我試圖將圖像預測到我的預訓練深度學習分類模型

Question

當我使用驗證數據進行測試時，當我在保存模型並預測輸出后嘗試發送單個圖像時，返回的輸出具有該形狀 [7.2918165e-06 3.2451030e-02 9.6753991e-01 1.7616502e-06]形狀 [0. 1. 0. 0.] 有什么問題

Answer 1

使用 model.predict( ) 應該返回與模型輸出相同的形狀，其有效值與輸入共同響應，並且當前訓練權重在可訓練的過程中具有微小的變化。

具有激活的最后一層的輸出，目標層輸出吸收優化器或損失 Fn。 當您使用層輸出時，一些輸出是二進制、熵、logits 或多個值。 （看形狀是 [ 2, 1 ] ）

[ 樣本 ]：

from os.path import exists

import tensorflow as tf
import matplotlib.pyplot as plt
import numpy as np

"""""""""""""""""""""""""""""""""""""""""""""""""""""""""
[PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]
None
"""""""""""""""""""""""""""""""""""""""""""""""""""""""""
physical_devices = tf.config.experimental.list_physical_devices('GPU')
assert len(physical_devices) > 0, "Not enough GPU hardware devices available"
config = tf.config.experimental.set_memory_growth(physical_devices[0], True)
print(physical_devices)
print(config)   

"""""""""""""""""""""""""""""""""""""""""""""""""""""""""
: Variables
"""""""""""""""""""""""""""""""""""""""""""""""""""""""""
learning_rate = 0.001
global_step = 0
vocabulary_size = 5000
start = 0
limit = 128
delta = 1
embedding_size = 16
n_sample = 16

tf.compat.v1.disable_eager_execution()

# Input data.
inputs = tf.compat.v1.get_variable('X', dtype = tf.int32, initializer = tf.random.uniform(shape=[1], maxval=1, dtype=tf.int32, seed=10))
labels = tf.compat.v1.get_variable('Y', dtype = tf.int32, initializer = tf.random.uniform(shape=[1, 1], maxval=1, dtype=tf.int32, seed=10))

# Look up embeddings for inputs.
embeddings = tf.Variable(
tf.random.uniform([vocabulary_size, embedding_size], -1.0, 1.0)
)
embed = tf.nn.embedding_lookup(embeddings, inputs)

# Construct the variables for the NCE loss
nce_weights = tf.Variable(
    tf.random.uniform(shape=[vocabulary_size, embedding_size], maxval=255, dtype=tf.float32,)
)
nce_biases = tf.Variable(tf.zeros([vocabulary_size]))

# time we evaluate the loss.
loss = tf.reduce_mean(
tf.nn.nce_loss(nce_weights, nce_biases,
               labels=labels,
               inputs=embed,
               num_sampled=n_sample,
               num_classes=vocabulary_size),
name='loss'
)
optimizer = tf.compat.v1.train.ProximalAdagradOptimizer(
    learning_rate,
    initial_accumulator_value=0.1,
    l1_regularization_strength=0.2,
    l2_regularization_strength=0.1,
    use_locking=False,
    name='ProximalAdagrad'
)
training_op = optimizer.minimize(loss, name='minimize') 


"""""""""""""""""""""""""""""""""""""""""""""""""""""""""
: DataSet / Input
"""""""""""""""""""""""""""""""""""""""""""""""""""""""""
X = np.reshape([ 500 ], (1))
Y = np.reshape([ 15 ], (1, 1))
history = [ ] 

"""""""""""""""""""""""""""""""""""""""""""""""""""""""""
: Training / Optimize
"""""""""""""""""""""""""""""""""""""""""""""""""""""""""
with tf.compat.v1.Session() as sess:
    sess.run(tf.compat.v1.global_variables_initializer())
    
    
    first_value = 500
    second_value = 15
    for i in range(1000):
        global_step = global_step + 1
        train_loss, temp = sess.run([loss, training_op], feed_dict={inputs:X, labels:Y})
        history.append(train_loss)
        
        print( 'steps: ' + str(i) )
        
        a = tf.constant([ first_value ])
        b = tf.constant([ second_value ])
        v = sess.run( [a, b] )
        print( v )
        
        first_value = first_value - i
        second_value = second_value - i
        
sess.close()

print( history )
plt.plot(history)
plt.show()
plt.close()

input('...')

[ 輸出 ]：