我创建了一个使用Dreamspark订阅在Azure上托管的Flask Web应用程序。它使用Numpy，Scipy和Theano等库。
当我使用简单的Theano计算部署应用程序时，它完美无缺。但是，当我将代码更改为更复杂的Theano计算时，我收到内部服务器错误。
以下是示例代码。当我调用simpleFunction（类似于Theano函数的总和）时，它可以工作，但是当调用complexFunction（类似于图像分类计算）时，它会创建一个内部服务器错误：
from flask import Flask
app = Flask(__name__)
wsgi_app = app.wsgi_app
from fileContainingTheanoCode import simpleFunction, complexFunction
@app.route('/')
def hello():
result = simpleFunction()
thisStr = str(result)
return """<html>
<head>
<title>Test website</title>
</head>
<body>
<h1>"""+thisStr+"""</h1>
</body>
</html>"""
if __name__ == '__main__':
HOST = os.environ.get('SERVER_HOST', 'localhost')
try:
PORT = int(os.environ.get('SERVER_PORT', '5555'))
except ValueError:
PORT = 5555
app.run(HOST, PORT)
为清楚起见，这里是文件theanoCode.py的代码（它是一个卷积神经网络分类器，加载模块用于从文件中读取mnist数据集）：
# Import Libraries----------------------------------------------------------
import os
import sys
import pickle
import theano
import numpy as np
from theano import tensor as T
from theano.tensor.nnet.conv import conv2d
from theano.tensor.signal.downsample import max_pool_2d
from theano.sandbox.rng_mrg import MRG_RandomStreams as RandomStreams
srng = RandomStreams()
import load
testSize = 10
trainSize = 2
#---------------------------------------------------------------------------
# Constants-----------------------------------------------------------------
weightsFile = 'mnist.weights'
#---------------------------------------------------------------------------
# Model---------------------------------------------------------------------
# Stuff variable into numpy array with theano float datatype
def floatX(X):
return np.asarray(X, dtype=theano.config.floatX)
# Initialize weights randomly
def init_weights(shape):
return theano.shared(np.asarray(np.random.randn(*shape) * 0.01, dtype=theano.config.floatX),borrow=True)
# ReLU - rectify linear function with Theano
def rectify(X):
return T.maximum(X, 0.)
# Softmax operation
def softmax(X):
e_x = T.exp(X - X.max(axis=1).dimshuffle(0, 'x'))
return e_x / e_x.sum(axis=1).dimshuffle(0, 'x')
# Implements random chance of ignoring a neuron output during training
def dropout(X, p=0.):
if p > 0:
retain_prob = 1 - p
X *= srng.binomial(X.shape, p=retain_prob, dtype=theano.config.floatX)
X /= retain_prob
return X
# Gradient Descent with regularization - parameter reduction
def RMSprop(cost, params, lr=0.001, rho=0.9, epsilon=1e-6):
grads = T.grad(cost=cost, wrt=params)
updates = []
for p, g in zip(params, grads):
acc = theano.shared(p.get_value() * 0.)
acc_new = rho * acc + (1 - rho) * g ** 2
gradient_scaling = T.sqrt(acc_new + epsilon)
g = g / gradient_scaling
updates.append((acc, acc_new))
updates.append((p, p - lr * g))
return updates
# Constructs the model
def model(X, w1, w2, w3, w4, p_drop_conv, p_drop_hidden):
l1a = rectify(conv2d(X, w1, border_mode='full'))
l1 = max_pool_2d(l1a, (2, 2),ignore_border=False)
l1 = dropout(l1, p_drop_conv)
l2a = rectify(conv2d(l1, w2))
l2 = max_pool_2d(l2a, (2, 2),ignore_border=False)
l2 = dropout(l2, p_drop_conv)
l3a = rectify(conv2d(l2, w3))
l3b = max_pool_2d(l3a, (2, 2),ignore_border=False)
l3 = T.flatten(l3b, outdim=2)
l3 = dropout(l3, p_drop_conv)
# Fully connected Layer
l4 = rectify(T.dot(l3, w4))
l4 = dropout(l4, p_drop_hidden)
# Classify the Output
pyx = softmax(T.dot(l4, w_o))
return l1, l2, l3, l4, pyx
# Load the data (tr = training, te = test)
trX, teX, trY, teY = load.mnist(ntrain=trainSize, ntest=testSize, onehot=True)
# Reshape training set to be 4-dimensional
# negative value is to get ordering right (rather than mirror image)
# Second parameter is color channels
trX = trX.reshape(-1, 1, 28, 28)
teX = teX.reshape(-1, 1, 28, 28)
# Input variables
X = T.dtensor4()
Y = T.fmatrix()
# WEIGHTS
if os.path.isfile(weightsFile):
# Go get saved weights from file
[w1, w2, w3, w4, w_o] = pickle.load(open(weightsFile,'rb'))
else:
# Initialize all layer weights
# (no. Inputs, no. Outputs, filter height, filter width)
# ver isto outra vez, minuto 48
w1 = init_weights((32, 1, 3, 3)) # 1 = num canais de cada imagem (cor)
w2 = init_weights((64, 32, 3, 3))
w3 = init_weights((128, 64, 3, 3))
w4 = init_weights((128 * 3 * 3, 625))
w_o = init_weights((625, 10)) # from fully connected layer to classifier
# This sets up the model graph and some vars for noise, which are the internal neurons
noise_l1, noise_l2, noise_l3, noise_l4, noise_py_x = model(X, w1, w2, w3, w4, 0.2, 0.5)
# No dropping neurons...so this will be used for prediction
l1, l2, l3, l4, py_x = model(X, w1, w2, w3, w4, 0., 0.)
# This makes predictions
y_x = T.argmax(py_x, axis=1)
# Compile prediction function - here named complexFunction
complexFunction = theano.function(inputs=[X], outputs=y_x, allow_input_downcast=True)
以下是详细错误消息：
Event: MODULE_SET_RESPONSE_ERROR_STATUS
ModuleName FastCgiModule
Notification EXECUTE_REQUEST_HANDLER
HttpStatus 500
HttpReason INTERNAL SERVER ERROR
HttpSubStatus 0
ErrorCode The operation completed successfully.
(0x0)
这是事件日志：
</Data></EventData></Event><Event><System><Provider Name="ASP.NET 4.0.30319.0"/><EventID>1325</EventID><Level>3</Level><Task>0</Task><Keywords>Keywords</Keywords><TimeCreated SystemTime="2016-06-05T19:34:20Z"/><EventRecordID>1604538437</EventRecordID><Channel>Application</Channel><Computer>RD000D3A218E0C</Computer><Security/></System><EventData><Data>An unhandled exception occurred and the process was terminated.
Application ID: /LM/W3SVC/1568611192/ROOT
Process ID: 647864
Exception: System.Configuration.ConfigurationErrorsException
Message: Couldn't find type for class Microsoft.WindowsAzure.Diagnostics.DiagnosticMonitorTraceListener, Microsoft.WindowsAzure.Diagnostics, Version=1.0.0.0, Culture=neutral, PublicKeyToken=31bf3856ad364e35.
StackTrace: at System.Diagnostics.TraceUtils.GetRuntimeObject(String className, Type baseType, String initializeData)
at System.Diagnostics.TypedElement.BaseGetRuntimeObject()
at System.Diagnostics.ListenerElement.GetRuntimeObject()
at System.Diagnostics.ListenerElementsCollection.GetRuntimeObject()
at System.Diagnostics.TraceInternal.get_Listeners()
at System.Diagnostics.TraceInternal.WriteLine(String message)
at System.Diagnostics.Debug.WriteLine(String message)
at Microsoft.Web.Compilation.Snapshots.SnapshotHelper.TakeSnapshotTimerCallback(Object stateInfo)
at System.Threading.TimerQueueTimer.CallCallbackInContext(Object state)
at System.Threading.ExecutionContext.RunInternal(ExecutionContext executionContext, ContextCallback callback, Object state, Boolean preserveSyncCtx)
at System.Threading.ExecutionContext.Run(ExecutionContext executionContext, ContextCallback callback, Object state, Boolean preserveSyncCtx)
at System.Threading.TimerQueueTimer.CallCallback()
at System.Threading.TimerQueueTimer.Fire()
at System.Threading.TimerQueue.FireNextTimers()
at System.Threading.TimerQueue.AppDomainTimerCallback()</Data></EventData></Event><Event><System><Provider Name=".NET Runtime"/><EventID>1026</EventID><Level>0</Level><Task>0</Task><Keywords>Keywords</Keywords><TimeCreated SystemTime="2016-06-05T19:34:20Z"/><EventRecordID>1604538453</EventRecordID><Channel>Application</Channel><Computer>RD000D3A218E0C</Computer><Security/></System><EventData><Data>Application: w3wp.exe
Framework Version: v4.0.30319
Description: The process was terminated due to an unhandled exception.
Exception Info: System.Configuration.ConfigurationErrorsException
at System.Diagnostics.TraceUtils.GetRuntimeObject(System.String, System.Type, System.String)
at System.Diagnostics.TypedElement.BaseGetRuntimeObject()
at System.Diagnostics.ListenerElement.GetRuntimeObject()
at System.Diagnostics.ListenerElementsCollection.GetRuntimeObject()
at System.Diagnostics.TraceInternal.get_Listeners()
at System.Diagnostics.TraceInternal.WriteLine(System.String)
at System.Diagnostics.Debug.WriteLine(System.String)
at Microsoft.Web.Compilation.Snapshots.SnapshotHelper.TakeSnapshotTimerCallback(System.Object)
at System.Threading.TimerQueueTimer.CallCallbackInContext(System.Object)
at System.Threading.ExecutionContext.RunInternal(System.Threading.ExecutionContext, System.Threading.ContextCallback, System.Object, Boolean)
at System.Threading.ExecutionContext.Run(System.Threading.ExecutionContext, System.Threading.ContextCallback, System.Object, Boolean)
at System.Threading.TimerQueueTimer.CallCallback()
at System.Threading.TimerQueueTimer.Fire()
at System.Threading.TimerQueue.FireNextTimers()
at System.Threading.TimerQueue.AppDomainTimerCallback()
</Data></EventData></Event></Events>
答案 0 :(得分：0)
导致theano问题的原因有很多，例如多线程，使用GPU，内存限制等。但根据代码和错误信息，我无法确定原因是什么。
我建议你可以尝试参考theano文件＆＃34; Debugging Theano: FAQ and Troubleshooting＆＃34;和文章＆＃34; Azure Web App sandbox＆＃34;找出由什么造成的真正问题。
你能分享一下complexFunction的代码吗？我认为它对分析问题非常有帮助，甚至尝试重现它以找出解决方案。
<强>更新：
根据您的代码和＆amp;事件日志，我认为导致这个问题的原因有两个。
某些代码逻辑错误导致内部服务器错误，但如果代码在本地运行良好，请忽略这种可能性。
您的网络应用程序web.config配置不正确，请参阅web.config部分，检查其配置是否正确。