TensorFlow中map_fn结合if条件处理列向量失败求助

Hey there! I see the problem with your second map_fn call—let's break it down and fix it.

The issue here is that in TensorFlow 1.x's graph execution mode (where you're using tf.Session), you can't use Python-native control flow or operators like if-else or % directly on TensorFlow tensors. These Python constructs run during graph construction time, not at runtime when the actual tensor values are available. TensorFlow requires you to use its own symbolic operations to build the computation graph correctly.

Here's the corrected code: #

import numpy as np
import tensorflow as tf

elems = np.array([[1.0], [2.0], [3.0]])

# Replace Python if-else with tf.cond, and % with tf.math.mod
tf_m2 = tf.map_fn(
    lambda x: tf.cond(
        tf.math.mod(x, 2) > 0,  # TensorFlow-compatible condition check
        lambda: x + 1,          # Branch if condition is True
        lambda: x               # Branch if condition is False
    ),
    elems
)

with tf.Session() as sess:
    res = sess.run(tf_m2)
    print(res)

Why this works: #

• tf.math.mod() replaces Python's % operator: it performs modular arithmetic directly on tensors, which integrates properly with TensorFlow's computation graph.
• tf.cond() replaces Python's if-else: it creates a conditional branch within the graph that evaluates the tensor condition at runtime, applying the correct operation to each element in your column vector.

Running this code will give you the expected output: [[2.] [2.] [4.]]

内容的提问来源于stack exchange，提问作者David Bolding