Cómo resolver un problema en Kaggle con TF-Hub

Ver en TensorFlow.org Ejecutar en Google Colab Ver en GitHub Descargar cuaderno Ver modelo TF Hub

TF-Hub es una plataforma para compartir experiencias de aprendizaje automático empaquetado en recursos reutilizables, en particular módulos pre-formados. En este tutorial, usaremos un módulo de incrustación de texto TF-Hub para entrenar un clasificador de sentimiento simple con una precisión de referencia razonable. Luego, enviaremos las predicciones a Kaggle.

Por tutorial más detallado de clasificación de texto con el TF-Hub y nuevas medidas para mejorar la exactitud, echar un vistazo a la clasificación del texto con el TF-Hub .

Configuración

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

Desde este tutorial va a utilizar un conjunto de datos de Kaggle, se requiere la creación de una API de emergencia para su cuenta Kaggle, y enviarlo al medio ambiente 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()

Empezando

Datos

Vamos a tratar de resolver el Sentiment Analysis sobre las revisiones de la película tarea de Kaggle. El conjunto de datos consta de subfrases sintácticas de las reseñas de películas de Rotten Tomatoes. La tarea es etiquetar las frases como negativo o positivo en la escala de 1 a 5.

Debe aceptar las normas de competencia antes de poder utilizar la API para descargar los datos.

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)

Entrenamiento de un modelo

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 [==============================] - 16s 3ms/step - loss: 1.0237 - accuracy: 0.5869 - val_loss: 1.0023 - val_accuracy: 0.5870
Epoch 2/25
4385/4385 [==============================] - 15s 3ms/step - loss: 0.9995 - accuracy: 0.5941 - val_loss: 0.9903 - val_accuracy: 0.5952
Epoch 3/25
4385/4385 [==============================] - 15s 3ms/step - loss: 0.9946 - accuracy: 0.5967 - val_loss: 0.9811 - val_accuracy: 0.6011
Epoch 4/25
4385/4385 [==============================] - 15s 3ms/step - loss: 0.9924 - accuracy: 0.5971 - val_loss: 0.9851 - val_accuracy: 0.5935
Epoch 5/25
4385/4385 [==============================] - 15s 3ms/step - loss: 0.9912 - accuracy: 0.5988 - val_loss: 0.9896 - val_accuracy: 0.5934
Epoch 6/25
4385/4385 [==============================] - 15s 3ms/step - loss: 0.9896 - accuracy: 0.5984 - val_loss: 0.9810 - val_accuracy: 0.5936
Epoch 7/25
4385/4385 [==============================] - 15s 3ms/step - loss: 0.9892 - accuracy: 0.5978 - val_loss: 0.9845 - val_accuracy: 0.5994
Epoch 8/25
4385/4385 [==============================] - 15s 3ms/step - loss: 0.9889 - accuracy: 0.5996 - val_loss: 0.9772 - val_accuracy: 0.6015
Epoch 9/25
4385/4385 [==============================] - 15s 3ms/step - loss: 0.9880 - accuracy: 0.5992 - val_loss: 0.9798 - val_accuracy: 0.5991
Epoch 10/25
4385/4385 [==============================] - 15s 3ms/step - loss: 0.9879 - accuracy: 0.6002 - val_loss: 0.9869 - val_accuracy: 0.5935
Epoch 11/25
4385/4385 [==============================] - 15s 3ms/step - loss: 0.9878 - accuracy: 0.5998 - val_loss: 0.9790 - val_accuracy: 0.5985
Epoch 12/25
4385/4385 [==============================] - 14s 3ms/step - loss: 0.9871 - accuracy: 0.5999 - val_loss: 0.9845 - val_accuracy: 0.5964
Epoch 13/25
4385/4385 [==============================] - 15s 3ms/step - loss: 0.9871 - accuracy: 0.6001 - val_loss: 0.9800 - val_accuracy: 0.5947
Epoch 14/25
4385/4385 [==============================] - 15s 3ms/step - loss: 0.9873 - accuracy: 0.6001 - val_loss: 0.9810 - val_accuracy: 0.5934
Epoch 15/25
4385/4385 [==============================] - 14s 3ms/step - loss: 0.9865 - accuracy: 0.5988 - val_loss: 0.9824 - val_accuracy: 0.5898
Epoch 16/25
4385/4385 [==============================] - 15s 3ms/step - loss: 0.9865 - accuracy: 0.5993 - val_loss: 0.9779 - val_accuracy: 0.5974
Epoch 17/25
4385/4385 [==============================] - 15s 3ms/step - loss: 0.9866 - accuracy: 0.5991 - val_loss: 0.9785 - val_accuracy: 0.5972
Epoch 18/25
4385/4385 [==============================] - 15s 3ms/step - loss: 0.9863 - accuracy: 0.6001 - val_loss: 0.9803 - val_accuracy: 0.5991
Epoch 19/25
4385/4385 [==============================] - 16s 4ms/step - loss: 0.9863 - accuracy: 0.5996 - val_loss: 0.9773 - val_accuracy: 0.5957
Epoch 20/25
4385/4385 [==============================] - 15s 3ms/step - loss: 0.9862 - accuracy: 0.5995 - val_loss: 0.9744 - val_accuracy: 0.6009
Epoch 21/25
4385/4385 [==============================] - 15s 3ms/step - loss: 0.9861 - accuracy: 0.5997 - val_loss: 0.9787 - val_accuracy: 0.5968
Epoch 22/25
4385/4385 [==============================] - 15s 3ms/step - loss: 0.9855 - accuracy: 0.5998 - val_loss: 0.9794 - val_accuracy: 0.5976
Epoch 23/25
4385/4385 [==============================] - 14s 3ms/step - loss: 0.9861 - accuracy: 0.5998 - val_loss: 0.9778 - val_accuracy: 0.5966
Epoch 24/25
4385/4385 [==============================] - 15s 3ms/step - loss: 0.9860 - accuracy: 0.5999 - val_loss: 0.9831 - val_accuracy: 0.5912
Epoch 25/25
4385/4385 [==============================] - 14s 3ms/step - loss: 0.9858 - accuracy: 0.5999 - val_loss: 0.9780 - val_accuracy: 0.5977

Predicción

Ejecute predicciones para el conjunto de validación y el conjunto de entrenamiento.

plt.plot(history.history['accuracy'])
plt.plot(history.history['val_accuracy'])
[<matplotlib.lines.Line2D at 0x7f62684da090>]

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 [==============================] - 14s 3ms/step - loss: 0.9834 - accuracy: 0.6007
493/493 [==============================] - 1s 2ms/step - loss: 0.9780 - accuracy: 0.5977
Training set accuracy: 0.6006770730018616
Validation set accuracy: 0.5976500511169434

Matriz de confusión

Otra estadística muy interesante, especialmente para los problemas multiclase, es la matriz de confusión . La matriz de confusión permite visualizar la proporción de ejemplos etiquetados correcta e incorrectamente. Podemos ver fácilmente cuánto está sesgado nuestro clasificador y si la distribución de etiquetas tiene sentido. Idealmente, la fracción más grande de predicciones debería distribuirse a lo largo de la 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

Podemos enviar fácilmente las predicciones a Kaggle pegando el siguiente código en una celda de código y ejecutándolo:

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")

Después de la presentación, compruebe la tabla de clasificación para ver cómo lo hizo.