BuiltinTuner.md 29 KB
Newer Older
Scarlett Li's avatar
Scarlett Li committed
1
# HyperParameter Tuning with NNI Built-in Tuners
Yan Ni's avatar
Yan Ni committed
2

Scarlett Li's avatar
Scarlett Li committed
3
To fit a machine/deep learning model into different tasks/problems, hyperparameters always need to be tuned. Automating the process of hyperparaeter tuning always requires a good tuning algorithm. NNI has provided state-of-the-art tuning algorithms as part of our built-in tuners and makes them easy to use. Below is the brief summary of NNI's current built-in tuners:
Yan Ni's avatar
Yan Ni committed
4

5
Note: Click the **Tuner's name** to get the Tuner's installation requirements, suggested scenario, and an example configuration. A link for a detailed description of each algorithm is located at the end of the suggested scenario for each tuner. Here is an [article](../CommunitySharings/HpoComparison.md) comparing different Tuners on several problems.
6

7
Currently, we support the following algorithms:
8

Yan Ni's avatar
Yan Ni committed
9
10
|Tuner|Brief Introduction of Algorithm|
|---|---|
Yan Ni's avatar
Yan Ni committed
11
12
13
|[__TPE__](#TPE)|The Tree-structured Parzen Estimator (TPE) is a sequential model-based optimization (SMBO) approach. SMBO methods sequentially construct models to approximate the performance of hyperparameters based on historical measurements, and then subsequently choose new hyperparameters to test based on this model. [Reference Paper](https://papers.nips.cc/paper/4443-algorithms-for-hyper-parameter-optimization.pdf)|
|[__Random Search__](#Random)|In Random Search for Hyper-Parameter Optimization show that Random Search might be surprisingly simple and effective. We suggest that we could use Random Search as the baseline when we have no knowledge about the prior distribution of hyper-parameters. [Reference Paper](http://www.jmlr.org/papers/volume13/bergstra12a/bergstra12a.pdf)|
|[__Anneal__](#Anneal)|This simple annealing algorithm begins by sampling from the prior, but tends over time to sample from points closer and closer to the best ones observed. This algorithm is a simple variation on the random search that leverages smoothness in the response surface. The annealing rate is not adaptive.|
14
15
|[__Naïve Evolution__](#Evolution)|Naïve Evolution comes from Large-Scale Evolution of Image Classifiers. It randomly initializes a population-based on search space. For each generation, it chooses better ones and does some mutation (e.g., change a hyperparameter, add/remove one layer) on them to get the next generation. Naïve Evolution requires many trials to work, but it's very simple and easy to expand new features. [Reference paper](https://arxiv.org/pdf/1703.01041.pdf)|
|[__SMAC__](#SMAC)|SMAC is based on Sequential Model-Based Optimization (SMBO). It adapts the most prominent previously used model class (Gaussian stochastic process models) and introduces the model class of random forests to SMBO, in order to handle categorical parameters. The SMAC supported by NNI is a wrapper on the SMAC3 GitHub repo. Notice, SMAC needs to be installed by `nnictl package` command. [Reference Paper,](https://www.cs.ubc.ca/~hutter/papers/10-TR-SMAC.pdf) [GitHub Repo](https://github.com/automl/SMAC3)|
Yan Ni's avatar
Yan Ni committed
16
|[__Batch tuner__](#Batch)|Batch tuner allows users to simply provide several configurations (i.e., choices of hyper-parameters) for their trial code. After finishing all the configurations, the experiment is done. Batch tuner only supports the type choice in search space spec.|
17
|[__Grid Search__](#GridSearch)|Grid Search performs an exhaustive searching through a manually specified subset of the hyperparameter space defined in the searchspace file. Note that the only acceptable types of search space are choice, quniform, randint. |
18
19
|[__Hyperband__](#Hyperband)|Hyperband tries to use limited resources to explore as many configurations as possible and returns the most promising ones as a final result. The basic idea is to generate many configurations and run them for a small number of trials. The half least-promising configurations are thrown out, the remaining are further trained along with a selection of new configurations. The size of these populations is sensitive to resource constraints (e.g. allotted search time). [Reference Paper](https://arxiv.org/pdf/1603.06560.pdf)|
|[__Network Morphism__](#NetworkMorphism)|Network Morphism provides functions to automatically search for deep learning architectures. It generates child networks that inherit the knowledge from their parent network which it is a morph from. This includes changes in depth, width, and skip-connections. Next, it estimates the value of a child network using historic architecture and metric pairs. Then it selects the most promising one to train. [Reference Paper](https://arxiv.org/abs/1806.10282)|
Yan Ni's avatar
Yan Ni committed
20
|[__Metis Tuner__](#MetisTuner)|Metis offers the following benefits when it comes to tuning parameters: While most tools only predict the optimal configuration, Metis gives you two outputs: (a) current prediction of optimal configuration, and (b) suggestion for the next trial. No more guesswork. While most tools assume training datasets do not have noisy data, Metis actually tells you if you need to re-sample a particular hyper-parameter. [Reference Paper](https://www.microsoft.com/en-us/research/publication/metis-robustly-tuning-tail-latencies-cloud-systems/)|
21
|[__BOHB__](#BOHB)|BOHB is a follow-up work to Hyperband. It targets the weakness of Hyperband that new configurations are generated randomly without leveraging finished trials. For the name BOHB, HB means Hyperband, BO means Bayesian Optimization. BOHB leverages finished trials by building multiple TPE models, a proportion of new configurations are generated through these models. [Reference Paper](https://arxiv.org/abs/1807.01774)|
Chi Song's avatar
Chi Song committed
22
|[__GP Tuner__](#GPTuner)|Gaussian Process Tuner is a sequential model-based optimization (SMBO) approach with Gaussian Process as the surrogate. [Reference Paper](https://papers.nips.cc/paper/4443-algorithms-for-hyper-parameter-optimization.pdf), [Github Repo](https://github.com/fmfn/BayesianOptimization)|
Chi Song's avatar
Chi Song committed
23
|[__PPO Tuner__](#PPOTuner)|PPO Tuner is a Reinforcement Learning tuner based on PPO algorithm. [Reference Paper](https://arxiv.org/abs/1707.06347)|
RayMeng8's avatar
RayMeng8 committed
24
|[__PBT Tuner__](#PBTTuner)|PBT Tuner is a simple asynchronous optimization algorithm which effectively utilizes a fixed computational budget to jointly optimize a population of models and their hyperparameters to maximize performance. [Reference Paper](https://arxiv.org/abs/1711.09846v1)|
Yan Ni's avatar
Yan Ni committed
25

Chi Song's avatar
Chi Song committed
26
## Usage of Built-in Tuners
Yan Ni's avatar
Yan Ni committed
27

28
Using a built-in tuner provided by the NNI SDK requires one to declare the  **builtinTunerName** and **classArgs** in the `config.yml` file. In this part, we will introduce each tuner along with information about usage and suggested scenarios, classArg requirements, and an example configuration.
Yan Ni's avatar
Yan Ni committed
29

30
Note: Please follow the format when you write your `config.yml` file. Some built-in tuners need to be installed using `nnictl package`, like SMAC.
Yan Ni's avatar
Yan Ni committed
31
32
33

<a name="TPE"></a>

34
### TPE
Yan Ni's avatar
Yan Ni committed
35

Chi Song's avatar
Chi Song committed
36
> Built-in Tuner Name: **TPE**
Yan Ni's avatar
Yan Ni committed
37
38
39

**Suggested scenario**

40
TPE, as a black-box optimization, can be used in various scenarios and shows good performance in general. Especially when you have limited computation resources and can only try a small number of trials. From a large amount of experiments, we found that TPE is far better than Random Search. [Detailed Description](./HyperoptTuner.md)
41

Yan Ni's avatar
Yan Ni committed
42

43
**classArgs Requirements:**
Yan Ni's avatar
Yan Ni committed
44

45
* **optimize_mode** (*maximize or minimize, optional, default = maximize*) - If 'maximize', the tuner will try to maximize metrics. If 'minimize', the tuner will try to minimize metrics.
Yan Ni's avatar
Yan Ni committed
46

47
Note: We have optimized the parallelism of TPE for large-scale trial concurrency. For the principle of optimization or turn-on optimization, please refer to [TPE document](./HyperoptTuner.md).
xuehui's avatar
xuehui committed
48

49
**Example Configuration:**
Yan Ni's avatar
Yan Ni committed
50

Yan Ni's avatar
Yan Ni committed
51
```yaml
Yan Ni's avatar
Yan Ni committed
52
53
54
55
56
57
58
59
60
61
62
# config.yml
tuner:
  builtinTunerName: TPE
  classArgs:
    optimize_mode: maximize
```

<br>

<a name="Random"></a>

63
### Random Search
Yan Ni's avatar
Yan Ni committed
64

Chi Song's avatar
Chi Song committed
65
> Built-in Tuner Name: **Random**
Yan Ni's avatar
Yan Ni committed
66
67
68

**Suggested scenario**

69
Random search is suggested when each trial does not take very long (e.g., each trial can be completed very quickly, or early stopped by the assessor), and you have enough computational resources. It's also useful if you want to uniformly explore the search space. Random Search can be considered a baseline search algorithm. [Detailed Description](./HyperoptTuner.md)
Yan Ni's avatar
Yan Ni committed
70

71
**Example Configuration:**
Yan Ni's avatar
Yan Ni committed
72

Yan Ni's avatar
Yan Ni committed
73
```yaml
Yan Ni's avatar
Yan Ni committed
74
75
76
77
78
79
80
81
82
# config.yml
tuner:
  builtinTunerName: Random
```

<br>

<a name="Anneal"></a>

83
### Anneal
Yan Ni's avatar
Yan Ni committed
84

Chi Song's avatar
Chi Song committed
85
> Built-in Tuner Name: **Anneal**
Yan Ni's avatar
Yan Ni committed
86
87
88

**Suggested scenario**

89
Anneal is suggested when each trial does not take very long and you have enough computation resources (very similar to Random Search). It's also useful when the variables in the search space can be sample from some prior distribution. [Detailed Description](./HyperoptTuner.md)
90

Yan Ni's avatar
Yan Ni committed
91

92
**classArgs Requirements:**
Yan Ni's avatar
Yan Ni committed
93

94
* **optimize_mode** (*maximize or minimize, optional, default = maximize*) - If 'maximize', the tuner will try to maximize metrics. If 'minimize', the tuner will try to minimize metrics.
Yan Ni's avatar
Yan Ni committed
95

96
**Example Configuration:**
Yan Ni's avatar
Yan Ni committed
97

Yan Ni's avatar
Yan Ni committed
98
```yaml
Yan Ni's avatar
Yan Ni committed
99
100
101
102
103
104
105
106
107
108
109
# config.yml
tuner:
  builtinTunerName: Anneal
  classArgs:
    optimize_mode: maximize
```

<br>

<a name="Evolution"></a>

110
### Naïve Evolution
Yan Ni's avatar
Yan Ni committed
111

Chi Song's avatar
Chi Song committed
112
> Built-in Tuner Name: **Evolution**
Yan Ni's avatar
Yan Ni committed
113
114
115

**Suggested scenario**

116
Its computational resource requirements are relatively high. Specifically, it requires a large initial population to avoid falling into a local optimum. If your trial is short or leverages assessor, this tuner is a good choice. It is also suggested when your trial code supports weight transfer; that is, the trial could inherit the converged weights from its parent(s). This can greatly speed up the training process. [Detailed Description](./EvolutionTuner.md)
117

118
**classArgs Requirements:**
xuehui's avatar
xuehui committed
119

120
* **optimize_mode** (*maximize or minimize, optional, default = maximize*) - If 'maximize', the tuner will try to maximize metrics. If 'minimize', the tuner will try to minimize metrics.
xuehui's avatar
xuehui committed
121

122
* **population_size** (*int value (should > 0), optional, default = 20*) - the initial size of the population (trial num) in the evolution tuner. It's suggested that `population_size` be much larger than `concurrency` so users can get the most out of the algorithm (and at least `concurrency`, or the tuner will fail on its first generation of parameters).
Yan Ni's avatar
Yan Ni committed
123

124
**Example Configuration:**
Yan Ni's avatar
Yan Ni committed
125

Yan Ni's avatar
Yan Ni committed
126
```yaml
Yan Ni's avatar
Yan Ni committed
127
128
129
130
131
# config.yml
tuner:
  builtinTunerName: Evolution
  classArgs:
    optimize_mode: maximize
xuehui's avatar
xuehui committed
132
    population_size: 100
Yan Ni's avatar
Yan Ni committed
133
134
135
136
137
138
```

<br>

<a name="SMAC"></a>

139
### SMAC
Yan Ni's avatar
Yan Ni committed
140

Chi Song's avatar
Chi Song committed
141
> Built-in Tuner Name: **SMAC**
Yan Ni's avatar
Yan Ni committed
142

143
**Please note that SMAC doesn't support running on Windows currently. For the specific reason, please refer to this [GitHub issue](https://github.com/automl/SMAC3/issues/483).**
144

145
146
**Installation**

147
SMAC needs to be installed by following command before the first usage. As a reminder, `swig` is required for SMAC: for Ubuntu `swig` can be installed with `apt`.
148
149
150
151
152

```bash
nnictl package install --name=SMAC
```

Yan Ni's avatar
Yan Ni committed
153
154
**Suggested scenario**

155
Similar to TPE, SMAC is also a black-box tuner that can be tried in various scenarios and is suggested when computational resources are limited. It is optimized for discrete hyperparameters, thus, it's suggested when most of your hyperparameters are discrete. [Detailed Description](./SmacTuner.md)
Yan Ni's avatar
Yan Ni committed
156

157
**classArgs Requirements:**
Yan Ni's avatar
Yan Ni committed
158

159
160
* **optimize_mode** (*maximize or minimize, optional, default = maximize*) - If 'maximize', the tuner will try to maximize metrics. If 'minimize', the tuner will try to minimize metrics.
* **config_dedup** (*True or False, optional, default = False*) - If True, the tuner will not generate a configuration that has been already generated. If False, a configuration may be generated twice, but it is rare for a relatively large search space.
Yan Ni's avatar
Yan Ni committed
161

162
**Example Configuration:**
Yan Ni's avatar
Yan Ni committed
163

Yan Ni's avatar
Yan Ni committed
164
```yaml
Yan Ni's avatar
Yan Ni committed
165
166
167
168
169
170
171
172
173
174
175
# config.yml
tuner:
  builtinTunerName: SMAC
  classArgs:
    optimize_mode: maximize
```

<br>

<a name="Batch"></a>

176
### Batch Tuner
Yan Ni's avatar
Yan Ni committed
177

Chi Song's avatar
Chi Song committed
178
> Built-in Tuner Name: BatchTuner
Yan Ni's avatar
Yan Ni committed
179
180
181

**Suggested scenario**

182
If the configurations you want to try have been decided beforehand, you can list them in search space file (using `choice`) and run them using batch tuner. [Detailed Description](./BatchTuner.md)
Yan Ni's avatar
Yan Ni committed
183

184
**Example Configuration:**
Yan Ni's avatar
Yan Ni committed
185

Yan Ni's avatar
Yan Ni committed
186
```yaml
Yan Ni's avatar
Yan Ni committed
187
188
189
190
191
192
193
# config.yml
tuner:
  builtinTunerName: BatchTuner
```

<br>

194
Note that the search space for BatchTuner should look like:
Yan Ni's avatar
Yan Ni committed
195
196
197
198
199
200
201
202
203

```json
{
    "combine_params":
    {
        "_type" : "choice",
        "_value" : [{"optimizer": "Adam", "learning_rate": 0.00001},
                    {"optimizer": "Adam", "learning_rate": 0.0001},
                    {"optimizer": "Adam", "learning_rate": 0.001},
204
205
                    {"optimizer": "SGD", "learning_rate": 0.01},
                    {"optimizer": "SGD", "learning_rate": 0.005},
Yan Ni's avatar
Yan Ni committed
206
207
208
209
210
                    {"optimizer": "SGD", "learning_rate": 0.0002}]
    }
}
```

211
The search space file should include the high-level key `combine_params`. The type of params in the search space must be `choice` and the `values` must include all the combined params values.
Yan Ni's avatar
Yan Ni committed
212
213
214

<a name="GridSearch"></a>

215
### Grid Search
Yan Ni's avatar
Yan Ni committed
216

Chi Song's avatar
Chi Song committed
217
> Built-in Tuner Name: **Grid Search**
Yan Ni's avatar
Yan Ni committed
218
219
220

**Suggested scenario**

221
Note that the only acceptable types within the search space are `choice`, `quniform`, and `randint`.
Yan Ni's avatar
Yan Ni committed
222

223
This is suggested when the search space is small. It's suggested when it is feasible to exhaustively sweep the whole search space. [Detailed Description](./GridsearchTuner.md)
Yan Ni's avatar
Yan Ni committed
224

225
**Example Configuration:**
Yan Ni's avatar
Yan Ni committed
226

Yan Ni's avatar
Yan Ni committed
227
```yaml
Yan Ni's avatar
Yan Ni committed
228
229
230
231
232
233
234
235
236
# config.yml
tuner:
  builtinTunerName: GridSearch
```

<br>

<a name="Hyperband"></a>

237
### Hyperband
Yan Ni's avatar
Yan Ni committed
238

Chi Song's avatar
Chi Song committed
239
> Built-in Advisor Name: **Hyperband**
Yan Ni's avatar
Yan Ni committed
240
241
242

**Suggested scenario**

243
This is suggested when you have limited computational resources but have a relatively large search space. It performs well in scenarios where intermediate results can indicate good or bad final results to some extent. For example, when models that are more accurate early on in training are also more accurate later on. [Detailed Description](./HyperbandAdvisor.md)
Yan Ni's avatar
Yan Ni committed
244

245
**classArgs Requirements:**
Yan Ni's avatar
Yan Ni committed
246

247
248
249
* **optimize_mode** (*maximize or minimize, optional, default = maximize*) - If 'maximize', the tuner will try to maximize metrics. If 'minimize', the tuner will try to minimize metrics.
* **R** (*int, optional, default = 60*) - the maximum budget given to a trial (could be the number of mini-batches or epochs). Each trial should use TRIAL_BUDGET to control how long they run.
* **eta** (*int, optional, default = 3*) - `(eta-1)/eta` is the proportion of discarded trials.
J-shang's avatar
J-shang committed
250
* **exec_mode** (*serial or parallelism, optional, default = parallelism*) - If 'parallelism', the tuner will try to use available resources to start new bucket immediately. If 'serial', the tuner will only start new bucket after the current bucket is done.
Yan Ni's avatar
Yan Ni committed
251

252
**Example Configuration:**
Yan Ni's avatar
Yan Ni committed
253

Yan Ni's avatar
Yan Ni committed
254
```yaml
Yan Ni's avatar
Yan Ni committed
255
256
257
258
259
260
261
262
263
264
265
266
267
# config.yml
advisor:
  builtinAdvisorName: Hyperband
  classArgs:
    optimize_mode: maximize
    R: 60
    eta: 3
```

<br>

<a name="NetworkMorphism"></a>

268
### Network Morphism
Yan Ni's avatar
Yan Ni committed
269

Chi Song's avatar
Chi Song committed
270
> Built-in Tuner Name: **NetworkMorphism**
Yan Ni's avatar
Yan Ni committed
271
272
273

**Installation**

274
NetworkMorphism requires [PyTorch](https://pytorch.org/get-started/locally) and [Keras](https://keras.io/#installation), so users should install them first. The corresponding requirements file is [here](https://github.com/microsoft/nni/blob/v1.9/examples/trials/network_morphism/requirements.txt).
Yan Ni's avatar
Yan Ni committed
275
276
277

**Suggested scenario**

278
This is suggested when you want to apply deep learning methods to your task but you have no idea how to choose or design a network. You may modify this [example](https://github.com/Microsoft/nni/tree/v1.9/examples/trials/network_morphism/cifar10/cifar10_keras.py) to fit your own dataset and your own data augmentation method. Also you can change the batch size, learning rate, or optimizer. Currently, this tuner only supports the computer vision domain. [Detailed Description](./NetworkmorphismTuner.md)
Yan Ni's avatar
Yan Ni committed
279

280
**classArgs Requirements:**
Yan Ni's avatar
Yan Ni committed
281

282
283
* **optimize_mode** (*maximize or minimize, optional, default = maximize*) - If 'maximize', the tuner will try to maximize metrics. If 'minimize', the tuner will try to minimize metrics.
* **task** (*('cv'), optional, default = 'cv'*) - The domain of the experiment. For now, this tuner only supports the computer vision (CV) domain.
Yan Ni's avatar
Yan Ni committed
284
285
286
287
* **input_width** (*int, optional, default = 32*) - input image width
* **input_channel** (*int, optional, default = 3*) - input image channel
* **n_output_node** (*int, optional, default = 10*) - number of classes

288
**Example Configuration:**
Yan Ni's avatar
Yan Ni committed
289

Yan Ni's avatar
Yan Ni committed
290
```yaml
Yan Ni's avatar
Yan Ni committed
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
# config.yml
tuner:
  builtinTunerName: NetworkMorphism
    classArgs:
      optimize_mode: maximize
      task: cv
      input_width: 32
      input_channel: 3
      n_output_node: 10
```

<br>

<a name="MetisTuner"></a>

306
### Metis Tuner
Yan Ni's avatar
Yan Ni committed
307

Chi Song's avatar
Chi Song committed
308
> Built-in Tuner Name: **MetisTuner**
Yan Ni's avatar
Yan Ni committed
309

310
Note that the only acceptable types of search space types are `quniform`, `uniform`, `randint`, and numerical `choice`. Only numerical values are supported since the values will be used to evaluate the 'distance' between different points.
Chi Song's avatar
Chi Song committed
311

Yan Ni's avatar
Yan Ni committed
312
313
**Suggested scenario**

314
Similar to TPE and SMAC, Metis is a black-box tuner. If your system takes a long time to finish each trial, Metis is more favorable than other approaches such as random search. Furthermore, Metis provides guidance on subsequent trials. Here is an [example](https://github.com/Microsoft/nni/tree/v1.9/examples/trials/auto-gbdt/search_space_metis.json) on the use of Metis. Users only need to send the final result, such as `accuracy`, to the tuner by calling the NNI SDK. [Detailed Description](./MetisTuner.md)
Yan Ni's avatar
Yan Ni committed
315

316
**classArgs Requirements:**
Yan Ni's avatar
Yan Ni committed
317

318
* **optimize_mode** (*'maximize' or 'minimize', optional, default = 'maximize'*) - If 'maximize', the tuner will try to maximize metrics. If 'minimize', the tuner will try to minimize metrics.
Yan Ni's avatar
Yan Ni committed
319

320
**Example Configuration:**
Yan Ni's avatar
Yan Ni committed
321

Yan Ni's avatar
Yan Ni committed
322
```yaml
Yan Ni's avatar
Yan Ni committed
323
324
325
326
327
328
# config.yml
tuner:
  builtinTunerName: MetisTuner
  classArgs:
    optimize_mode: maximize
```
Shufan Huang's avatar
Shufan Huang committed
329
330
331
332
333

<br>

<a name="BOHB"></a>

334
### BOHB Advisor
Shufan Huang's avatar
Shufan Huang committed
335

Chi Song's avatar
Chi Song committed
336
> Built-in Tuner Name: **BOHB**
Shufan Huang's avatar
Shufan Huang committed
337
338
339

**Installation**

340
BOHB advisor requires [ConfigSpace](https://github.com/automl/ConfigSpace) package. ConfigSpace can be installed using the following command.
Shufan Huang's avatar
Shufan Huang committed
341
342
343
344
345
346
347

```bash
nnictl package install --name=BOHB
```

**Suggested scenario**

348
Similar to Hyperband, BOHB is suggested when you have limited computational resources but have a relatively large search space. It performs well in scenarios where intermediate results can indicate good or bad final results to some extent. In this case, it may converge to a better configuration than Hyperband due to its usage of Bayesian optimization. [Detailed Description](./BohbAdvisor.md)
Shufan Huang's avatar
Shufan Huang committed
349

350
**classArgs Requirements:**
Shufan Huang's avatar
Shufan Huang committed
351

352
353
354
* **optimize_mode** (*maximize or minimize, optional, default = maximize*) - If 'maximize', tuners will try to maximize metrics. If 'minimize', tuner will try to minimize metrics.
* **min_budget** (*int, optional, default = 1*) - The smallest budget to assign to a trial job, (budget can be the number of mini-batches or epochs). Needs to be positive.
* **max_budget** (*int, optional, default = 3*) - The largest budget to assign to a trial job, (budget can be the number of mini-batches or epochs). Needs to be larger than min_budget.
Shufan Huang's avatar
Shufan Huang committed
355
* **eta** (*int, optional, default = 3*) - In each iteration, a complete run of sequential halving is executed. In it, after evaluating each configuration on the same subset size, only a fraction of 1/eta of them 'advances' to the next round. Must be greater or equal to 2.
356
357
358
* **min_points_in_model**(*int, optional, default = None*): number of observations to start building a KDE. Default 'None' means dim+1; when the number of completed trials in this budget is equal to or larger than `max{dim+1, min_points_in_model}`, BOHB will start to build a KDE model of this budget then use said KDE model to guide configuration selection. Needs to be positive. (dim means the number of hyperparameters in search space)
* **top_n_percent**(*int, optional, default = 15*): percentage (between 1 and 99) of the observations which are considered good. Good points and bad points are used for building KDE models. For example, if you have 100 observed trials and top_n_percent is 15, then the top 15% of points will be used for building the good points models "l(x)". The remaining 85% of points will be used for building the bad point models "g(x)".
* **num_samples**(*int, optional, default = 64*): number of samples to optimize EI (default 64). In this case, we will sample "num_samples" points and compare the result of l(x)/g(x). Then we will return the one with the maximum l(x)/g(x) value as the next configuration if the optimize_mode is `maximize`. Otherwise, we return the smallest one.
Shufan Huang's avatar
Shufan Huang committed
359
* **random_fraction**(*float, optional, default = 0.33*): fraction of purely random configurations that are sampled from the prior without the model.
360
361
* **bandwidth_factor**(*float, optional, default = 3.0*): to encourage diversity, the points proposed to optimize EI are sampled from a 'widened' KDE where the bandwidth is multiplied by this factor. We suggest using the default value if you are not familiar with KDE.
* **min_bandwidth**(*float, optional, default = 0.001*): to keep diversity, even when all (good) samples have the same value for one of the parameters, a minimum bandwidth (default: 1e-3) is used instead of zero. We suggest using the default value if you are not familiar with KDE.
Shufan Huang's avatar
Shufan Huang committed
362

363
*Please note that the float type currently only supports decimal representations. You have to use 0.333 instead of 1/3 and 0.001 instead of 1e-3.*
Shufan Huang's avatar
Shufan Huang committed
364

365
**Example Configuration:**
Shufan Huang's avatar
Shufan Huang committed
366

367
```yaml
Shufan Huang's avatar
Shufan Huang committed
368
369
370
371
372
373
374
advisor:
  builtinAdvisorName: BOHB
  classArgs:
    optimize_mode: maximize
    min_budget: 1
    max_budget: 27
    eta: 3
375
```
Guoxin's avatar
Guoxin committed
376
377
378

<a name="GPTuner"></a>

379
### GP Tuner
Guoxin's avatar
Guoxin committed
380

Chi Song's avatar
Chi Song committed
381
> Built-in Tuner Name: **GPTuner**
Guoxin's avatar
Guoxin committed
382

383
Note that the only acceptable types within the search space are `randint`, `uniform`, `quniform`,  `loguniform`, `qloguniform`, and numerical `choice`. Only numerical values are supported since the values will be used to evaluate the 'distance' between different points.
Guoxin's avatar
Guoxin committed
384
385
386

**Suggested scenario**

387
As a strategy in a Sequential Model-based Global Optimization (SMBO) algorithm, GP Tuner uses a proxy optimization problem (finding the maximum of the acquisition function) that, albeit still a hard problem, is cheaper (in the computational sense) to solve and common tools can be employed to solve it. Therefore, GP Tuner is most adequate for situations where the function to be optimized is very expensive to evaluate. GP can be used when computational resources are limited. However, GP Tuner has a computational cost that grows at *O(N^3)* due to the requirement of inverting the Gram matrix, so it's not suitable when lots of trials are needed. [Detailed Description](./GPTuner.md)
Guoxin's avatar
Guoxin committed
388

389
**classArgs Requirements:**
Guoxin's avatar
Guoxin committed
390

391
* **optimize_mode** (*'maximize' or 'minimize', optional, default = 'maximize'*) - If 'maximize', the tuner will try to maximize metrics. If 'minimize', the tuner will try to minimize metrics.
392
* **utility** (*'ei', 'ucb' or 'poi', optional, default = 'ei'*) - The utility function (acquisition function). 'ei', 'ucb', and 'poi' correspond to 'Expected Improvement', 'Upper Confidence Bound', and 'Probability of Improvement', respectively.
393
394
395
396
397
398
* **kappa** (*float, optional, default = 5*) - Used by the 'ucb' utility function. The bigger `kappa` is, the more exploratory the tuner will be.
* **xi** (*float, optional, default = 0*) - Used by the 'ei' and 'poi' utility functions. The bigger `xi` is, the more exploratory the tuner will be.
* **nu** (*float, optional, default = 2.5*) - Used to specify the Matern kernel. The smaller nu, the less smooth the approximated function is.
* **alpha** (*float, optional, default = 1e-6*) - Used to specify the Gaussian Process Regressor. Larger values correspond to an increased noise level in the observations.
* **cold_start_num** (*int, optional, default = 10*) - Number of random explorations to perform before the Gaussian Process. Random exploration can help by diversifying the exploration space.
* **selection_num_warm_up** (*int, optional, default = 1e5*) - Number of random points to evaluate when getting the point which maximizes the acquisition function.
Chi Song's avatar
Chi Song committed
399
* **selection_num_starting_points** (*int, optional, default = 250*) - Number of times to run L-BFGS-B from a random starting point after the warmup.
Guoxin's avatar
Guoxin committed
400

401
**Example Configuration:**
Guoxin's avatar
Guoxin committed
402
403
404
405
406
407
408

```yaml
# config.yml
tuner:
  builtinTunerName: GPTuner
  classArgs:
    optimize_mode: maximize
Guoxin's avatar
Guoxin committed
409
410
411
    utility: 'ei'
    kappa: 5.0
    xi: 0.0
Guoxin's avatar
Guoxin committed
412
413
414
415
416
417
    nu: 2.5
    alpha: 1e-6
    cold_start_num: 10
    selection_num_warm_up: 100000
    selection_num_starting_points: 250
```
418
419
420

<a name="PPOTuner"></a>

421
### PPO Tuner
422
423
424

> Built-in Tuner Name: **PPOTuner**

425
Note that the only acceptable types within the search space are `layer_choice` and `input_choice`. For `input_choice`, `n_chosen` can only be 0, 1, or [0, 1]. Note, the search space file for NAS is usually automatically generated through the command [`nnictl ss_gen`](../Tutorial/Nnictl.md).
426
427
428

**Suggested scenario**

429
PPOTuner is a Reinforcement Learning tuner based on the PPO algorithm. PPOTuner can be used when using the NNI NAS interface to do neural architecture search. In general, the Reinforcement Learning algorithm needs more computing resources, though the PPO algorithm is relatively more efficient than others. It's recommended to use this tuner when you have a large amount of computional resources available. You could try it on a very simple task, such as the [mnist-nas](https://github.com/microsoft/nni/tree/v1.9/examples/trials/mnist-nas) example. [See details](./PPOTuner.md)
430

431
**classArgs Requirements:**
432

433
* **optimize_mode** (*'maximize' or 'minimize'*) - If 'maximize', the tuner will try to maximize metrics. If 'minimize', the tuner will try to minimize metrics.
Chi Song's avatar
Chi Song committed
434
* **trials_per_update** (*int, optional, default = 20*) - The number of trials to be used for one update. It must be divisible by minibatch_size. `trials_per_update` is recommended to be an exact multiple of `trialConcurrency` for better concurrency of trials.
435
* **epochs_per_update** (*int, optional, default = 4*) - The number of epochs for one update.
436
* **minibatch_size** (*int, optional, default = 4*) - Mini-batch size (i.e., number of trials for a mini-batch) for the update. Note that trials_per_update must be divisible by minibatch_size.
437
* **ent_coef** (*float, optional, default = 0.0*) - Policy entropy coefficient in the optimization objective.
438
* **lr** (*float, optional, default = 3e-4*) - Learning rate of the model (lstm network); constant.
439
440
441
442
443
444
* **vf_coef** (*float, optional, default = 0.5*) - Value function loss coefficient in the optimization objective.
* **max_grad_norm** (*float, optional, default = 0.5*) - Gradient norm clipping coefficient.
* **gamma** (*float, optional, default = 0.99*) - Discounting factor.
* **lam** (*float, optional, default = 0.95*) - Advantage estimation discounting factor (lambda in the paper).
* **cliprange** (*float, optional, default = 0.2*) - Cliprange in the PPO algorithm, constant.

445
**Example Configuration:**
446
447
448
449
450
451
452

```yaml
# config.yml
tuner:
  builtinTunerName: PPOTuner
  classArgs:
    optimize_mode: maximize
Chi Song's avatar
Chi Song committed
453
```
RayMeng8's avatar
RayMeng8 committed
454
455
456

<a name="PBTTuner"></a>

457
### PBT Tuner
RayMeng8's avatar
RayMeng8 committed
458
459
460
461
462

> Built-in Tuner Name: **PBTTuner**

**Suggested scenario**

463
Population Based Training (PBT) bridges and extends parallel search methods and sequential optimization methods. It requires relatively small computation resource, by inheriting weights from currently good-performing ones to explore better ones periodically. With PBTTuner, users finally get a trained model, rather than a configuration that could reproduce the trained model by training the model from scratch. This is because model weights are inherited periodically through the whole search process. PBT can also be seen as a training approach. If you don't need to get a specific configuration, but just expect a good model, PBTTuner is a good choice. [See details](./PBTTuner.md)
RayMeng8's avatar
RayMeng8 committed
464
465
466
467
468

**classArgs requirements:**

* **optimize_mode** (*'maximize' or 'minimize'*) - If 'maximize', the tuner will target to maximize metrics. If 'minimize', the tuner will target to minimize metrics.
* **all_checkpoint_dir** (*str, optional, default = None*) - Directory for trials to load and save checkpoint, if not specified, the directory would be "~/nni/checkpoint/<exp-id>". Note that if the experiment is not local mode, users should provide a path in a shared storage which can be accessed by all the trials.
469
* **population_size** (*int, optional, default = 10*) - Number of trials in a population. Each step has this number of trials. In our implementation, one step is running each trial by specific training epochs set by users.
RayMeng8's avatar
RayMeng8 committed
470
471
472
473
474
475
476
477
478
479
480
481
* **factors** (*tuple, optional, default = (1.2, 0.8)*) - Factors for perturbation of hyperparameters.
* **fraction** (*float, optional, default = 0.2*) - Fraction for selecting bottom and top trials.

**Usage example**

```yaml
# config.yml
tuner:
  builtinTunerName: PBTTuner
  classArgs:
    optimize_mode: maximize
```
482
483
484
485

Note that, to use this tuner, your trial code should be modified accordingly, please refer to [the document of PBTTuner](./PBTTuner.md) for details.


486
487
488
## **Reference and Feedback**
* To [report a bug](https://github.com/microsoft/nni/issues/new?template=bug-report.md) for this feature in GitHub;
* To [file a feature or improvement request](https://github.com/microsoft/nni/issues/new?template=enhancement.md) for this feature in GitHub;
489
490
491
* To know more about [Feature Engineering with NNI](https://github.com/microsoft/nni/blob/v1.9/docs/en_US/FeatureEngineering/Overview.md);
* To know more about [NAS with NNI](https://github.com/microsoft/nni/blob/v1.9/docs/en_US/NAS/Overview.md);
* To know more about [Model Compression with NNI](https://github.com/microsoft/nni/blob/v1.9/docs/en_US/Compression/Overview.md);