ValueError：传递值的长度是 1445，索引意味着 1

Question

我正在尝试进行一个非常简单的梯度下降实现，但在执行该函数时出现此错误，特别是指向在每个循环开始时计算预测值的 np.dot 函数。它给了我一些“ValueError：传递值的长度是 1445，索引意味着 1。”尽管点积是正确的，但它是 (1445, 4) * (4,1)。循环成功完成第一次迭代，然后抛出错误

代码如下：`

def gredientDescent(inputs, outputs, learning_rate) : 

weights = np.zeros((4,1))
bias = 0



for i in range(num_observations) :
    
    print(weights)

    
    predicted = np.dot(inputs, weights) +  bias



    
    deltas = predicted - outputs        
    
    
    cost = np.sum(deltas ** 2) / num_observations
    
    
    dw = np.dot(inputs.T, deltas)
    db = np.sum(deltas)
    
    
    
    weights =  weights - (learning_rate * dw)
    bias = bias - (learning_rate * db)
    
    print(weights)
            
    

    
gredientDescent(inputs,outputs, 0.001)

`

以及出现的错误：

ValueError                                Traceback (most recent call last)
<ipython-input-177-5517d5583095> in <module>
     38 
     39 
---> 40 gredientDescent(inputs,outputs, 0.001)

<ipython-input-177-5517d5583095> in gredientDescent(inputs, outputs, learning_rate)
     11 
     12 
---> 13         predicted = np.dot(inputs, weights) +  bias
     14 
     15 

C:\anaconda3\envs\py3tf2\lib\site-packages\pandas\core\series.py in __array_ufunc__(self, ufunc, method, *inputs, **kwargs)
    634         # for binary ops, use our custom dunder methods
    635         result = ops.maybe_dispatch_ufunc_to_dunder_op(
--> 636             self, ufunc, method, *inputs, **kwargs
    637         )
    638         if result is not NotImplemented:

pandas\_libs\ops_dispatch.pyx in pandas._libs.ops_dispatch.maybe_dispatch_ufunc_to_dunder_op()

C:\anaconda3\envs\py3tf2\lib\site-packages\pandas\core\ops\common.py in new_method(self, other)
     62         other = item_from_zerodim(other)
     63 
---> 64         return method(self, other)
     65 
     66     return new_method

C:\anaconda3\envs\py3tf2\lib\site-packages\pandas\core\ops\__init__.py in wrapper(left, right)
    503         result = arithmetic_op(lvalues, rvalues, op, str_rep)
    504 
--> 505         return _construct_result(left, result, index=left.index, name=res_name)
    506 
    507     wrapper.__name__ = op_name

C:\anaconda3\envs\py3tf2\lib\site-packages\pandas\core\ops\__init__.py in _construct_result(left, result, index, name)
    476     # We do not pass dtype to ensure that the Series constructor
    477     #  does inference in the case where `result` has object-dtype.
--> 478     out = left._constructor(result, index=index)
    479     out = out.__finalize__(left)
    480 

C:\anaconda3\envs\py3tf2\lib\site-packages\pandas\core\series.py in __init__(self, data, index, dtype, name, copy, fastpath)
    290                     if len(index) != len(data):
    291                         raise ValueError(
--> 292                             f"Length of passed values is {len(data)}, "
    293                             f"index implies {len(index)}."
    294                         )

ValueError: Length of passed values is 1445, index implies 1.

Answer 1

这应该可行：

weights = np.zeros((4,1))
bias = 0



for i in range(num_observations) :

    print(weights)


    predicted = np.dot(inputs, weights.T) +  bias




    deltas = predicted - outputs        


    cost = np.sum(deltas ** 2) / num_observations


    dw = np.dot(inputs.T, deltas)
    db = np.sum(deltas)



    weights =  weights - (learning_rate * dw)
    bias = bias - (learning_rate * db)

    print(weights)
        



gredientDescent(inputs,outputs, 0.001)

注意：

1. predicted = np.dot(inputs, weights) + bias 更改为 predicted = np.dot(inputs, weights.T) + bias
2. np.dot(inputs.T, deltas) 改为 np.dot(inputs, deltas.T)

，因为您希望通过其权重获得每个输入的乘法总和，而不是每个输入的权重乘法总和。 希望这可以帮助。干杯。