在 TensorFlow.org 上檢視

在 Google Colab 中執行

在 GitHub 上檢視

下載筆記本

查看 TF Hub 模型

TF-Hub 是一個平台，用於分享封裝在可重複使用資源中的機器學習專業知識，特別是預先訓練的模組。在本教學課程中，我們將使用 TF-Hub 文字嵌入模組來訓練一個簡單的情感分類器，並具有合理的基本準確度。然後，我們將預測結果提交到 Kaggle。

如需有關使用 TF-Hub 進行文字分類的更詳細教學課程，以及進一步提高準確度的步驟，請參閱使用 TF-Hub 進行文字分類。

設定

pip install -q kaggle

import tensorflow as tf
import tensorflow_hub as hub
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import seaborn as sns
import zipfile

from sklearn import model_selection

由於本教學課程將使用來自 Kaggle 的資料集，因此需要為您的 Kaggle 帳戶建立 API 金鑰，並將其上傳到 Colab 環境。

import os
import pathlib

# Upload the API token.
def get_kaggle():
  try:
    import kaggle
    return kaggle
  except OSError:
    pass

  token_file = pathlib.Path("~/.kaggle/kaggle.json").expanduser()
  token_file.parent.mkdir(exist_ok=True, parents=True)

  try:
    from google.colab import files
  except ImportError:
    raise ValueError("Could not find kaggle token.")

  uploaded = files.upload()
  token_content = uploaded.get('kaggle.json', None)
  if token_content:
    token_file.write_bytes(token_content)
    token_file.chmod(0o600)
  else:
    raise ValueError('Need a file named "kaggle.json"')

  import kaggle
  return kaggle


kaggle = get_kaggle()

開始使用

資料

我們將嘗試解決 Kaggle 上的電影評論情感分析任務。該資料集包含爛番茄電影評論的語法子短語。任務是在 1 到 5 的量表上將短語標記為負面或正面。

您必須接受競賽規則，才能使用 API 下載資料。

SENTIMENT_LABELS = [
    "negative", "somewhat negative", "neutral", "somewhat positive", "positive"
]

# Add a column with readable values representing the sentiment.
def add_readable_labels_column(df, sentiment_value_column):
  df["SentimentLabel"] = df[sentiment_value_column].replace(
      range(5), SENTIMENT_LABELS)

# Download data from Kaggle and create a DataFrame.
def load_data_from_zip(path):
  with zipfile.ZipFile(path, "r") as zip_ref:
    name = zip_ref.namelist()[0]
    with zip_ref.open(name) as zf:
      return pd.read_csv(zf, sep="\t", index_col=0)


# The data does not come with a validation set so we'll create one from the
# training set.
def get_data(competition, train_file, test_file, validation_set_ratio=0.1):
  data_path = pathlib.Path("data")
  kaggle.api.competition_download_files(competition, data_path)
  competition_path = (data_path/competition)
  competition_path.mkdir(exist_ok=True, parents=True)
  competition_zip_path = competition_path.with_suffix(".zip")

  with zipfile.ZipFile(competition_zip_path, "r") as zip_ref:
    zip_ref.extractall(competition_path)

  train_df = load_data_from_zip(competition_path/train_file)
  test_df = load_data_from_zip(competition_path/test_file)

  # Add a human readable label.
  add_readable_labels_column(train_df, "Sentiment")

  # We split by sentence ids, because we don't want to have phrases belonging
  # to the same sentence in both training and validation set.
  train_indices, validation_indices = model_selection.train_test_split(
      np.unique(train_df["SentenceId"]),
      test_size=validation_set_ratio,
      random_state=0)

  validation_df = train_df[train_df["SentenceId"].isin(validation_indices)]
  train_df = train_df[train_df["SentenceId"].isin(train_indices)]
  print("Split the training data into %d training and %d validation examples." %
        (len(train_df), len(validation_df)))

  return train_df, validation_df, test_df


train_df, validation_df, test_df = get_data(
    "sentiment-analysis-on-movie-reviews",
    "train.tsv.zip", "test.tsv.zip")

Split the training data into 140315 training and 15745 validation examples.

train_df.head(20)

訓練模型

class MyModel(tf.keras.Model):
  def __init__(self, hub_url):
    super().__init__()
    self.hub_url = hub_url
    self.embed = hub.load(self.hub_url).signatures['default']
    self.sequential = tf.keras.Sequential([
      tf.keras.layers.Dense(500),
      tf.keras.layers.Dense(100),
      tf.keras.layers.Dense(5),
    ])

  def call(self, inputs):
    phrases = inputs['Phrase'][:,0]
    embedding = 5*self.embed(phrases)['default']
    return self.sequential(embedding)

  def get_config(self):
    return {"hub_url":self.hub_url}

model = MyModel("https://tfhub.dev/google/nnlm-en-dim128/1")
model.compile(
    loss = tf.losses.SparseCategoricalCrossentropy(from_logits=True),
    optimizer=tf.optimizers.Adam(), 
    metrics = [tf.keras.metrics.SparseCategoricalAccuracy(name="accuracy")])

2024-03-09 13:31:20.808764: E external/local_xla/xla/stream_executor/cuda/cuda_driver.cc:282] failed call to cuInit: CUDA_ERROR_NO_DEVICE: no CUDA-capable device is detected

history = model.fit(x=dict(train_df), y=train_df['Sentiment'],
          validation_data=(dict(validation_df), validation_df['Sentiment']),
          epochs = 25)

Epoch 1/25
4385/4385 ━━━━━━━━━━━━━━━━━━━━ 17s 4ms/step - accuracy: 0.5711 - loss: 1.0662 - val_accuracy: 0.5910 - val_loss: 1.0030
Epoch 2/25
4385/4385 ━━━━━━━━━━━━━━━━━━━━ 16s 4ms/step - accuracy: 0.5939 - loss: 1.0012 - val_accuracy: 0.5965 - val_loss: 0.9947
Epoch 3/25
4385/4385 ━━━━━━━━━━━━━━━━━━━━ 16s 4ms/step - accuracy: 0.5967 - loss: 0.9949 - val_accuracy: 0.5940 - val_loss: 0.9910
Epoch 4/25
4385/4385 ━━━━━━━━━━━━━━━━━━━━ 16s 4ms/step - accuracy: 0.5978 - loss: 0.9944 - val_accuracy: 0.5910 - val_loss: 0.9921
Epoch 5/25
4385/4385 ━━━━━━━━━━━━━━━━━━━━ 16s 4ms/step - accuracy: 0.5959 - loss: 0.9923 - val_accuracy: 0.5906 - val_loss: 0.9898
Epoch 6/25
4385/4385 ━━━━━━━━━━━━━━━━━━━━ 16s 4ms/step - accuracy: 0.5984 - loss: 0.9890 - val_accuracy: 0.5950 - val_loss: 0.9861
Epoch 7/25
4385/4385 ━━━━━━━━━━━━━━━━━━━━ 16s 4ms/step - accuracy: 0.5967 - loss: 0.9887 - val_accuracy: 0.5883 - val_loss: 0.9970
Epoch 8/25
4385/4385 ━━━━━━━━━━━━━━━━━━━━ 16s 4ms/step - accuracy: 0.5968 - loss: 0.9893 - val_accuracy: 0.6026 - val_loss: 0.9858
Epoch 9/25
4385/4385 ━━━━━━━━━━━━━━━━━━━━ 16s 4ms/step - accuracy: 0.5982 - loss: 0.9883 - val_accuracy: 0.5940 - val_loss: 0.9888
Epoch 10/25
4385/4385 ━━━━━━━━━━━━━━━━━━━━ 16s 4ms/step - accuracy: 0.5986 - loss: 0.9910 - val_accuracy: 0.6008 - val_loss: 0.9815
Epoch 11/25
4385/4385 ━━━━━━━━━━━━━━━━━━━━ 16s 4ms/step - accuracy: 0.5988 - loss: 0.9869 - val_accuracy: 0.5960 - val_loss: 0.9871
Epoch 12/25
4385/4385 ━━━━━━━━━━━━━━━━━━━━ 16s 4ms/step - accuracy: 0.6008 - loss: 0.9863 - val_accuracy: 0.5942 - val_loss: 0.9863
Epoch 13/25
4385/4385 ━━━━━━━━━━━━━━━━━━━━ 16s 4ms/step - accuracy: 0.5996 - loss: 0.9878 - val_accuracy: 0.5990 - val_loss: 0.9886
Epoch 14/25
4385/4385 ━━━━━━━━━━━━━━━━━━━━ 16s 4ms/step - accuracy: 0.6022 - loss: 0.9852 - val_accuracy: 0.5969 - val_loss: 0.9831
Epoch 15/25
4385/4385 ━━━━━━━━━━━━━━━━━━━━ 16s 4ms/step - accuracy: 0.5960 - loss: 0.9910 - val_accuracy: 0.5971 - val_loss: 0.9851
Epoch 16/25
4385/4385 ━━━━━━━━━━━━━━━━━━━━ 16s 4ms/step - accuracy: 0.5991 - loss: 0.9887 - val_accuracy: 0.6022 - val_loss: 0.9840
Epoch 17/25
4385/4385 ━━━━━━━━━━━━━━━━━━━━ 16s 4ms/step - accuracy: 0.6007 - loss: 0.9856 - val_accuracy: 0.5987 - val_loss: 0.9832
Epoch 18/25
4385/4385 ━━━━━━━━━━━━━━━━━━━━ 16s 4ms/step - accuracy: 0.6010 - loss: 0.9873 - val_accuracy: 0.5994 - val_loss: 0.9848
Epoch 19/25
4385/4385 ━━━━━━━━━━━━━━━━━━━━ 16s 4ms/step - accuracy: 0.5986 - loss: 0.9877 - val_accuracy: 0.5969 - val_loss: 0.9938
Epoch 20/25
4385/4385 ━━━━━━━━━━━━━━━━━━━━ 16s 4ms/step - accuracy: 0.6006 - loss: 0.9857 - val_accuracy: 0.5914 - val_loss: 0.9933
Epoch 21/25
4385/4385 ━━━━━━━━━━━━━━━━━━━━ 16s 4ms/step - accuracy: 0.5986 - loss: 0.9870 - val_accuracy: 0.5956 - val_loss: 0.9882
Epoch 22/25
4385/4385 ━━━━━━━━━━━━━━━━━━━━ 16s 4ms/step - accuracy: 0.5997 - loss: 0.9866 - val_accuracy: 0.5966 - val_loss: 0.9882
Epoch 23/25
4385/4385 ━━━━━━━━━━━━━━━━━━━━ 16s 4ms/step - accuracy: 0.5989 - loss: 0.9837 - val_accuracy: 0.5910 - val_loss: 0.9869
Epoch 24/25
4385/4385 ━━━━━━━━━━━━━━━━━━━━ 16s 4ms/step - accuracy: 0.5981 - loss: 0.9872 - val_accuracy: 0.6011 - val_loss: 0.9812
Epoch 25/25
4385/4385 ━━━━━━━━━━━━━━━━━━━━ 16s 4ms/step - accuracy: 0.6018 - loss: 0.9834 - val_accuracy: 0.5976 - val_loss: 0.9854

預測

針對驗證集和訓練集執行預測。

plt.plot(history.history['accuracy'])
plt.plot(history.history['val_accuracy'])

[<matplotlib.lines.Line2D at 0x7fcfc99b88e0>]

train_eval_result = model.evaluate(dict(train_df), train_df['Sentiment'])
validation_eval_result = model.evaluate(dict(validation_df), validation_df['Sentiment'])

print(f"Training set accuracy: {train_eval_result[1]}")
print(f"Validation set accuracy: {validation_eval_result[1]}")

4385/4385 ━━━━━━━━━━━━━━━━━━━━ 16s 4ms/step - accuracy: 0.6212 - loss: 0.9463
493/493 ━━━━━━━━━━━━━━━━━━━━ 1s 1ms/step - accuracy: 0.6105 - loss: 0.9561
Training set accuracy: 0.6025798916816711
Validation set accuracy: 0.5975865125656128

混淆矩陣

另一個非常有趣的統計數據，特別是對於多類別問題，是混淆矩陣。混淆矩陣可以視覺化正確和不正確標記範例的比例。我們可以輕鬆地看出我們的分類器有多少偏差，以及標籤的分佈是否合理。理想情況下，最大部分的預測應該沿著對角線分佈。

predictions = model.predict(dict(validation_df))
predictions = tf.argmax(predictions, axis=-1)
predictions

493/493 ━━━━━━━━━━━━━━━━━━━━ 1s 1ms/step
<tf.Tensor: shape=(15745,), dtype=int64, numpy=array([1, 1, 2, ..., 2, 2, 2])>

cm = tf.math.confusion_matrix(validation_df['Sentiment'], predictions)
cm = cm/cm.numpy().sum(axis=1)[:, tf.newaxis]

sns.heatmap(
    cm, annot=True,
    xticklabels=SENTIMENT_LABELS,
    yticklabels=SENTIMENT_LABELS)
plt.xlabel("Predicted")
plt.ylabel("True")

Text(50.72222222222221, 0.5, 'True')

我們可以透過將以下程式碼貼到程式碼儲存格並執行它，輕鬆地將預測結果提交回 Kaggle

test_predictions = model.predict(dict(test_df))
test_predictions = np.argmax(test_predictions, axis=-1)

result_df = test_df.copy()

result_df["Predictions"] = test_predictions

result_df.to_csv(
    "predictions.csv",
    columns=["Predictions"],
    header=["Sentiment"])
kaggle.api.competition_submit("predictions.csv", "Submitted from Colab",
                              "sentiment-analysis-on-movie-reviews")

提交後，查看排行榜以了解您的表現。