Top-k准确率

- 日理万妓 2022-09-08 00:08 149阅读 0赞

# Top-k准确率 #

import  tensorflow as tf
    import  os
    
    os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2' #限制控制台打印日志级别
    tf.random.set_seed(2467)

def accuracy(output, target, topk=(1,)):
        # output [10,6]
        maxk = max(topk)
        batch_size = target.shape[0]
        
        pred = tf.math.top_k(output, maxk).indices # 前K个最大值的索引 [10,maxk]
    # print('每行top-6 最大值',tf.math.top_k(output, maxk).values)
        print('每行top-6 最大值下标',tf.math.top_k(output, maxk).indices)
        pred = tf.transpose(pred, perm=[1, 0]) # 转置 [maxk,10] 方便统计比较
        print('转置 每行top-6 最大值下标',pred)
        
        target_ = tf.broadcast_to(target, pred.shape) # target [10]广播成相同的形状 [maxk,10]
        print('准确值',target_)
        correct = tf.equal(pred, target_)
        print(correct)
        res = []
        for k in topk:
            # 分别计算top-1 到 top-k的准确率
            correct_k = tf.cast(tf.reshape(correct[:k], [-1]), dtype=tf.float32)
            # 展成行统计1的个数
            correct_k = tf.reduce_sum(correct_k)
            acc = float(correct_k* (100.0 / batch_size) )
            res.append(acc)
    
        return res

output = tf.random.normal([10, 6])
    output = tf.math.softmax(output, axis=1)  # 使得每行概率之和为1
    target = tf.random.uniform([10], maxval=6, dtype=tf.int32) # 生成6类样本 因此top-6准确率一定为100%
    print('prob:', output.numpy())
    
    pred = tf.argmax(output, axis=1) # argmax() 每行最大值的索引
    print('pred:', pred.numpy())
    print('label:', target.numpy())
    
    acc = accuracy(output, target, topk=(1,2,3,4,5,6))
    print('\ntop-1-6 acc:', acc)

prob: [[0.22184627 0.1221462  0.07409586 0.34996933 0.07511458 0.15682773]
     [0.14504947 0.04924896 0.21993972 0.07217006 0.08854685 0.42504498]
     [0.1542809  0.14255568 0.07593964 0.09910513 0.395387   0.13273162]
     [0.16668463 0.11025342 0.11436021 0.03997175 0.28128707 0.28744298]
     [0.05896802 0.02670637 0.3851833  0.13821954 0.27897805 0.11194469]
     [0.16363983 0.24467127 0.08584066 0.08326607 0.3861453  0.03643688]
     [0.28560603 0.22363636 0.06274495 0.26323685 0.03589749 0.12887837]
     [0.38431036 0.09477527 0.1130109  0.0730463  0.15465215 0.18020503]
     [0.05116006 0.15413527 0.16139452 0.32532734 0.14692572 0.16105707]
     [0.10868432 0.05033949 0.30445468 0.13251573 0.3949189  0.00908681]]
    pred: [3 5 4 5 2 4 0 0 3 4]
    label: [1 3 5 2 1 0 4 4 1 1]
    每行top-6 最大值下标 tf.Tensor(
    [[3 0 5 1 4 2]
     [5 2 0 4 3 1]
     [4 0 1 5 3 2]
     [5 4 0 2 1 3]
     [2 4 3 5 0 1]
     [4 1 0 2 3 5]
     [0 3 1 5 2 4]
     [0 5 4 2 1 3]
     [3 2 5 1 4 0]
     [4 2 3 0 1 5]], shape=(10, 6), dtype=int32)
    转置 每行top-6 最大值下标 tf.Tensor(
    [[3 5 4 5 2 4 0 0 3 4]
     [0 2 0 4 4 1 3 5 2 2]
     [5 0 1 0 3 0 1 4 5 3]
     [1 4 5 2 5 2 5 2 1 0]
     [4 3 3 1 0 3 2 1 4 1]
     [2 1 2 3 1 5 4 3 0 5]], shape=(6, 10), dtype=int32)
    准确值 tf.Tensor(
    [[1 3 5 2 1 0 4 4 1 1]
     [1 3 5 2 1 0 4 4 1 1]
     [1 3 5 2 1 0 4 4 1 1]
     [1 3 5 2 1 0 4 4 1 1]
     [1 3 5 2 1 0 4 4 1 1]
     [1 3 5 2 1 0 4 4 1 1]], shape=(6, 10), dtype=int32)
    tf.Tensor(
    [[False False False False False False False False False False]
     [False False False False False False False False False False]
     [False False False False False  True False  True False False]
     [ True False  True  True False False False False  True False]
     [False  True False False False False False False False  True]
     [False False False False  True False  True False False False]], shape=(6, 10), dtype=bool)
    
    top-1-6 acc: [0.0, 0.0, 20.0, 60.0, 80.0, 100.0]