How to solve a problem on Kaggle with TF-Hub

View on TensorFlow.org Run in Google Colab View on GitHub Download notebook See TF Hub model

TF-Hub is a platform to share machine learning expertise packaged in reusable resources, notably pre-trained modules. In this tutorial, we will use a TF-Hub text embedding module to train a simple sentiment classifier with a reasonable baseline accuracy. We will then submit the predictions to Kaggle.

For more detailed tutorial on text classification with TF-Hub and further steps for improving the accuracy, take a look at Text classification with TF-Hub.

Setup

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

Since this tutorial will be using a dataset from Kaggle, it requires creating an API Token for your Kaggle account, and uploading it to the Colab environment.

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()

Getting started

Data

We will try to solve the Sentiment Analysis on Movie Reviews task from Kaggle. The dataset consists of syntactic subphrases of the Rotten Tomatoes movie reviews. The task is to label the phrases as negative or positive on the scale from 1 to 5.

You must accept the competition rules before you can use the API to download the 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)

Training an 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: 1.0238 - accuracy: 0.5858 - val_loss: 0.9840 - val_accuracy: 0.6025
Epoch 2/25
4385/4385 [==============================] - 16s 4ms/step - loss: 1.0001 - accuracy: 0.5951 - val_loss: 0.9906 - val_accuracy: 0.5864
Epoch 3/25
4385/4385 [==============================] - 16s 4ms/step - loss: 0.9955 - accuracy: 0.5961 - val_loss: 0.9908 - val_accuracy: 0.5891
Epoch 4/25
4385/4385 [==============================] - 17s 4ms/step - loss: 0.9931 - accuracy: 0.5971 - val_loss: 0.9857 - val_accuracy: 0.5968
Epoch 5/25
4385/4385 [==============================] - 16s 4ms/step - loss: 0.9917 - accuracy: 0.5970 - val_loss: 0.9768 - val_accuracy: 0.5970
Epoch 6/25
4385/4385 [==============================] - 16s 4ms/step - loss: 0.9908 - accuracy: 0.5976 - val_loss: 0.9913 - val_accuracy: 0.5907
Epoch 7/25
4385/4385 [==============================] - 17s 4ms/step - loss: 0.9899 - accuracy: 0.5983 - val_loss: 1.0171 - val_accuracy: 0.5777
Epoch 8/25
4385/4385 [==============================] - 16s 4ms/step - loss: 0.9896 - accuracy: 0.5987 - val_loss: 0.9837 - val_accuracy: 0.5976
Epoch 9/25
4385/4385 [==============================] - 16s 4ms/step - loss: 0.9889 - accuracy: 0.5996 - val_loss: 0.9825 - val_accuracy: 0.5956
Epoch 10/25
4385/4385 [==============================] - 17s 4ms/step - loss: 0.9885 - accuracy: 0.5987 - val_loss: 0.9826 - val_accuracy: 0.6007
Epoch 11/25
4385/4385 [==============================] - 17s 4ms/step - loss: 0.9879 - accuracy: 0.5996 - val_loss: 0.9850 - val_accuracy: 0.5931
Epoch 12/25
4385/4385 [==============================] - 16s 4ms/step - loss: 0.9876 - accuracy: 0.5987 - val_loss: 0.9872 - val_accuracy: 0.5932
Epoch 13/25
4385/4385 [==============================] - 16s 4ms/step - loss: 0.9876 - accuracy: 0.5989 - val_loss: 0.9817 - val_accuracy: 0.5963
Epoch 14/25
4385/4385 [==============================] - 16s 4ms/step - loss: 0.9870 - accuracy: 0.5996 - val_loss: 0.9803 - val_accuracy: 0.5951
Epoch 15/25
4385/4385 [==============================] - 16s 4ms/step - loss: 0.9869 - accuracy: 0.5993 - val_loss: 0.9786 - val_accuracy: 0.5979
Epoch 16/25
4385/4385 [==============================] - 16s 4ms/step - loss: 0.9867 - accuracy: 0.5998 - val_loss: 0.9781 - val_accuracy: 0.5938
Epoch 17/25
4385/4385 [==============================] - 16s 4ms/step - loss: 0.9864 - accuracy: 0.5996 - val_loss: 0.9790 - val_accuracy: 0.5963
Epoch 18/25
4385/4385 [==============================] - 16s 4ms/step - loss: 0.9863 - accuracy: 0.5988 - val_loss: 0.9828 - val_accuracy: 0.6006
Epoch 19/25
4385/4385 [==============================] - 16s 4ms/step - loss: 0.9865 - accuracy: 0.5996 - val_loss: 0.9771 - val_accuracy: 0.5997
Epoch 20/25
4385/4385 [==============================] - 16s 4ms/step - loss: 0.9862 - accuracy: 0.5994 - val_loss: 0.9742 - val_accuracy: 0.5940
Epoch 21/25
4385/4385 [==============================] - 16s 4ms/step - loss: 0.9862 - accuracy: 0.6004 - val_loss: 0.9784 - val_accuracy: 0.6003
Epoch 22/25
4385/4385 [==============================] - 16s 4ms/step - loss: 0.9860 - accuracy: 0.5994 - val_loss: 0.9830 - val_accuracy: 0.5907
Epoch 23/25
4385/4385 [==============================] - 16s 4ms/step - loss: 0.9857 - accuracy: 0.6006 - val_loss: 0.9814 - val_accuracy: 0.5946
Epoch 24/25
4385/4385 [==============================] - 16s 4ms/step - loss: 0.9856 - accuracy: 0.6001 - val_loss: 0.9741 - val_accuracy: 0.6039
Epoch 25/25
4385/4385 [==============================] - 16s 4ms/step - loss: 0.9857 - accuracy: 0.6004 - val_loss: 0.9776 - val_accuracy: 0.6018

Prediction

Run predictions for the validation set and training set.

plt.plot(history.history['accuracy'])
plt.plot(history.history['val_accuracy'])
[<matplotlib.lines.Line2D at 0x7fe728d627f0>]

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 [==============================] - 15s 4ms/step - loss: 0.9844 - accuracy: 0.6026
493/493 [==============================] - 1s 2ms/step - loss: 0.9776 - accuracy: 0.6018
Training set accuracy: 0.6025514006614685
Validation set accuracy: 0.6017783284187317

Confusion matrix

Another very interesting statistic, especially for multiclass problems, is the confusion matrix. The confusion matrix allows visualization of the proportion of correctly and incorrectly labelled examples. We can easily see how much our classifier is biased and whether the distribution of labels makes sense. Ideally the largest fraction of predictions should be distributed along the 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

We can easily submit the predictions back to Kaggle by pasting the following code to a code cell and executing it:

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")

After submitting, check the leaderboard to see how you did.