Halaman ini diterjemahkan oleh Cloud Translation API.
Switch to English

Bagaimana mengatasi masalah di Kaggle dengan TF-Hub

Lihat di TensorFlow.org Jalankan di Google Colab Lihat di GitHub Unduh buku catatan Lihat model TF Hub

TF-Hub adalah platform untuk berbagi keahlian pembelajaran mesin yang dikemas dalam sumber daya yang dapat digunakan kembali, terutama modul yang telah dilatih sebelumnya. Dalam tutorial ini, kita akan menggunakan modul embedding teks TF-Hub untuk melatih pengklasifikasi sentimen sederhana dengan akurasi baseline yang wajar. Kami kemudian akan mengirimkan prediksi ke Kaggle.

Untuk tutorial lebih rinci tentang klasifikasi teks dengan TF-Hub dan langkah-langkah lebih lanjut untuk meningkatkan akurasi, lihat Klasifikasi teks dengan TF-Hub .

Mendirikan

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

Karena tutorial ini akan menggunakan kumpulan data dari Kaggle, ini memerlukan pembuatan Token API untuk akun Kaggle Anda, dan mengunggahnya ke lingkungan Colab.

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()

Mulai

Data

Kami akan mencoba menyelesaikan tugas Analisis Sentimen pada Ulasan Film dari Kaggle. Dataset terdiri dari subfrasa sintaksis dari ulasan film Rotten Tomatoes. Tugasnya adalah memberi label frasa sebagai negatif atau positif pada skala dari 1 sampai 5.

Anda harus menerima aturan kompetisi sebelum Anda dapat menggunakan API untuk mengunduh data.

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)

Melatih Model

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")])
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 - loss: 1.0237 - accuracy: 0.5860 - val_loss: 0.9896 - val_accuracy: 0.5923
Epoch 2/25
4385/4385 [==============================] - 17s 4ms/step - loss: 0.9997 - accuracy: 0.5944 - val_loss: 0.9805 - val_accuracy: 0.6005
Epoch 3/25
4385/4385 [==============================] - 17s 4ms/step - loss: 0.9951 - accuracy: 0.5957 - val_loss: 0.9825 - val_accuracy: 0.5922
Epoch 4/25
4385/4385 [==============================] - 16s 4ms/step - loss: 0.9930 - accuracy: 0.5969 - val_loss: 0.9848 - val_accuracy: 0.5957
Epoch 5/25
4385/4385 [==============================] - 17s 4ms/step - loss: 0.9917 - accuracy: 0.5981 - val_loss: 0.9857 - val_accuracy: 0.5912
Epoch 6/25
4385/4385 [==============================] - 17s 4ms/step - loss: 0.9908 - accuracy: 0.5983 - val_loss: 0.9849 - val_accuracy: 0.5975
Epoch 7/25
4385/4385 [==============================] - 17s 4ms/step - loss: 0.9894 - accuracy: 0.5986 - val_loss: 0.9787 - val_accuracy: 0.6015
Epoch 8/25
4385/4385 [==============================] - 17s 4ms/step - loss: 0.9891 - accuracy: 0.5994 - val_loss: 0.9892 - val_accuracy: 0.5891
Epoch 9/25
4385/4385 [==============================] - 17s 4ms/step - loss: 0.9884 - accuracy: 0.5986 - val_loss: 0.9769 - val_accuracy: 0.5986
Epoch 10/25
4385/4385 [==============================] - 17s 4ms/step - loss: 0.9885 - accuracy: 0.5995 - val_loss: 0.9786 - val_accuracy: 0.5933
Epoch 11/25
4385/4385 [==============================] - 17s 4ms/step - loss: 0.9877 - accuracy: 0.5987 - val_loss: 0.9782 - val_accuracy: 0.6041
Epoch 12/25
4385/4385 [==============================] - 17s 4ms/step - loss: 0.9876 - accuracy: 0.5988 - val_loss: 0.9798 - val_accuracy: 0.5964
Epoch 13/25
4385/4385 [==============================] - 17s 4ms/step - loss: 0.9873 - accuracy: 0.5991 - val_loss: 0.9762 - val_accuracy: 0.5978
Epoch 14/25
4385/4385 [==============================] - 17s 4ms/step - loss: 0.9871 - accuracy: 0.5990 - val_loss: 0.9799 - val_accuracy: 0.5971
Epoch 15/25
4385/4385 [==============================] - 17s 4ms/step - loss: 0.9868 - accuracy: 0.5995 - val_loss: 0.9795 - val_accuracy: 0.5971
Epoch 16/25
4385/4385 [==============================] - 17s 4ms/step - loss: 0.9871 - accuracy: 0.5996 - val_loss: 0.9754 - val_accuracy: 0.5987
Epoch 17/25
4385/4385 [==============================] - 17s 4ms/step - loss: 0.9867 - accuracy: 0.5993 - val_loss: 0.9800 - val_accuracy: 0.5948
Epoch 18/25
4385/4385 [==============================] - 17s 4ms/step - loss: 0.9865 - accuracy: 0.5994 - val_loss: 0.9756 - val_accuracy: 0.5975
Epoch 19/25
4385/4385 [==============================] - 17s 4ms/step - loss: 0.9864 - accuracy: 0.5998 - val_loss: 0.9785 - val_accuracy: 0.5982
Epoch 20/25
4385/4385 [==============================] - 17s 4ms/step - loss: 0.9864 - accuracy: 0.5998 - val_loss: 0.9869 - val_accuracy: 0.5946
Epoch 21/25
4385/4385 [==============================] - 17s 4ms/step - loss: 0.9863 - accuracy: 0.6005 - val_loss: 0.9798 - val_accuracy: 0.5961
Epoch 22/25
4385/4385 [==============================] - 17s 4ms/step - loss: 0.9862 - accuracy: 0.5996 - val_loss: 0.9823 - val_accuracy: 0.5915
Epoch 23/25
4385/4385 [==============================] - 17s 4ms/step - loss: 0.9861 - accuracy: 0.6002 - val_loss: 0.9812 - val_accuracy: 0.5909
Epoch 24/25
4385/4385 [==============================] - 17s 4ms/step - loss: 0.9859 - accuracy: 0.5996 - val_loss: 0.9850 - val_accuracy: 0.5942
Epoch 25/25
4385/4385 [==============================] - 17s 4ms/step - loss: 0.9858 - accuracy: 0.6003 - val_loss: 0.9785 - val_accuracy: 0.5985

Ramalan

Jalankan prediksi untuk set validasi dan set pelatihan.

plt.plot(history.history['accuracy'])
plt.plot(history.history['val_accuracy'])
[<matplotlib.lines.Line2D at 0x7f1ec30faeb8>]

png

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 - loss: 0.9851 - accuracy: 0.6028
493/493 [==============================] - 1s 2ms/step - loss: 0.9785 - accuracy: 0.5985
Training set accuracy: 0.6027723550796509
Validation set accuracy: 0.5984756946563721

Matriks kebingungan

Statistik lain yang sangat menarik, terutama untuk masalah multikelas, adalah matriks konfusi . Matriks kebingungan memungkinkan visualisasi proporsi contoh yang diberi label dengan benar dan salah. Kita dapat dengan mudah melihat seberapa besar bias pengklasifikasi kita dan apakah distribusi label masuk akal. Idealnya, pecahan terbesar dari prediksi harus didistribusikan di sepanjang diagonal.

predictions = model.predict(dict(validation_df))
predictions = tf.argmax(predictions, axis=-1)
predictions
<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(32.99999999999999, 0.5, 'True')

png

Kami dapat dengan mudah mengirimkan prediksi kembali ke Kaggle dengan menempelkan kode berikut ke sel kode dan menjalankannya:

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")

Setelah mengirim, periksa papan peringkat untuk melihat kinerja Anda.