1, padding=2)],
+                                    key='mid_conv')
+        self.conv2 = nn.Conv2d(20, 50, 5, 1)
+        self.fc1 = nn.Linear(4*4*50, hidden_size)
+        self.fc2 = nn.Linear(hidden_size, 10)
+        # skip connection over mid_conv
+        self.input_switch = InputChoice(n_candidates=2,
+                                        n_chosen=1,
+                                        key='skip')
+
+    def forward(self, x):
+        x = F.relu(self.conv1(x))
+        x = F.max_pool2d(x, 2, 2)
+        old_x = x
+        x = F.relu(self.mid_conv(x))
+        zero_x = torch.zeros_like(old_x)
+        skip_x = self.input_switch([zero_x, old_x])
+        x = torch.add(x, skip_x)
+        x = F.relu(self.conv2(x))
+        x = F.max_pool2d(x, 2, 2)
+        x = x.view(-1, 4*4*50)
+        x = F.relu(self.fc1(x))
+        x = self.fc2(x)
+        return F.log_softmax(x, dim=1)
+
+
+def train(args, model, device, train_loader, optimizer, epoch):
+    model.train()
+    for batch_idx, (data, target) in enumerate(train_loader):
+        data, target = data.to(device), target.to(device)
+        optimizer.zero_grad()
+        output = model(data)
+        loss = F.nll_loss(output, target)
+        loss.backward()
+        optimizer.step()
+        if batch_idx % args['log_interval'] == 0:
+            logger.info('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(
+                epoch, batch_idx * len(data), len(train_loader.dataset),
+                100. * batch_idx / len(train_loader), loss.item()))
+
+
+def test(args, model, device, test_loader):
+    model.eval()
+    test_loss = 0
+    correct = 0
+    with torch.no_grad():
+        for data, target in test_loader:
+            data, target = data.to(device), target.to(device)
+            output = model(data)
+            # sum up batch loss
+            test_loss += F.nll_loss(output, target, reduction='sum').item()
+            # get the index of the max log-probability
+            pred = output.argmax(dim=1, keepdim=True)
+            correct += pred.eq(target.view_as(pred)).sum().item()
+
+    test_loss /= len(test_loader.dataset)
+
+    accuracy = 100. * correct / len(test_loader.dataset)
+
+    logger.info('\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.0f}%)\n'.format(
+        test_loss, correct, len(test_loader.dataset), accuracy))
+
+    return accuracy
+
+
+def main(args):
+    use_cuda = not args['no_cuda'] and torch.cuda.is_available()
+
+    torch.manual_seed(args['seed'])
+
+    device = torch.device("cuda" if use_cuda else "cpu")
+
+    kwargs = {'num_workers': 1, 'pin_memory': True} if use_cuda else {}
+
+    #data_dir = os.path.join(args['data_dir'], nni.get_trial_id())
+    data_dir = os.path.join(args['data_dir'], 'data')
+
+    train_loader = torch.utils.data.DataLoader(
+        datasets.MNIST(data_dir, train=True, download=True,
+                       transform=transforms.Compose([
+                           transforms.ToTensor(),
+                           transforms.Normalize((0.1307,), (0.3081,))
+                       ])),
+        batch_size=args['batch_size'], shuffle=True, **kwargs)
+    test_loader = torch.utils.data.DataLoader(
+        datasets.MNIST(data_dir, train=False, transform=transforms.Compose([
+            transforms.ToTensor(),
+            transforms.Normalize((0.1307,), (0.3081,))
+        ])),
+        batch_size=1000, shuffle=True, **kwargs)
+
+    hidden_size = args['hidden_size']
+
+    model = Net(hidden_size=hidden_size).to(device)
+    get_and_apply_next_architecture(model)
+    optimizer = optim.SGD(model.parameters(), lr=args['lr'],
+                          momentum=args['momentum'])
+
+    for epoch in range(1, args['epochs'] + 1):
+        train(args, model, device, train_loader, optimizer, epoch)
+        test_acc = test(args, model, device, test_loader)
+
+        if epoch < args['epochs']:
+            # report intermediate result
+            nni.report_intermediate_result(test_acc)
+            logger.debug('test accuracy %g', test_acc)
+            logger.debug('Pipe send intermediate result done.')
+        else:
+            # report final result
+            nni.report_final_result(test_acc)
+            logger.debug('Final result is %g', test_acc)
+            logger.debug('Send final result done.')
+
+
+def get_params():
+    # Training settings
+    parser = argparse.ArgumentParser(description='PyTorch MNIST Example')
+    parser.add_argument("--data_dir", type=str,
+                        default='/tmp/tensorflow/mnist/input_data', help="data directory")
+    parser.add_argument('--batch_size', type=int, default=64, metavar='N',
+                        help='input batch size for training (default: 64)')
+    parser.add_argument("--hidden_size", type=int, default=512, metavar='N',
+                        help='hidden layer size (default: 512)')
+    parser.add_argument('--lr', type=float, default=0.01, metavar='LR',
+                        help='learning rate (default: 0.01)')
+    parser.add_argument('--momentum', type=float, default=0.5, metavar='M',
+                        help='SGD momentum (default: 0.5)')
+    parser.add_argument('--epochs', type=int, default=10, metavar='N',
+                        help='number of epochs to train (default: 10)')
+    parser.add_argument('--seed', type=int, default=1, metavar='S',
+                        help='random seed (default: 1)')
+    parser.add_argument('--no_cuda', action='store_true', default=False,
+                        help='disables CUDA training')
+    parser.add_argument('--log_interval', type=int, default=1000, metavar='N',
+                        help='how many batches to wait before logging training status')
+
+
+    args, _ = parser.parse_known_args()
+    return args
+
+
+if __name__ == '__main__':
+    try:
+        params = vars(get_params())
+        main(params)
+    except Exception as exception:
+        logger.exception(exception)
+        raise

examples/tuners/random_nas_tuner/random_nas_tuner.py

@@ -6,36 +6,27 @@ from nni.tuner import Tuner
 def random_archi_generator(nas_ss, random_state):
     '''random
     '''
-    chosen_archi = {}
-    for block_name, block_value in nas_ss.items():
-        assert block_value['_type'] == "mutable_layer", \
-            "Random NAS Tuner only receives NAS search space whose _type is 'mutable_layer'"
-        block = block_value['_value']
-        tmp_block = {}
-        for layer_name, layer in block.items():
-            tmp_layer = {}
-            for key, value in layer.items():
-                if key == 'layer_choice':
-                    index = random_state.randint(len(value))
-                    tmp_layer['chosen_layer'] = value[index]
-                elif key == 'optional_inputs':
-                    tmp_layer['chosen_inputs'] = []
-                    if layer['optional_inputs']:
-                        if isinstance(layer['optional_input_size'], int):
-                            choice_num = layer['optional_input_size']
-                        else:
-                            choice_range = layer['optional_input_size']
-                            choice_num = random_state.randint(choice_range[0], choice_range[1] + 1)
-                        for _ in range(choice_num):
-                            index = random_state.randint(len(layer['optional_inputs']))
-                            tmp_layer['chosen_inputs'].append(layer['optional_inputs'][index])
-                elif key == 'optional_input_size':
-                    pass
-                else:
-                    raise ValueError('Unknown field %s in layer %s of block %s' % (key, layer_name, block_name))
-            tmp_block[layer_name] = tmp_layer
-        chosen_archi[block_name] = tmp_block
-    return chosen_archi
+    chosen_arch = {}
+    for key, val in nas_ss.items():
+        assert val['_type'] in ['layer_choice', 'input_choice'], \
+            "Random NAS Tuner only receives NAS search space whose _type is 'layer_choice' or 'input_choice'"
+        if val['_type'] == 'layer_choice':
+            choices = val['_value']
+            index = random_state.randint(len(choices))
+            chosen_arch[key] = {'_value': choices[index], '_idx': index}
+        elif val['_type'] == 'input_choice':
+            choices = val['_value']['candidates']
+            n_chosen = val['_value']['n_chosen']
+            chosen = []
+            idxs = []
+            for _ in range(n_chosen):
+                index = random_state.randint(len(choices))
+                chosen.append(choices[index])
+                idxs.append(index)
+            chosen_arch[key] = {'_value': chosen, '_idx': idxs}
+        else:
+            raise ValueError('Unknown key %s and value %s' % (key, val))
+    return chosen_arch
 
 
 class RandomNASTuner(Tuner):

src/sdk/pynni/nni/nas/pytorch/classic_nas/__init__.py

+from .mutator import get_and_apply_next_architecture

src/sdk/pynni/nni/nas/pytorch/classic_nas/mutator.py

+import os
+import sys
+import json
+import logging
+import torch
+import nni
+from nni.env_vars import trial_env_vars
+from nni.nas.pytorch.base_mutator import BaseMutator
+from nni.nas.pytorch.mutables import LayerChoice, InputChoice
+
+logger = logging.getLogger(__name__)
+
+def get_and_apply_next_architecture(model):
+    """
+    Wrapper of ClassicMutator to make it more meaningful,
+    similar to ```get_next_parameter``` for HPO.
+
+    Parameters
+    ----------
+    model : pytorch model
+        user's model with search space (e.g., LayerChoice, InputChoice) embedded in it
+    """
+    ClassicMutator(model)
+
+class ClassicMutator(BaseMutator):
+    """
+    This mutator is to apply the architecture chosen from tuner.
+    It implements the forward function of LayerChoice and InputChoice,
+    to only activate the chosen ones
+    """
+    def __init__(self, model):
+        """
+        Generate search space based on ```model```.
+        If env ```NNI_GEN_SEARCH_SPACE``` exists, this is in dry run mode for
+        generating search space for the experiment.
+        If not, there are still two mode, one is nni experiment mode where users
+        use ```nnictl``` to start an experiment. The other is standalone mode
+        where users directly run the trial command, this mode chooses the first
+        one(s) for each LayerChoice and InputChoice.
+
+        Parameters
+        ----------
+        model : pytorch model
+            user's model with search space (e.g., LayerChoice, InputChoice) embedded in it
+        """
+        super(ClassicMutator, self).__init__(model)
+        self.chosen_arch = {}
+        self.search_space = self._generate_search_space()
+        if 'NNI_GEN_SEARCH_SPACE' in os.environ:
+            # dry run for only generating search space
+            self._dump_search_space(self.search_space, os.environ.get('NNI_GEN_SEARCH_SPACE'))
+            sys.exit(0)
+        # get chosen arch from tuner
+        self.chosen_arch = nni.get_next_parameter()
+        if not self.chosen_arch and trial_env_vars.NNI_PLATFORM is None:
+            logger.warning('This is in standalone mode, the chosen are the first one(s)')
+            self.chosen_arch = self._standalone_generate_chosen()
+        self._validate_chosen_arch()
+
+    def _validate_chosen_arch(self):
+        pass
+
+    def _standalone_generate_chosen(self):
+        """
+        Generate the chosen architecture for standalone mode,
+        i.e., choose the first one(s) for LayerChoice and InputChoice
+
+        { key_name: {'_value': "conv1",
+                     '_idx': 0} }
+
+        { key_name: {'_value': ["in1"],
+                     '_idx': [0]} }
+
+        Returns
+        -------
+        dict
+            the chosen architecture
+        """
+        chosen_arch = {}
+        for key, val in self.search_space.items():
+            if val['_type'] == 'layer_choice':
+                choices = val['_value']
+                chosen_arch[key] = {'_value': choices[0], '_idx': 0}
+            elif val['_type'] == 'input_choice':
+                choices = val['_value']['candidates']
+                n_chosen = val['_value']['n_chosen']
+                chosen_arch[key] = {'_value': choices[:n_chosen], '_idx': list(range(n_chosen))}
+            else:
+                raise ValueError('Unknown key %s and value %s' % (key, val))
+        return chosen_arch
+
+    def _generate_search_space(self):
+        """
+        Generate search space from mutables.
+        Here is the search space format:
+
+        { key_name: {'_type': 'layer_choice',
+                     '_value': ["conv1", "conv2"]} }
+
+        { key_name: {'_type': 'input_choice',
+                     '_value': {'candidates': ["in1", "in2"],
+                                'n_chosen': 1}} }
+
+        Returns
+        -------
+        dict
+            the generated search space
+        """
+        search_space = {}
+        for mutable in self.mutables:
+            # for now we only generate flattened search space
+            if isinstance(mutable, LayerChoice):
+                key = mutable.key
+                val = [repr(choice) for choice in mutable.choices]
+                search_space[key] = {"_type": "layer_choice", "_value": val}
+            elif isinstance(mutable, InputChoice):
+                key = mutable.key
+                search_space[key] = {"_type": "input_choice",
+                                     "_value": {"candidates": mutable.choose_from,
+                                                "n_chosen": mutable.n_chosen}}
+            else:
+                raise TypeError('Unsupported mutable type: %s.' % type(mutable))
+        return search_space
+
+    def _dump_search_space(self, search_space, file_path):
+        with open(file_path, 'w') as ss_file:
+            json.dump(search_space, ss_file)
+
+    def _tensor_reduction(self, reduction_type, tensor_list):
+        if tensor_list == "none":
+            return tensor_list
+        if not tensor_list:
+            return None  # empty. return None for now
+        if len(tensor_list) == 1:
+            return tensor_list[0]
+        if reduction_type == "sum":
+            return sum(tensor_list)
+        if reduction_type == "mean":
+            return sum(tensor_list) / len(tensor_list)
+        if reduction_type == "concat":
+            return torch.cat(tensor_list, dim=1)
+        raise ValueError("Unrecognized reduction policy: \"{}\"".format(reduction_type))
+
+    def on_forward_layer_choice(self, mutable, *inputs):
+        """
+        Implement the forward of LayerChoice
+
+        Parameters
+        ----------
+        mutable: LayerChoice
+        inputs: list of torch.Tensor
+
+        Returns
+        -------
+        tuple
+            return of the chosen op, the index of the chosen op
+
+        """
+        assert mutable.key in self.chosen_arch
+        val = self.chosen_arch[mutable.key]
+        assert isinstance(val, dict)
+        idx = val['_idx']
+        assert self.search_space[mutable.key]['_value'][idx] == val['_value']
+        return mutable.choices[idx](*inputs), idx
+
+    def on_forward_input_choice(self, mutable, tensor_list):
+        """
+        Implement the forward of InputChoice
+
+        Parameters
+        ----------
+        mutable: InputChoice
+        tensor_list: list of torch.Tensor
+        tags: list of string
+
+        Returns
+        -------
+        tuple of torch.Tensor and list
+            reduced tensor, mask list
+
+        """
+        assert mutable.key in self.chosen_arch
+        val = self.chosen_arch[mutable.key]
+        assert isinstance(val, dict)
+        mask = [0 for _ in range(mutable.n_candidates)]
+        out = []
+        for i, idx in enumerate(val['_idx']):
+            # check whether idx matches the chosen candidate name
+            assert self.search_space[mutable.key]['_value']['candidates'][idx] == val['_value'][i]
+            out.append(tensor_list[idx])
+            mask[idx] = 1
+        return self._tensor_reduction(mutable.reduction, out), mask

src/sdk/pynni/nni/ppo_tuner/ppo_tuner.py

@@ -351,74 +351,33 @@ class PPOTuner(Tuner):
         self.send_trial_callback = None
         logger.info('Finished PPOTuner initialization')
 
-    def _process_one_nas_space(self, block_name, block_space):
-        """
-        Process nas space to determine observation space and action space
-
-        Parameters
-        ----------
-        block_name : str
-            The name of the mutable block
-        block_space : dict
-            Search space of this mutable block
-
-        Returns
-        -------
-        actions_spaces : list
-            List of the space of each action
-        actions_to_config : list
-            The mapping from action to generated configuration
-        """
-        actions_spaces = []
-        actions_to_config = []
-
-        block_arch_temp = {}
-        for l_name, layer in block_space.items():
-            chosen_layer_temp = {}
-
-            if len(layer['layer_choice']) > 1:
-                actions_spaces.append(layer['layer_choice'])
-                actions_to_config.append((block_name, l_name, 'chosen_layer'))
-                chosen_layer_temp['chosen_layer'] = None
-            else:
-                assert len(layer['layer_choice']) == 1
-                chosen_layer_temp['chosen_layer'] = layer['layer_choice'][0]
-
-            if layer['optional_input_size'] not in [0, 1, [0, 1]]:
-                raise ValueError('Optional_input_size can only be 0, 1, or [0, 1], but the pecified one is %s'
-                                 % (layer['optional_input_size']))
-            if isinstance(layer['optional_input_size'], list):
-                actions_spaces.append(["None", *layer['optional_inputs']])
-                actions_to_config.append((block_name, l_name, 'chosen_inputs'))
-                chosen_layer_temp['chosen_inputs'] = None
-            elif layer['optional_input_size'] == 1:
-                actions_spaces.append(layer['optional_inputs'])
-                actions_to_config.append((block_name, l_name, 'chosen_inputs'))
-                chosen_layer_temp['chosen_inputs'] = None
-            elif layer['optional_input_size'] == 0:
-                chosen_layer_temp['chosen_inputs'] = []
-            else:
-                raise ValueError('invalid type and value of optional_input_size')
-
-            block_arch_temp[l_name] = chosen_layer_temp
-
-        self.chosen_arch_template[block_name] = block_arch_temp
-
-        return actions_spaces, actions_to_config
-
     def _process_nas_space(self, search_space):
-        """
-        Process nas search space to get action/observation space
-        """
         actions_spaces = []
         actions_to_config = []
-        for b_name, block in search_space.items():
-            if block['_type'] != 'mutable_layer':
-                raise ValueError('PPOTuner only accept mutable_layer type in search space, but the current one is %s'%(block['_type']))
-            block = block['_value']
-            act, act_map = self._process_one_nas_space(b_name, block)
-            actions_spaces.extend(act)
-            actions_to_config.extend(act_map)
+        for key, val in search_space.items():
+            if val['_type'] == 'layer_choice':
+                actions_to_config.append((key, 'layer_choice'))
+                actions_spaces.append(val['_value'])
+                self.chosen_arch_template[key] = None
+            elif val['_type'] == 'input_choice':
+                candidates = val['_value']['candidates']
+                n_chosen = val['_value']['n_chosen']
+                if n_chosen not in [0, 1, [0, 1]]:
+                    raise ValueError('Optional_input_size can only be 0, 1, or [0, 1], but the pecified one is %s'
+                                     % (n_chosen))
+                if isinstance(n_chosen, list):
+                    actions_to_config.append((key, 'input_choice'))
+                    # FIXME: risk, candidates might also have None
+                    actions_spaces.append(['None', *candidates])
+                    self.chosen_arch_template[key] = None
+                elif n_chosen == 1:
+                    actions_to_config.append((key, 'input_choice'))
+                    actions_spaces.append(candidates)
+                    self.chosen_arch_template[key] = None
+                elif n_chosen == 0:
+                    self.chosen_arch_template[key] = []
+            else:
+                raise ValueError('Unsupported search space type: %s' % (val['_type']))
 
         # calculate observation space
         dedup = {}
@@ -428,7 +387,6 @@ class PPOTuner(Tuner):
         full_act_space = [act for act, _ in dedup.items()]
         assert len(full_act_space) == len(dedup)
         observation_space = len(full_act_space)
-
         nsteps = len(actions_spaces)
 
         return actions_spaces, actions_to_config, full_act_space, observation_space, nsteps
@@ -470,7 +428,7 @@ class PPOTuner(Tuner):
             Search space for NAS
             the format could be referred to search space spec (https://nni.readthedocs.io/en/latest/Tutorial/SearchSpaceSpec.html).
         """
-        logger.info('=== update search space %s', search_space)
+        logger.info('update search space %s', search_space)
         assert self.search_space is None
         self.search_space = search_space
 
@@ -496,16 +454,19 @@ class PPOTuner(Tuner):
         chosen_arch = copy.deepcopy(self.chosen_arch_template)
         for cnt, act in enumerate(actions):
             act_name = self.full_act_space[act]
-            (block_name, layer_name, key) = self.actions_to_config[cnt]
-            if key == 'chosen_inputs':
+            (_key, _type) = self.actions_to_config[cnt]
+            if _type == 'input_choice':
                 if act_name == 'None':
-                    chosen_arch[block_name][layer_name][key] = []
+                    chosen_arch[_key] = {'_value': [], '_idx': []}
                 else:
-                    chosen_arch[block_name][layer_name][key] = [act_name]
-            elif key == 'chosen_layer':
-                chosen_arch[block_name][layer_name][key] = act_name
+                    candidates = self.search_space[_key]['_value']['candidates']
+                    idx = candidates.index(act_name)
+                    chosen_arch[_key] = {'_value': [act_name], '_idx': [idx]}
+            elif _type == 'layer_choice':
+                idx = self.search_space[_key]['_value'].index(act_name)
+                chosen_arch[_key] = {'_value': act_name, '_idx': idx}
             else:
-                raise ValueError('unrecognized key: {0}'.format(key))
+                raise ValueError('unrecognized key: {0}'.format(_type))
         return chosen_arch
 
     def generate_multiple_parameters(self, parameter_id_list, **kwargs):
@@ -561,6 +522,7 @@ class PPOTuner(Tuner):
 
         trial_info_idx, actions = self.trials_info.get_next()
         if trial_info_idx is None:
+            logger.debug('Credit added by one in parameters request')
             self.credit += 1
             self.param_ids.append(parameter_id)
             raise nni.NoMoreTrialError('no more parameters now.')
@@ -573,6 +535,7 @@ class PPOTuner(Tuner):
         """
         Run a inference to generate next batch of configurations
         """
+        logger.debug('Start next round inference...')
         self.finished_trials = 0
         self.model.compute_rewards(self.trials_info, self.trials_result)
         self.model.train(self.trials_info, self.inf_batch_size)
@@ -584,6 +547,7 @@ class PPOTuner(Tuner):
                                       mb_values, mb_neglogpacs,
                                       mb_dones, last_values,
                                       self.inf_batch_size)
+        logger.debug('Next round inference complete.')
         # check credit and submit new trials
         for _ in range(self.credit):
             trial_info_idx, actions = self.trials_info.get_next()
@@ -596,6 +560,7 @@ class PPOTuner(Tuner):
             new_config = self._actions_to_config(actions)
             self.send_trial_callback(param_id, new_config)
             self.credit -= 1
+            logger.debug('Send new trial (%d, %s) for reducing credit', param_id, new_config)
 
     def receive_trial_result(self, parameter_id, parameters, value, **kwargs):
         """
@@ -621,7 +586,10 @@ class PPOTuner(Tuner):
         self.trials_result[trial_info_idx] = value
         self.finished_trials += 1
 
+        logger.debug('receive_trial_result, parameter_id %d, trial_info_idx %d, finished_trials %d, inf_batch_size %d',
+                     parameter_id, trial_info_idx, self.finished_trials, self.inf_batch_size)
         if self.finished_trials == self.inf_batch_size:
+            logger.debug('Start next round inference in receive_trial_result')
             self._next_round_inference()
 
     def trial_end(self, parameter_id, success, **kwargs):
@@ -650,6 +618,7 @@ class PPOTuner(Tuner):
             self.trials_result[trial_info_idx] = (sum(values) / len(values)) if values else 0
             self.finished_trials += 1
             if self.finished_trials == self.inf_batch_size:
+                logger.debug('Start next round inference in trial_end')
                 self._next_round_inference()
 
     def import_data(self, data):

tools/nni_cmd/nnictl.py

@@ -10,7 +10,8 @@ from .launcher import create_experiment, resume_experiment, view_experiment
 from .updater import update_searchspace, update_concurrency, update_duration, update_trialnum, import_data
 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
+                          monitor_experiment, export_trials_data, trial_codegen, webui_url, \
+                          get_config, log_stdout, log_stderr, search_space_auto_gen
 from .package_management import package_install, package_show
 from .constants import DEFAULT_REST_PORT
 from .tensorboard_utils import start_tensorboard, stop_tensorboard
@@ -38,6 +39,13 @@ def parse_args():
     # create subparsers for args with sub values
     subparsers = parser.add_subparsers()
 
+    # parse the command of auto generating search space
+    parser_start = subparsers.add_parser('ss_gen', help='automatically generate search space file from trial code')
+    parser_start.add_argument('--trial_command', '-t', required=True, dest='trial_command', help='the command for running trial code')
+    parser_start.add_argument('--trial_dir', '-d', default='./', dest='trial_dir', help='the directory for running the command')
+    parser_start.add_argument('--file', '-f', default='nni_auto_gen_search_space.json', dest='file', help='the path of search space file')
+    parser_start.set_defaults(func=search_space_auto_gen)
+
     # parse start command
     parser_start = subparsers.add_parser('create', help='create a new experiment')
     parser_start.add_argument('--config', '-c', required=True, dest='config', help='the path of yaml config file')

tools/nni_cmd/nnictl_utils.py

@@ -9,6 +9,7 @@ import re
 import shutil
 from datetime import datetime, timezone
 from pathlib import Path
+from subprocess import Popen
 from pyhdfs import HdfsClient
 from nni_annotation import expand_annotations
 from .rest_utils import rest_get, rest_delete, check_rest_server_quick, check_response
@@ -675,3 +676,16 @@ def export_trials_data(args):
             print_error('Export failed...')
     else:
         print_error('Restful server is not Running')
+
+def search_space_auto_gen(args):
+    '''dry run trial code to generate search space file'''
+    trial_dir = os.path.expanduser(args.trial_dir)
+    file_path = os.path.expanduser(args.file)
+    if not os.path.isabs(file_path):
+        abs_file_path = os.path.join(os.getcwd(), file_path)
+    assert os.path.exists(trial_dir)
+    if os.path.exists(abs_file_path):
+        print_warning('%s already exits, will be over written' % abs_file_path)
+    print_normal('Dry run to generate search space...')
+    Popen(args.trial_command, cwd=trial_dir, env=dict(os.environ, NNI_GEN_SEARCH_SPACE=abs_file_path), shell=True).wait()
+    print_normal('Dry run to generate search space, Done')
\ No newline at end of file