Fine-tuning for HuggingFace Sentiment Classifier https://lppier.github.io/NLP-PK!-HuggingFace-vs-AWS-ML-Services

HF_CustomSentimentClassifier.py

from pathlib import Path
from sklearn.model_selection import train_test_split
from transformers import DistilBertTokenizerFast
import torch
from transformers import DistilBertForSequenceClassification, Trainer, TrainingArguments
import torch.nn.functional as F
from sklearn.metrics import accuracy_score, precision_recall_fscore_support

# IMDB Dataset can be found here
# wget http://ai.stanford.edu/~amaas/data/sentiment/aclImdb_v1.tar.gz
# tar -xf aclImdb_v1.tar.gz


def compute_metrics(pred):
    labels = pred.label_ids
    preds = pred.predictions.argmax(-1)
    precision, recall, f1, _ = precision_recall_fscore_support(
        labels, preds, average="binary"
    )
    acc = accuracy_score(labels, preds)
    return {"accuracy": acc, "f1": f1, "precision": precision, "recall": recall}


def read_imdb_split(split_dir):
    split_dir = Path(split_dir)
    texts = []
    labels = []
    for label_dir in ["pos", "neg"]:
        for text_file in (split_dir / label_dir).iterdir():
            texts.append(text_file.read_text(encoding="utf8"))
            labels.append(0 if label_dir is "neg" else 1)

    return texts, labels


train_texts, train_labels = read_imdb_split("data/aclImdb/train")
test_texts, test_labels = read_imdb_split("data/aclImdb/test")

# Further split training set to get a validation set
train_texts, val_texts, train_labels, val_labels = train_test_split(
    train_texts, train_labels, test_size=0.1
)

# Get BERT Tokens
tokenizer = DistilBertTokenizerFast.from_pretrained("distilbert-base-uncased")
train_encodings = tokenizer(train_texts, truncation=True, padding=True)
val_encodings = tokenizer(val_texts, truncation=True, padding=True)
test_encodings = tokenizer(test_texts, truncation=True, padding=True)


class IMDbDataset(torch.utils.data.Dataset):
    def __init__(self, encodings, labels):
        self.encodings = encodings
        self.labels = labels

    def __getitem__(self, idx):
        item = {key: torch.tensor(val[idx]) for key, val in self.encodings.items()}
        item["labels"] = torch.tensor(self.labels[idx])
        return item

    def __len__(self):
        return len(self.labels)


train_dataset = IMDbDataset(train_encodings, train_labels)
val_dataset = IMDbDataset(val_encodings, val_labels)
test_dataset = IMDbDataset(test_encodings, test_labels)

training_args = TrainingArguments(
    output_dir="./results",  # output directory
    num_train_epochs=1,  # total number of training epochs
    per_device_train_batch_size=16,  # batch size per device during training
    per_device_eval_batch_size=64,  # batch size for evaluation
    warmup_steps=500,  # number of warmup steps for learning rate scheduler
    weight_decay=0.01,  # strength of weight decay
    logging_dir="./logs",  # directory for storing logs
    logging_steps=10,
)

model = DistilBertForSequenceClassification.from_pretrained("distilbert-base-uncased")

trainer = Trainer(
    model=model,  # the instantiated 🤗 Transformers model to be trained
    args=training_args,  # training arguments, defined above
    train_dataset=train_dataset,  # training dataset
    eval_dataset=val_dataset,  # evaluation dataset
    compute_metrics=compute_metrics,
)

trainer.train()  # this saves the models in the checkpoints under results folder

# Inference Code

import pandas as pd
model = DistilBertForSequenceClassification.from_pretrained("./results_old/checkpoint-3500")
df_labels = pd.read_csv("data/comprehendimdbtest.csv", header=None)
test_data = df_labels.iloc[:,1].to_list()

predictions = []
# Need this loop otherwise my cpu memory just loads up
for text in test_data:
    encoding = tokenizer(text, return_tensors="pt", padding=True, truncation=True)
    outputs = model(**encoding)
    pt_predictions = F.softmax(outputs[0], dim=-1)
    predictions.append(pt_predictions.argmax(-1).item()) # warning, don't append the tensor too memory intensive!
    print(pt_predictions)


precision, recall, f1, _ = precision_recall_fscore_support(
    df_labels.iloc[:, 0], predictions, average="binary"
)
acc = accuracy_score(df_labels.iloc[:, 0], predictions)
metrics = {"accuracy": acc, "f1": f1, "precision": precision, "recall": recall}
print(metrics)