探索 TF-Hub CORD-19 Swivel 嵌入向量

在 TensorFlow.org 上查看

在 Google Colab 中运行

在 GitHub 上查看源代码

下载笔记本

查看 TF Hub 模型

TF-Hub 上的 CORD-19 Swivel 文本嵌入向量模块 (https://tfhub.dev/tensorflow/cord-19/swivel-128d/1) 旨在支持研究人员分析与 COVID-19 相关的自然语言文本。这些嵌入向量针对 CORD-19 数据集中文章的标题、作者、摘要、正文文本和参考文献标题进行了训练。

在此 Colab 中，我们将进行以下操作：

分析嵌入向量空间中语义相似的单词
使用 CORD-19 嵌入向量在 SciCite 数据集上训练分类器

设置

import functools
import itertools
import matplotlib.pyplot as plt
import numpy as np
import seaborn as sns
import pandas as pd

import tensorflow.compat.v1 as tf
tf.disable_eager_execution()
tf.logging.set_verbosity('ERROR')

import tensorflow_datasets as tfds
import tensorflow_hub as hub

try:
  from google.colab import data_table
  def display_df(df):
    return data_table.DataTable(df, include_index=False)
except ModuleNotFoundError:
  # If google-colab is not available, just display the raw DataFrame
  def display_df(df):
    return df

2022-12-14 21:42:43.193433: W tensorflow/compiler/xla/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libnvinfer.so.7'; dlerror: libnvinfer.so.7: cannot open shared object file: No such file or directory
2022-12-14 21:42:43.193555: W tensorflow/compiler/xla/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libnvinfer_plugin.so.7'; dlerror: libnvinfer_plugin.so.7: cannot open shared object file: No such file or directory
2022-12-14 21:42:43.193586: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Cannot dlopen some TensorRT libraries. If you would like to use Nvidia GPU with TensorRT, please make sure the missing libraries mentioned above are installed properly.

分析嵌入向量

首先，我们通过计算和绘制不同术语之间的相关矩阵来分析嵌入向量。如果嵌入向量学会了成功捕获不同单词的含义，则语义相似的单词的嵌入向量应相互靠近。我们来看一些与 COVID-19 相关的术语。

# Use the inner product between two embedding vectors as the similarity measure
def plot_correlation(labels, features):
  corr = np.inner(features, features)
  corr /= np.max(corr)
  sns.heatmap(corr, xticklabels=labels, yticklabels=labels)


with tf.Graph().as_default():
  # Load the module
  query_input = tf.placeholder(tf.string)
  module = hub.Module('https://tfhub.dev/tensorflow/cord-19/swivel-128d/1')
  embeddings = module(query_input)

  with tf.train.MonitoredTrainingSession() as sess:

    # Generate embeddings for some terms
    queries = [
        # Related viruses
        "coronavirus", "SARS", "MERS",
        # Regions
        "Italy", "Spain", "Europe",
        # Symptoms
        "cough", "fever", "throat"
    ]

    features = sess.run(embeddings, feed_dict={query_input: queries})
    plot_correlation(queries, features)

2022-12-14 21:42:44.865767: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.

png

可以看到，嵌入向量成功捕获了不同术语的含义。每个单词都与其所在簇的其他单词相似（即“coronavirus”与“SARS”和“MERS”高度相关），但与其他簇的术语不同（即“SARS”与“Spain”之间的相似度接近于 0）。

现在，我们来看看如何使用这些嵌入向量解决特定任务。

SciCite：引用意图分类

本部分介绍了将嵌入向量用于下游任务（如文本分类）的方法。我们将使用 TensorFlow 数据集中的 SciCite 数据集对学术论文中的引文意图进行分类。给定一个带有学术论文引文的句子，对引用的主要意图进行分类：是背景信息、使用方法，还是比较结果。

Set up the dataset from TFDS

class Dataset:
  """Build a dataset from a TFDS dataset."""
  def __init__(self, tfds_name, feature_name, label_name):
    self.dataset_builder = tfds.builder(tfds_name)
    self.dataset_builder.download_and_prepare()
    self.feature_name = feature_name
    self.label_name = label_name

  def get_data(self, for_eval):
    splits = THE_DATASET.dataset_builder.info.splits
    if tfds.Split.TEST in splits:
      split = tfds.Split.TEST if for_eval else tfds.Split.TRAIN
    else:
      SPLIT_PERCENT = 80
      split = "train[{}%:]".format(SPLIT_PERCENT) if for_eval else "train[:{}%]".format(SPLIT_PERCENT)
    return self.dataset_builder.as_dataset(split=split)

  def num_classes(self):
    return self.dataset_builder.info.features[self.label_name].num_classes

  def class_names(self):
    return self.dataset_builder.info.features[self.label_name].names

  def preprocess_fn(self, data):
    return data[self.feature_name], data[self.label_name]

  def example_fn(self, data):
    feature, label = self.preprocess_fn(data)
    return {'feature': feature, 'label': label}, label


def get_example_data(dataset, num_examples, **data_kw):
  """Show example data"""
  with tf.Session() as sess:
    batched_ds = dataset.get_data(**data_kw).take(num_examples).map(dataset.preprocess_fn).batch(num_examples)
    it = tf.data.make_one_shot_iterator(batched_ds).get_next()
    data = sess.run(it)
  return data


TFDS_NAME = 'scicite'
TEXT_FEATURE_NAME = 'string'
LABEL_NAME = 'label'
THE_DATASET = Dataset(TFDS_NAME, TEXT_FEATURE_NAME, LABEL_NAME)

Let's take a look at a few labeled examples from the training set

NUM_EXAMPLES = 20 
data = get_example_data(THE_DATASET, NUM_EXAMPLES, for_eval=False)
display_df(
    pd.DataFrame({
        TEXT_FEATURE_NAME: [ex.decode('utf8') for ex in data[0]],
        LABEL_NAME: [THE_DATASET.class_names()[x] for x in data[1]]
    }))

训练引用意图分类器

我们将使用 Estimator 在 SciCite 数据集上对分类器进行训练。让我们设置 input_fns，将数据集读取到模型中。

def preprocessed_input_fn(for_eval):
  data = THE_DATASET.get_data(for_eval=for_eval)
  data = data.map(THE_DATASET.example_fn, num_parallel_calls=1)
  return data


def input_fn_train(params):
  data = preprocessed_input_fn(for_eval=False)
  data = data.repeat(None)
  data = data.shuffle(1024)
  data = data.batch(batch_size=params['batch_size'])
  return data


def input_fn_eval(params):
  data = preprocessed_input_fn(for_eval=True)
  data = data.repeat(1)
  data = data.batch(batch_size=params['batch_size'])
  return data


def input_fn_predict(params):
  data = preprocessed_input_fn(for_eval=True)
  data = data.batch(batch_size=params['batch_size'])
  return data

我们构建一个模型，该模型使用 CORD-19 嵌入向量，并在顶部具有一个分类层。

def model_fn(features, labels, mode, params):
  # Embed the text
  embed = hub.Module(params['module_name'], trainable=params['trainable_module'])
  embeddings = embed(features['feature'])

  # Add a linear layer on top
  logits = tf.layers.dense(
      embeddings, units=THE_DATASET.num_classes(), activation=None)
  predictions = tf.argmax(input=logits, axis=1)

  if mode == tf.estimator.ModeKeys.PREDICT:
    return tf.estimator.EstimatorSpec(
        mode=mode,
        predictions={
            'logits': logits,
            'predictions': predictions,
            'features': features['feature'],
            'labels': features['label']
        })

  # Set up a multi-class classification head
  loss = tf.nn.sparse_softmax_cross_entropy_with_logits(
      labels=labels, logits=logits)
  loss = tf.reduce_mean(loss)

  if mode == tf.estimator.ModeKeys.TRAIN:
    optimizer = tf.train.GradientDescentOptimizer(learning_rate=params['learning_rate'])
    train_op = optimizer.minimize(loss, global_step=tf.train.get_or_create_global_step())
    return tf.estimator.EstimatorSpec(mode=mode, loss=loss, train_op=train_op)

  elif mode == tf.estimator.ModeKeys.EVAL:
    accuracy = tf.metrics.accuracy(labels=labels, predictions=predictions)
    precision = tf.metrics.precision(labels=labels, predictions=predictions)
    recall = tf.metrics.recall(labels=labels, predictions=predictions)

    return tf.estimator.EstimatorSpec(
        mode=mode,
        loss=loss,
        eval_metric_ops={
            'accuracy': accuracy,
            'precision': precision,
            'recall': recall,
        })

Hyperparmeters

EMBEDDING = 'https://tfhub.dev/tensorflow/cord-19/swivel-128d/1' 
TRAINABLE_MODULE = False 
STEPS =   8000
EVAL_EVERY = 200 
BATCH_SIZE = 10 
LEARNING_RATE = 0.01 

params = {
    'batch_size': BATCH_SIZE,
    'learning_rate': LEARNING_RATE,
    'module_name': EMBEDDING,
    'trainable_module': TRAINABLE_MODULE
}

训练并评估模型

让我们训练并评估模型以查看在 SciCite 任务上的性能。

estimator = tf.estimator.Estimator(functools.partial(model_fn, params=params))
metrics = []

for step in range(0, STEPS, EVAL_EVERY):
  estimator.train(input_fn=functools.partial(input_fn_train, params=params), steps=EVAL_EVERY)
  step_metrics = estimator.evaluate(input_fn=functools.partial(input_fn_eval, params=params))
  print('Global step {}: loss {:.3f}, accuracy {:.3f}'.format(step, step_metrics['loss'], step_metrics['accuracy']))
  metrics.append(step_metrics)

2022-12-14 21:42:49.944792: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
/tmpfs/tmp/ipykernel_55626/393120678.py:7: UserWarning: `tf.layers.dense` is deprecated and will be removed in a future version. Please use `tf.keras.layers.Dense` instead.
logits = tf.layers.dense(
2022-12-14 21:42:51.606735: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 0: loss 0.797, accuracy 0.659
2022-12-14 21:42:52.907787: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:42:54.348922: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 200: loss 0.724, accuracy 0.703
2022-12-14 21:42:55.250551: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:42:56.592261: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 400: loss 0.673, accuracy 0.732
2022-12-14 21:42:57.531646: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:42:58.849640: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 600: loss 0.650, accuracy 0.739
2022-12-14 21:42:59.774741: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:43:01.085012: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 800: loss 0.626, accuracy 0.757
2022-12-14 21:43:02.507653: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:43:03.903652: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 1000: loss 0.614, accuracy 0.761
2022-12-14 21:43:04.818431: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:43:06.357206: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 1200: loss 0.609, accuracy 0.769
2022-12-14 21:43:07.310568: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:43:08.633451: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 1400: loss 0.586, accuracy 0.782
2022-12-14 21:43:09.557532: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:43:10.827364: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 1600: loss 0.581, accuracy 0.783
2022-12-14 21:43:11.789486: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:43:13.185712: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 1800: loss 0.583, accuracy 0.776
2022-12-14 21:43:14.101954: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:43:15.381008: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 2000: loss 0.571, accuracy 0.789
2022-12-14 21:43:16.333631: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:43:17.628058: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 2200: loss 0.573, accuracy 0.774
2022-12-14 21:43:18.569037: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:43:19.991966: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 2400: loss 0.559, accuracy 0.793
2022-12-14 21:43:20.936393: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:43:22.288659: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 2600: loss 0.568, accuracy 0.782
2022-12-14 21:43:23.221558: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:43:24.532759: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 2800: loss 0.562, accuracy 0.784
2022-12-14 21:43:25.520853: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:43:26.875396: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 3000: loss 0.566, accuracy 0.780
2022-12-14 21:43:27.829207: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:43:29.180763: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 3200: loss 0.556, accuracy 0.785
2022-12-14 21:43:30.125850: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:43:31.514597: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 3400: loss 0.556, accuracy 0.787
2022-12-14 21:43:32.457847: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:43:34.149400: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 3600: loss 0.558, accuracy 0.778
2022-12-14 21:43:35.143225: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:43:36.462402: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 3800: loss 0.556, accuracy 0.783
2022-12-14 21:43:37.454875: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:43:38.807818: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 4000: loss 0.554, accuracy 0.782
2022-12-14 21:43:39.770718: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:43:41.074971: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 4200: loss 0.549, accuracy 0.784
2022-12-14 21:43:42.004150: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:43:43.315299: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 4400: loss 0.551, accuracy 0.784
2022-12-14 21:43:44.238165: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:43:45.674709: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 4600: loss 0.544, accuracy 0.786
2022-12-14 21:43:46.644853: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:43:47.970517: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 4800: loss 0.539, accuracy 0.793
2022-12-14 21:43:48.923991: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:43:50.355042: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 5000: loss 0.545, accuracy 0.787
2022-12-14 21:43:51.311623: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:43:52.737111: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 5200: loss 0.542, accuracy 0.789
2022-12-14 21:43:53.681737: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:43:55.050884: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 5400: loss 0.543, accuracy 0.790
2022-12-14 21:43:55.994610: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:43:57.381780: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 5600: loss 0.539, accuracy 0.791
2022-12-14 21:43:58.291692: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:43:59.827555: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 5800: loss 0.539, accuracy 0.791
2022-12-14 21:44:00.875266: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:44:02.150170: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 6000: loss 0.533, accuracy 0.801
2022-12-14 21:44:03.069729: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:44:04.490394: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 6200: loss 0.540, accuracy 0.791
2022-12-14 21:44:05.437356: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:44:06.754939: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 6400: loss 0.537, accuracy 0.791
2022-12-14 21:44:07.864979: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:44:09.281294: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 6600: loss 0.542, accuracy 0.788
2022-12-14 21:44:10.251222: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:44:11.561167: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 6800: loss 0.538, accuracy 0.787
2022-12-14 21:44:12.524633: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:44:13.882438: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 7000: loss 0.529, accuracy 0.793
2022-12-14 21:44:14.833350: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:44:16.280598: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 7200: loss 0.540, accuracy 0.792
2022-12-14 21:44:17.252976: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:44:18.604710: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 7400: loss 0.539, accuracy 0.788
2022-12-14 21:44:19.542903: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:44:20.944670: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 7600: loss 0.539, accuracy 0.789
2022-12-14 21:44:21.886642: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
2022-12-14 21:44:23.215022: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
Global step 7800: loss 0.539, accuracy 0.790

global_steps = [x['global_step'] for x in metrics]
fig, axes = plt.subplots(ncols=2, figsize=(20,8))

for axes_index, metric_names in enumerate([['accuracy', 'precision', 'recall'],
                                            ['loss']]):
  for metric_name in metric_names:
    axes[axes_index].plot(global_steps, [x[metric_name] for x in metrics], label=metric_name)
  axes[axes_index].legend()
  axes[axes_index].set_xlabel("Global Step")

png

可以看到，损失迅速减小，而准确率迅速提高。我们绘制一些样本来检查预测与真实标签的关系：

predictions = estimator.predict(functools.partial(input_fn_predict, params))

first_10_predictions = list(itertools.islice(predictions, 10))

display_df(
  pd.DataFrame({
      TEXT_FEATURE_NAME: [pred['features'].decode('utf8') for pred in first_10_predictions],
      LABEL_NAME: [THE_DATASET.class_names()[pred['labels']] for pred in first_10_predictions],
      'prediction': [THE_DATASET.class_names()[pred['predictions']] for pred in first_10_predictions]
  }))

2022-12-14 21:44:24.825812: W tensorflow/core/common_runtime/graph_constructor.cc:1526] Importing a graph with a lower producer version 27 into an existing graph with producer version 1286. Shape inference will have run different parts of the graph with different producer versions.
/tmpfs/tmp/ipykernel_55626/393120678.py:7: UserWarning: `tf.layers.dense` is deprecated and will be removed in a future version. Please use `tf.keras.layers.Dense` instead.
  logits = tf.layers.dense(

可以看到，对于此随机样本，模型大多数时候都会预测正确的标签，这表明它可以很好地嵌入科学句子。

后续计划

现在，您已经对 TF-Hub 中的 CORD-19 Swivel 嵌入向量有了更多了解，我们鼓励您参加 CORD-19 Kaggle 竞赛，为从 COVID-19 相关学术文本中获得更深入的科学洞见做出贡献。

参加 CORD-19 Kaggle Challenge
详细了解 COVID-19 开放研究数据集 (CORD-19)
访问 https://tfhub.dev/tensorflow/cord-19/swivel-128d/1，参阅文档并详细了解 TF-Hub 嵌入向量
使用 TensorFlow Embedding Projector 探索 CORD-19 嵌入向量空间