Google Machine Learning API Issue - google-cloud-ml-engine

Question

I'm trying to use the Google Machine Learning API and I'm facing two problems.
In the API explorer I put the correct information and I get a response error:
Code 200
"error": "Missing \"instances\" field in request body: {\n \"httpBody\": \n
{\n \"data\": \"\\"instances\\" : \\"teste\\"\",\n
\"contentType\": \"application/json\"\n }\n}"
The request find my model (if I change the value in field name I get another error) but don't understand my json. That's the json:
{"instances" : [{"key":"0", "image_bytes": {"b64": "mybase64"} }]}
When I do the predict on the command line using gcloud, I get no errors and everything seems ok. The Json that I was create for gcloud is a little bit different:
{"key":"0", "image_bytes": {"b64": "mybase64"} }
I already tryied that one in the API explorer and no success.
So, I decided to use the .Net Api to try the predict and I get other situation: The Response is Empty (???).
Here is my code:
'get the service credential that I created
Dim credential = Await GetCredential()
Dim myService As New CloudMachineLearningEngineService(New BaseClientService.Initializer() With {
.ApplicationName = "my Project Name (Is That It???)",
.ApiKey = "my API Key",
.HttpClientInitializer = credential
})
Dim myBase64 As String = GetBase64("my image path to convert into a base64 String")
Dim myJsonRequest As String = "{""instances"" : [{""key"":""0"", ""image_bytes"": {""b64"": """ + myBase64 + """}}]}"
Dim myRequest = New GoogleCloudMlV1PredictRequest With {
.HttpBody = New GoogleApiHttpBody With {.Data = myJsonRequest,
.ContentType = "application/json"
}
}
'If I change the model name I get error
Dim myPredictRequest = myService.Projects.Predict(myRequest, "projects/myProject/models/myModel/versions/v1")
myPredictRequest.AccessToken = credential.Token.AccessToken
myPredictRequest.OauthToken = credential.Token.AccessToken
myPredictRequest.Key = "my API Key
'Execute the request
Dim myResponse = myPredictRequest.Execute()
'at this point, myResponse is Empty (myResponse.ContentType Is Nothing, myResponse.Data Is Nothing And myResponse.ETag Is Nothing)
If I change the model name I get a error informing that my model was not found, so my credentials are right.
I don't know what I'm doing wrong. Someboby can help with any of this issues?
Thanks!
UPDATE: --------------------------
I changed this Execute Command:
Dim myResponse = myPredictRequest.Execute()
To This One:
Dim s = StreamToString(myPredictRequest.ExecuteAsStream())
and Now I can get the same error with .Net API and google developers interface (Missing instances field...).
So If someboby just Know what is wrong with my Json request, It will help a lot.

Answer 1

The JSON you put in the API explorer is indeed correct (assuming, of course, your model has inputs key and image_bytes). This appears to be a bug with the explorer I will report.
The reason you are getting the error you are in the .NET code is because you are using an .HttpBody field. This code:
Dim myJsonRequest As String = "{""instances"" : [{""key"":""0"", ""image_bytes"": {""b64"": """ + myBase64 + """}}]}"
Dim myRequest = New GoogleCloudMlV1PredictRequest With {
.HttpBody = New GoogleApiHttpBody With {.Data = myJsonRequest,
.ContentType = "application/json"
}
}
Will produce a JSON request that looks like this:
{
"httpBody": {
"data": "{\"instances\" : [{\"key\":\"0\", \"image_bytes\": {\"b64\": \"mybase64\"} }]}",
"contentType": "application\/json"
}
}
When what you really need is:
{"instances" : [{"key":"0", "image_bytes": {"b64": "mybase64"} }]}
Hence the error message you see.
I don't know how to generate the correct response using the .NET library; based on the Python example in the docs, I would guess:
Dim myJsonRequest As String = "{""instances"" : [{""key"":""0"", ""image_bytes"": {""b64"": """ + myBase64 + """}}]}"
Dim myPredictRequest = myService.Projects.Predict(myJsonRequest, "projects/myProject/models/myModel/versions/v1")
But I don't have a good way of testing that. For reference, the Python equivalent is:
response = service.projects().predict(
name=name,
body=myJsonRequest
).execute()

Answer 2

I solved the problem with .Net API.
I created two new classes Inherits the Google API's classes.
Something like that:
Imports Google.Apis.CloudMachineLearningEngine.v1.Data
Imports Newtonsoft.Json
Public Class myGoogleCloudMlV1PredictRequest
Inherits GoogleCloudMlV1PredictRequest
<JsonProperty("instances")>
Public Property MyHttpBody As List(Of myGoogleApiHttpBody)
End Class
Imports Google.Apis.CloudMachineLearningEngine.v1.Data
Imports Newtonsoft.Json
Public Class myGoogleApiHttpBody
Inherits GoogleApiHttpBody
<JsonProperty("image_bytes")>
Public Property MyData As image_byte
<JsonProperty("key")>
Public Property key As String
End Class
So, in my original code I change this part:
Dim myBase64 As String = GetBase64("my_image_path_to_convert_into_a _base64_String")
Dim myJsonRequest As String = "{""instances"" : [{""key"":""0"", ""image_bytes"": {""b64"": """ + myBase64 + """}}]}"
Dim myRequest = New GoogleCloudMlV1PredictRequest With {
.HttpBody = New GoogleApiHttpBody With {.Data = myJsonRequest,
.ContentType = "application/json"
}
}
For this one:
Dim myBase64 As String = GetBase64("my_image_path_to_convert_into_a _base64_String")
Dim myRequest = New myGoogleCloudMlV1PredictRequest With {
.MyHttpBody = New List(Of myGoogleApiHttpBody)()
}
Dim item As myGoogleApiHttpBody = New myGoogleApiHttpBody With {
.key = "0",
.MyData = New image_byte With {
.b64 = myBase64
}
}
myRequest.MyHttpBody.Add(item)
And voilá, It's working!
Thanks for everyone!!

Answer 3

Github issue #1068 shows two work-arounds for this problem.
In summary, use service.ModifyRequest to insert the raw JSON content.
Or use service.HttpClient.PostAsync(...) directly.