Tensorflow Debug 介绍

from tensorflow.python import debug as tf_debug
sess = tf_debug.LocalCLIDebugWrapperSession(sess)

import numpy as np
import tensorflow as tf

# Truth values for the linear model.
k_true = [[1, -1], [3, -3], [2, -2]]
b_true = [-5, 5]
num_examples = 128

with tf.Session() as sess:
    # Input place holders.
    x = tf.placeholder(tf.float32, shape=[None, 3], name="x")
    y = tf.placeholder(tf.float32, shape=[None, 2], name="y")

    # Deine model architecture, loss and training operator.
    dense_layer = tf.keras.layers.Dense(2, use_bias=True)
    y_hat = dense_layer(x)
    loss = tf.reduce_mean(
        tf.keras.losses.mean_squared_error(y, y_hat), name="loss")
    train_op = tf.train.GradientDescentOptimizer(0.05).minimize(loss)

    # Initializer model variables.
    sess.run(tf.global_variables_initializer())
    # Saver
    saver = tf.train.Saver()

    for i in range(50):
        # Generate synthetic training data.
        xs = np.random.randn(num_examples, 3)
        ys = np.matmul(xs, k_true) + b_true

        save_path = saver.save(sess, '/tmp/logdir/lin_dbg.ckpt')

        loss_val, _ = sess.run([loss, train_op], feed_dict={x: xs, y: ys})
        print("Iter %d: loss = %g" % (i, loss_val))

import numpy as np
import tensorflow as tf
from tensorflow.python import debug as tf_debug

# Truth values for the linear model.
k_true = [[1, -1], [3, -3], [2, -2]]
b_true = [-5, 5]
num_examples = 128

# Input place holders.
x = tf.placeholder(tf.float32, shape=[None, 3], name="x")
y = tf.placeholder(tf.float32, shape=[None, 2], name="y")

# Deine model architecture, loss and training operator.
dense_layer = tf.keras.layers.Dense(2, use_bias=True)
y_hat = dense_layer(x)
loss = tf.reduce_mean(
    tf.keras.losses.mean_squared_error(y, y_hat), name="loss")
train_op = tf.train.GradientDescentOptimizer(0.05).minimize(loss)

# Init OP
init_op = tf.global_variables_initializer()

# Saver
# saver = tf.train.Saver()

with tf.Session() as sess:
    # Initializer model variables.
    sess.run(init_op)

    # sess = tf_debug.TensorBoardDebugWrapperSession(sess, "yiddiEdge:7000")

    for i in range(50):
        # Generate synthetic training data.
        xs = np.random.randn(num_examples, 3)
        ys = np.matmul(xs, k_true) + b_true

        # save_path = saver.save(sess, '/tmp/logdir/lin_dbg.ckpt')

        loss_val, _ = sess.run([loss, train_op], feed_dict={x: xs, y: ys})
        print("Iter %d: loss = %g" % (i, loss_val))

sess = tf.Session()
sess = tf_debug.TensorBoardDebugWrapperSession(sess, "yiddiEdge:7000")
sess.run(my_fetches)

# To connect to the debugger using hooks, e.g., from tf.Estimator:
from tensorflow.python import debug as tf_debug
hook = tf_debug.TensorBoardDebugHook("yiddiEdge:7000")
my_estimator.fit(x=x_data, y=y_data, steps=1000, monitors=[hook])

<<get-pid>>
<<kill-pid>>
<<del-graph-summary>>
<<tensorboard-run>>
<<run-tensorboard>>

ps -aux | grep "python" | grep -E "(default|tfdbg|tensorboard)" | grep -v "grep" | awk '{print $2}'

;; 取元素
(defun r1l(tbl)
  (mapcar (lambda (x) (number-to-string (car x))) tbl)
  )
;; (print pid)
;; (print (reduce-one-layer pid))
(mapcar #'shell-command-to-string
        (mapcar (lambda (x) (concat "kill " x)) (r1l pid))))

rm -rf /tmp/logdir/*
ls /tmp/logdir

python -m tensorflow.python.debug.examples.debug_mnist