如何为已完成One-Hot编码的标签设置TensorFlow占位符

Hey there! Let's work through how to fix that labels_ placeholder based on your existing One-Hot encoding setup.

First, let's recap what your encoded labels look like:

• You’ve converted your 11-class (0-10) ratings into One-Hot encoded vectors using OneHotEncoder(sparse=False). This means your labels variable is now a dense array with shape [number_of_samples, 11]—each row is an 11-dimensional vector where exactly one position is 1 (matching the class) and the rest are 0.
• By default, scikit-learn's OneHotEncoder outputs float64 values, though TensorFlow commonly uses float32 for efficiency (we can align this in the placeholder).

Correct Placeholder Setup #

Replace your incomplete labels_ line with this:

labels_ = tf.placeholder(tf.float32, [None, 11], name='labels')

Why This Works: #

• Data Type: tf.float32 aligns with the numeric type of your encoded labels (you can use tf.float64 if you want to stick to scikit-learn’s default, but float32 is standard in TensorFlow for most models to save memory and speed up computations).
• Shape:
  • The first dimension None lets you feed batches of any size during training or inference.
  • The second dimension 11 matches exactly the number of One-Hot encoded classes you have (0 to 10, totaling 11 categories).

Quick Verification Tip #

If you want to double-check the shape of your encoded labels before defining the placeholder, run this:

print(labels.shape)  # Should output (number_of_samples, 11)

This confirms the second dimension of your placeholder matches your actual data.

内容的提问来源于stack exchange，提问作者Philip Seyfi