認知アーキテクチャを統合したbertによる感情分析について

pip install transformers datasets torch

import pandas as pd
from sklearn.metrics import accuracy_score, precision_recall_fscore_support
from transformers import BertTokenizer, BertForSequenceClassification, Trainer, TrainingArguments
import torch
from torch.utils.data import Dataset

# WRIMEデータセットのダウンロード
!wget https://github.com/ids-cv/wrime/raw/master/wrime-ver1.tsv

# データセットの読み込み
df = pd.read_table('wrime-ver1.tsv')
print(df.head())

# データフレームの列名を確認
print(df.columns)

# Googleドライブのマウント
from google.colab import drive
drive.mount('/content/drive')

# Hugging Face Hubにログイン
!huggingface-cli login

# トークンを入力し、Gitクレデンシャルとして追加するプロンプトに従う

# トークナイザーの読み込み
tokenizer = BertTokenizer.from_pretrained('cl-tohoku/bert-base-japanese')

# データセットを訓練用とテスト用に分割
df_groups = df.groupby('Train/Dev/Test')
df_train = df_groups.get_group('train')
df_test = pd.concat([df_groups.get_group('dev'), df_groups.get_group('test')])
print('train :', len(df_train))  # train : 17104
print('test :', len(df_test))    # test : 1133

# 訓練データとテストデータのトークン化
train_encodings = tokenizer(df_train['Sentence'].tolist(), padding='max_length', truncation=True, return_tensors='pt')
test_encodings = tokenizer(df_test['Sentence'].tolist(), padding='max_length', truncation=True, return_tensors='pt')

# データセットクラスの定義
class WRIMEDataset(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)

# 訓練データセットとテストデータセットの作成
train_labels = df_train['Avg. Readers_Joy'].tolist()
test_labels = df_test['Avg. Readers_Joy'].tolist()
train_dataset = WRIMEDataset(train_encodings, train_labels)
test_dataset = WRIMEDataset(test_encodings, test_labels)

# モデルの準備
model = BertForSequenceClassification.from_pretrained('cl-tohoku/bert-base-japanese', num_labels=8)

training_args = TrainingArguments(
    output_dir='./results',
    evaluation_strategy='epoch',
    learning_rate=2e-5,
    per_device_train_batch_size=16,
    per_device_eval_batch_size=16,
    num_train_epochs=3,
    weight_decay=0.01,
)

trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=train_dataset,
    eval_dataset=test_dataset
)

# モデルの訓練
trainer.train()

# モデルの評価
predictions = trainer.predict(test_dataset)
preds = predictions.predictions.argmax(-1)

# 精度、適合率、再現率、F1スコアの計算
accuracy = accuracy_score(test_labels, preds)
precision, recall, f1, _ = precision_recall_fscore_support(test_labels, preds, average='weighted')

print(f"Accuracy: {accuracy}")
print(f"Precision: {precision}")
print(f"Recall: {recall}")
print(f"F1 Score: {f1}")

# 感情分析結果を取得
predictions = trainer.predict(test_dataset)
predicted_labels = predictions.predictions.argmax(-1)

# ACT-Rモデルにフィードバック
class ACTRModel:
    def __init__(self):
        self.memory = []

    def update_memory(self, emotion):
        self.memory.append(emotion)

    def predict(self, input_data):
        # ここでACT-Rモデルを用いた予測ロジックを実装
        # 現在は単にメモリから最新の感情を返すだけ
        if self.memory:
            return self.memory[-1]
        else:
            return 0  # デフォルト値

actr_model = ACTRModel()

# 予測ラベルをACT-Rモデルにフィードバック
for label in predicted_labels:
    actr_model.update_memory(label)

# ACT-Rモデルを用いた新しい予測を生成
actr_predictions = [actr_model.predict(label) for label in predicted_labels]

# 評価指標の計算
from sklearn.metrics import accuracy_score, f1_score, precision_score, recall_score

# 実際の感情ラベル
true_labels = test_labels

# 精度の計算
accuracy = accuracy_score(true_labels, actr_predictions)
f1 = f1_score(true_labels, actr_predictions, average='weighted')
precision = precision_score(true_labels, actr_predictions, average='weighted')
recall = recall_score(true_labels, actr_predictions, average='weighted')

print(f"Accuracy: {accuracy}")
print(f"F1 Score: {f1}")
print(f"Precision: {precision}")
print(f"Recall: {recall}")