Note
Go to the end to download the full example code.
Hyperparameter search for text classification#
Author(s): Romain Egele, Brett Eiffert.
In this tutorial we present how to use hyperparameter optimization on a text classification analysis example from the Pytorch documentation.
Reference: This tutorial is based on materials from the Pytorch Documentation: Text classification with the torchtext library
%%bash
pip install deephyper
pip install ray
pip install torch torchtext torchdata
Imports#
import ray
import json
import pandas as pd
from functools import partial
import torch
from torchtext.data.utils import get_tokenizer
from torchtext.data.functional import to_map_style_dataset
from torchtext.vocab import build_vocab_from_iterator
from torch.utils.data import DataLoader
from torch.utils.data.dataset import random_split
from torch import nn
Note
The following can be used to detect if <b>CUDA</b> devices are available on the current host. Therefore, this notebook will automatically adapt the parallel execution based on the ressources available locally. However, it will not be the case if many compute nodes are requested.
The dataset#
The torchtext library provides a few raw dataset iterators, which yield the raw text strings. For example, the AG_NEWS dataset iterators yield the raw data as a tuple of label and text. It has four labels (1 : World 2 : Sports 3 : Business 4 : Sci/Tec).
from torchtext.datasets import AG_NEWS
def load_data(train_ratio, fast=False):
train_iter, test_iter = AG_NEWS()
train_dataset = to_map_style_dataset(train_iter)
test_dataset = to_map_style_dataset(test_iter)
num_train = int(len(train_dataset) * train_ratio)
split_train, split_valid = \
random_split(train_dataset, [num_train, len(train_dataset) - num_train])
## downsample
if fast:
split_train, _ = random_split(split_train, [int(len(split_train)*.05), int(len(split_train)*.95)])
split_valid, _ = random_split(split_valid, [int(len(split_valid)*.05), int(len(split_valid)*.95)])
test_dataset, _ = random_split(test_dataset, [int(len(test_dataset)*.05), int(len(test_dataset)*.95)])
return split_train, split_valid, test_dataset
Preprocessing pipelines and Batch generation#
Here is an example for typical NLP data processing with tokenizer and vocabulary. The first step is to build a vocabulary with the raw training dataset. Here we use built in
factory function build_vocab_from_iterator which accepts iterator that yield list or iterator of tokens. Users can also pass any special symbols to be added to the
vocabulary.
The vocabulary block converts a list of tokens into integers.
vocab(['here', 'is', 'an', 'example'])
>>> [475, 21, 30, 5286]
The text pipeline converts a text string into a list of integers based on the lookup table defined in the vocabulary. The label pipeline converts the label into integers. For example,
text_pipeline('here is the an example')
>>> [475, 21, 2, 30, 5286]
label_pipeline('10')
>>> 9
train_iter = AG_NEWS(split='train')
num_class = 4
tokenizer = get_tokenizer('basic_english')
def yield_tokens(data_iter):
for _, text in data_iter:
yield tokenizer(text)
vocab = build_vocab_from_iterator(yield_tokens(train_iter), specials=["<unk>"])
vocab.set_default_index(vocab["<unk>"])
vocab_size = len(vocab)
text_pipeline = lambda x: vocab(tokenizer(x))
label_pipeline = lambda x: int(x) - 1
def collate_batch(batch, device):
label_list, text_list, offsets = [], [], [0]
for (_label, _text) in batch:
label_list.append(label_pipeline(_label))
processed_text = torch.tensor(text_pipeline(_text), dtype=torch.int64)
text_list.append(processed_text)
offsets.append(processed_text.size(0))
label_list = torch.tensor(label_list, dtype=torch.int64)
offsets = torch.tensor(offsets[:-1]).cumsum(dim=0)
text_list = torch.cat(text_list)
return label_list.to(device), text_list.to(device), offsets.to(device)
Note
The collate_fn function works on a batch of samples generated from DataLoader. The input to collate_fn is a batch of data with the batch size in DataLoader, and collate_fn processes them according to the data processing pipelines declared previously.
Define the model#
The model is composed of the nn.EmbeddingBag layer plus a linear layer for the classification purpose.
class TextClassificationModel(nn.Module):
def __init__(self, vocab_size, embed_dim, num_class):
super(TextClassificationModel, self).__init__()
self.embedding = nn.EmbeddingBag(vocab_size, embed_dim, sparse=False)
self.fc = nn.Linear(embed_dim, num_class)
self.init_weights()
def init_weights(self):
initrange = 0.5
self.embedding.weight.data.uniform_(-initrange, initrange)
self.fc.weight.data.uniform_(-initrange, initrange)
self.fc.bias.data.zero_()
def forward(self, text, offsets):
embedded = self.embedding(text, offsets)
return self.fc(embedded)
Define functions to train the model and evaluate results.#
def train(model, criterion, optimizer, dataloader):
model.train()
for _, (label, text, offsets) in enumerate(dataloader):
optimizer.zero_grad()
predicted_label = model(text, offsets)
loss = criterion(predicted_label, label)
loss.backward()
torch.nn.utils.clip_grad_norm_(model.parameters(), 0.1)
optimizer.step()
def evaluate(model, dataloader):
model.eval()
total_acc, total_count = 0, 0
with torch.no_grad():
for _, (label, text, offsets) in enumerate(dataloader):
predicted_label = model(text, offsets)
total_acc += (predicted_label.argmax(1) == label).sum().item()
total_count += label.size(0)
return total_acc/total_count
Define the run-function#
The run-function defines how the objective that we want to maximize is computed. It takes a config dictionary as input and often returns a scalar value that we want to maximize. The config contains a sample value of hyperparameters that we want to tune. In this example we will search for:
num_epochs(default value:10)batch_size(default value:64)learning_rate(default value:5)
A hyperparameter value can be acessed easily in the dictionary through the corresponding key, for example config["units"].
def get_run(train_ratio=0.95):
def run(config: dict):
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
embed_dim = 64
collate_fn = partial(collate_batch, device=device)
split_train, split_valid, _ = load_data(train_ratio, fast=True) # set fast=false for longer running, more accurate example
train_dataloader = DataLoader(split_train, batch_size=int(config["batch_size"]),
shuffle=True, collate_fn=collate_fn)
valid_dataloader = DataLoader(split_valid, batch_size=int(config["batch_size"]),
shuffle=True, collate_fn=collate_fn)
model = TextClassificationModel(vocab_size, int(embed_dim), num_class).to(device)
criterion = torch.nn.CrossEntropyLoss()
optimizer = torch.optim.SGD(model.parameters(), lr=config["learning_rate"])
for _ in range(1, int(config["num_epochs"]) + 1):
train(model, criterion, optimizer, train_dataloader)
accu_test = evaluate(model, valid_dataloader)
return accu_test
return run
We create two versions of run, one quicker to evaluate for the search, with a small training dataset, and another one, for performance evaluation, which uses a normal training/validation ratio.
quick_run = get_run(train_ratio=0.3)
perf_run = get_run(train_ratio=0.95)
Note
The objective maximised by DeepHyper is the scalar value returned by the run-function.
In this tutorial it corresponds to the validation accuracy of the model after training.
Define the Hyperparameter optimization problem#
Hyperparameter ranges are defined using the following syntax:
Discrete integer ranges are generated from a tuple
(lower: int, upper: int)Continuous prarameters are generated from a tuple
(lower: float, upper: float)Categorical or nonordinal hyperparameter ranges can be given as a list of possible values
[val1, val2, ...]
We provide the default configuration of hyperparameters as a starting point of the problem.
from deephyper.hpo import HpProblem
problem = HpProblem()
# Discrete hyperparameter (sampled with uniform prior)
problem.add_hyperparameter((5, 20), "num_epochs", default_value=10)
# Discrete and Real hyperparameters (sampled with log-uniform)
problem.add_hyperparameter((8, 512, "log-uniform"), "batch_size", default_value=64)
problem.add_hyperparameter((0.1, 10, "log-uniform"), "learning_rate", default_value=5)
problem
Configuration space object:
Hyperparameters:
batch_size, Type: UniformInteger, Range: [8, 512], Default: 64, on log-scale
learning_rate, Type: UniformFloat, Range: [0.1, 10.0], Default: 5.0, on log-scale
num_epochs, Type: UniformInteger, Range: [5, 20], Default: 10
Evaluate a default configuration#
We evaluate the performance of the default set of hyperparameters provided in the Pytorch tutorial.
#We launch the Ray run-time and execute the `run` function
#with the default configuration
if is_gpu_available:
if not(ray.is_initialized()):
ray.init(num_cpus=n_gpus, num_gpus=n_gpus, log_to_driver=False)
run_default = ray.remote(num_cpus=1, num_gpus=1)(perf_run)
objective_default = ray.get(run_default.remote(problem.default_configuration))
else:
if not(ray.is_initialized()):
ray.init(num_cpus=1, log_to_driver=False)
run_default = perf_run
objective_default = run_default(problem.default_configuration)
print(f"Accuracy Default Configuration: {objective_default:.3f}")
2025-03-18 00:44:17,179 INFO worker.py:1832 -- Started a local Ray instance. View the dashboard at http://127.0.0.1:8265
Accuracy Default Configuration: 0.867
Define the evaluator object#
The Evaluator object allows to change the parallelization backend used by DeepHyper.
It is a standalone object which schedules the execution of remote tasks. All evaluators needs a run_function to be instantiated.
Then a keyword method defines the backend (e.g., "ray") and the method_kwargs corresponds to keyword arguments of this chosen method.
evaluator = Evaluator.create(run_function, method, method_kwargs)
Once created the evaluator.num_workers gives access to the number of available parallel workers.
Finally, to submit and collect tasks to the evaluator one just needs to use the following interface:
configs = [...]
evaluator.submit(configs)
...
tasks_done = evaluator.get("BATCH", size=1) # For asynchronous
tasks_done = evaluator.get("ALL") # For batch synchronous
Warning
Each Evaluator saves its own state, therefore it is crucial to create a new evaluator when launching a fresh search.
from deephyper.evaluator import Evaluator
from deephyper.evaluator.callback import TqdmCallback
def get_evaluator(run_function):
# Default arguments for Ray: 1 worker and 1 worker per evaluation
method_kwargs = {
"num_cpus": 1,
"num_cpus_per_task": 1,
"callbacks": [TqdmCallback()]
}
# If GPU devices are detected then it will create 'n_gpus' workers
# and use 1 worker for each evaluation
if is_gpu_available:
method_kwargs["num_cpus"] = n_gpus
method_kwargs["num_gpus"] = n_gpus
method_kwargs["num_cpus_per_task"] = 1
method_kwargs["num_gpus_per_task"] = 1
evaluator = Evaluator.create(
run_function,
method="ray",
method_kwargs=method_kwargs
)
print(f"Created new evaluator with {evaluator.num_workers} worker{'s' if evaluator.num_workers > 1 else ''} and config: {method_kwargs}", )
return evaluator
evaluator_1 = get_evaluator(quick_run)
Created new evaluator with 1 worker and config: {'num_cpus': 1, 'num_cpus_per_task': 1, 'callbacks': [<deephyper.evaluator.callback.TqdmCallback object at 0x1339d1e20>]}
Define and run the Centralized Bayesian Optimization search (CBO)#
We create the CBO using the problem and evaluator defined above.
from deephyper.hpo import CBO
Instanciate the search with the problem and a specific evaluator
search = CBO(problem, evaluator_1)
WARNING:root:Results file already exists, it will be renamed to /Users/35e/Projects/DeepHyper/deephyper/examples/examples_hpo/results_20250318-004434.csv
Note
- All DeepHyper’s search algorithm have two stopping criteria:
max_evals (int): Defines the maximum number of evaluations that we want to perform. Default to-1for an infinite number.timeout (int): Defines a time budget (in seconds) before stopping the search. Default toNonefor an infinite time budget.
results = search.search(max_evals=30)
0%| | 0/30 [00:00<?, ?it/s]
3%|▎ | 1/30 [00:00<00:00, 1901.32it/s, failures=0, objective=0.357]
7%|▋ | 2/30 [00:04<01:08, 2.46s/it, failures=0, objective=0.357]
7%|▋ | 2/30 [00:04<01:08, 2.46s/it, failures=0, objective=0.357]
10%|█ | 3/30 [00:09<01:29, 3.30s/it, failures=0, objective=0.357]
10%|█ | 3/30 [00:09<01:29, 3.30s/it, failures=0, objective=0.357]
13%|█▎ | 4/30 [00:13<01:29, 3.45s/it, failures=0, objective=0.357]
13%|█▎ | 4/30 [00:13<01:29, 3.45s/it, failures=0, objective=0.576]
17%|█▋ | 5/30 [00:48<06:05, 14.63s/it, failures=0, objective=0.576]
17%|█▋ | 5/30 [00:48<06:05, 14.63s/it, failures=0, objective=0.586]
20%|██ | 6/30 [01:02<05:41, 14.24s/it, failures=0, objective=0.586]
20%|██ | 6/30 [01:02<05:41, 14.24s/it, failures=0, objective=0.586]
23%|██▎ | 7/30 [01:04<04:03, 10.57s/it, failures=0, objective=0.586]
23%|██▎ | 7/30 [01:04<04:03, 10.57s/it, failures=0, objective=0.586]
27%|██▋ | 8/30 [01:08<03:06, 8.48s/it, failures=0, objective=0.586]
27%|██▋ | 8/30 [01:08<03:06, 8.48s/it, failures=0, objective=0.586]
30%|███ | 9/30 [01:12<02:26, 6.97s/it, failures=0, objective=0.586]
30%|███ | 9/30 [01:12<02:26, 6.97s/it, failures=0, objective=0.586]
33%|███▎ | 10/30 [01:20<02:25, 7.28s/it, failures=0, objective=0.586]
33%|███▎ | 10/30 [01:20<02:25, 7.28s/it, failures=0, objective=0.776]
37%|███▋ | 11/30 [01:25<02:07, 6.71s/it, failures=0, objective=0.776]
37%|███▋ | 11/30 [01:25<02:07, 6.71s/it, failures=0, objective=0.776]
40%|████ | 12/30 [01:32<02:01, 6.78s/it, failures=0, objective=0.776]
40%|████ | 12/30 [01:32<02:01, 6.78s/it, failures=0, objective=0.776]
43%|████▎ | 13/30 [01:41<02:03, 7.27s/it, failures=0, objective=0.776]
43%|████▎ | 13/30 [01:41<02:03, 7.27s/it, failures=0, objective=0.776]
47%|████▋ | 14/30 [01:50<02:05, 7.83s/it, failures=0, objective=0.776]
47%|████▋ | 14/30 [01:50<02:05, 7.83s/it, failures=0, objective=0.794]
50%|█████ | 15/30 [01:56<01:51, 7.44s/it, failures=0, objective=0.794]
50%|█████ | 15/30 [01:56<01:51, 7.44s/it, failures=0, objective=0.797]
53%|█████▎ | 16/30 [02:03<01:38, 7.04s/it, failures=0, objective=0.797]
53%|█████▎ | 16/30 [02:03<01:38, 7.04s/it, failures=0, objective=0.805]
57%|█████▋ | 17/30 [02:08<01:24, 6.48s/it, failures=0, objective=0.805]
57%|█████▋ | 17/30 [02:08<01:24, 6.48s/it, failures=0, objective=0.805]
60%|██████ | 18/30 [02:10<01:03, 5.27s/it, failures=0, objective=0.805]
60%|██████ | 18/30 [02:10<01:03, 5.27s/it, failures=0, objective=0.805]
63%|██████▎ | 19/30 [02:16<00:59, 5.43s/it, failures=0, objective=0.805]
63%|██████▎ | 19/30 [02:16<00:59, 5.43s/it, failures=0, objective=0.805]
67%|██████▋ | 20/30 [02:23<00:58, 5.84s/it, failures=0, objective=0.805]
67%|██████▋ | 20/30 [02:23<00:58, 5.84s/it, failures=0, objective=0.805]
70%|███████ | 21/30 [02:28<00:50, 5.65s/it, failures=0, objective=0.805]
70%|███████ | 21/30 [02:28<00:50, 5.65s/it, failures=0, objective=0.805]
73%|███████▎ | 22/30 [02:33<00:44, 5.58s/it, failures=0, objective=0.805]
73%|███████▎ | 22/30 [02:33<00:44, 5.58s/it, failures=0, objective=0.805]
77%|███████▋ | 23/30 [02:39<00:40, 5.73s/it, failures=0, objective=0.805]
77%|███████▋ | 23/30 [02:39<00:40, 5.73s/it, failures=0, objective=0.805]
80%|████████ | 24/30 [03:01<01:03, 10.59s/it, failures=0, objective=0.805]
80%|████████ | 24/30 [03:01<01:03, 10.59s/it, failures=0, objective=0.814]
83%|████████▎ | 25/30 [03:27<01:15, 15.03s/it, failures=0, objective=0.814]
83%|████████▎ | 25/30 [03:27<01:15, 15.03s/it, failures=0, objective=0.814]
87%|████████▋ | 26/30 [03:50<01:09, 17.36s/it, failures=0, objective=0.814]
87%|████████▋ | 26/30 [03:50<01:09, 17.36s/it, failures=0, objective=0.814]
90%|█████████ | 27/30 [04:13<00:57, 19.06s/it, failures=0, objective=0.814]
90%|█████████ | 27/30 [04:13<00:57, 19.06s/it, failures=0, objective=0.814]
93%|█████████▎| 28/30 [04:31<00:37, 18.84s/it, failures=0, objective=0.814]
93%|█████████▎| 28/30 [04:31<00:37, 18.84s/it, failures=0, objective=0.814]
97%|█████████▋| 29/30 [05:12<00:25, 25.55s/it, failures=0, objective=0.814]
97%|█████████▋| 29/30 [05:12<00:25, 25.55s/it, failures=0, objective=0.814]
100%|██████████| 30/30 [05:32<00:00, 23.86s/it, failures=0, objective=0.814]
100%|██████████| 30/30 [05:32<00:00, 23.86s/it, failures=0, objective=0.814]
The returned results is a Pandas Dataframe where columns are hyperparameters and information stored by the evaluator:
job_idis a unique identifier corresponding to the order of creation of tasksobjectiveis the value returned by the run-functiontimestamp_submitis the time (in seconds) when the hyperparameter configuration was submitted by theEvaluatorrelative to the creation of the evaluator.timestamp_gatheris the time (in seconds) when the hyperparameter configuration was collected by theEvaluatorrelative to the creation of the evaluator.
Evaluate the best configuration#
Now that the search is over, let us print the best configuration found during this run and evaluate it on the full training dataset.
i_max = results.objective.argmax()
best_config = results.iloc[i_max][:-3].to_dict()
best_config = {k[2:]: v for k, v in best_config.items() if k.startswith("p:")}
print(f"The default configuration has an accuracy of {objective_default:.3f}. \n"
f"The best configuration found by DeepHyper has an accuracy {results['objective'].iloc[i_max]:.3f}, \n"
f"finished after {results['m:timestamp_gather'].iloc[i_max]:.2f} secondes of search.\n")
print(json.dumps(best_config, indent=4))
The default configuration has an accuracy of 0.867.
The best configuration found by DeepHyper has an accuracy 0.814,
finished after 192.00 secondes of search.
{
"batch_size": 22,
"learning_rate": 6.480428119634669,
"num_epochs": 19
}
objective_best = perf_run(best_config)
print(f"Accuracy Best Configuration: {objective_best:.3f}")
Accuracy Best Configuration: 0.903
Total running time of the script: (7 minutes 12.166 seconds)