Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Activity
lqdev
@lqdev:lqdev.tech
[m]
Also if you're doing batch storing, you want to use the Transform method of you rmodel. This allows you to make multiple predictions.
uzfm
@uzfm

ML.NET support CUDA 11
Loading model from: F:\V2Sorter\DataSet\03.01.2021\imageClassifier_2.zip
2021-04-16 16:06:33.457209: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1720] Found device 0 with properties:
pciBusID: 0000:03:00.0 name: GeForce RTX 3080 computeCapability: 8.6
coreClock: 1.905GHz coreCount: 68 deviceMemorySize: 10.00GiB deviceMemoryBandwidth: 707.88GiB/s
2021-04-16 16:06:33.458465: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library cudart64_110.dll
2021-04-16 16:06:33.459114: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library cublas64_11.dll
2021-04-16 16:06:33.459639: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library cublasLt64_11.dll
2021-04-16 16:06:33.460164: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library cufft64_10.dll
2021-04-16 16:06:33.460705: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library curand64_10.dll
2021-04-16 16:06:33.461270: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library cusolver64_10.dll
2021-04-16 16:06:33.461946: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library cusparse64_11.dll
2021-04-16 16:06:33.462589: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library cudnn64_8.dll
2021-04-16 16:06:33.463409: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1862] Adding visible gpu devices: 0
2021-04-16 16:06:33.464111: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1261] Device interconnect StreamExecutor with strength 1 edge matrix:
2021-04-16 16:06:33.464830: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1267] 0
2021-04-16 16:06:33.465221: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1280] 0: N
2021-04-16 16:06:33.465788: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1406] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 7745 MB memory) -> physical GPU (device: 0, name: GeForce RTX 3080, pci bus id: 0000:03:00.0, compute capability: 8.6)
Exception thrown: 'System.EntryPointNotFoundException' in Microsoft.ML.Vision.dll
device is disconnected
Exception thrown: 'System.IO.FileNotFoundException' in mscorlib.dll
Exception thrown: 'System.IO.FileNotFoundException' in mscorlib.dll
The thread 0x2f80 has exited with code 0 (0x0).

I'm using "SciSharp.TensorFlow.Redist-Windows-GPU" with CUDA 2.4.0 support but get the exception "Exception thrown: 'System.EntryPointNotFoundException' in Microsoft.ML.Vision.dll"
what could be the problem?

lqdev
@lqdev:lqdev.tech
[m]
I believe ML.NET works with Cuda 10.1, not 11
uzfm
@uzfm
I am using a GeForce RTX 3080 GPU it only works with CUDA 11.
When will ML.NET use CUDA 11?
HashGrammer
@HashGrammer
This may be too "general knowledge" for here, so let me know and I'll remove it if so, but does anyone know how to get intellisense recommendations in VS 2017 like they are in 2019? For example, I'm designing a GUI to allow input to train the model. When I add a <Grid>, the first two recommendations when I open a bracket in 2019 are Grid.RowDefinitions and Grid.CollumnDefinitions with little stars next to them THEN the A-Z list. In 2017 I get only A-Z, with no "star" choices.
I'm even willing to use a marketplace suggestion (something free, not resharper) that can do this if it isn't available by default in anything older than 2019. I have googled this to no avail. I don't think I refined my search enough or something, but I have tried several queries and come up empty or with unrelated results.
HashGrammer
@HashGrammer
Apparently it's called intellicode and all the ways that I found that used to add it to 2017 no longer exist. It's not in Extensions and Updates under the latest version of 2017 and the URL that used to lead to the marketplace for a VSX no longer exists. Strange.
Super bummer. Oh well.
uzfm
@uzfm
When will ML.NET use CUDA 11?
Eternity-8
@Eternity-8
Hi everyone! I started looking into DeepLearning_ImageClassification_Binary solution and I can't find a way to control the amount of images to classify. Is there a property in the code which is responsible for that?
2 replies
evo11x
@evo11x

When will ML.NET use CUDA 11?

Good question, I am also interested in that

Z3roCoder
@Z3roCoder
Hello! Sorry if this is a stupid question, but would you deem MLNet ready for production use? Considering also the SciSharp-Stack but many of the Libraries seem little maintained with many open, uncommented issues; It would be awesome to hear some opinions on this
1 reply
Arafat Tehsin
@arafattehsin
Hi @Z3roCoder - Not speaking on behalf of others but I have used ML.NET in production and I have seen a few folks using it.
MacNeacail
@MacNeacail
Michael: I still cannot work out how to get the centroids out of the latest ML model you showed me -- nothing pops up when I try all the different extensions -- a small hand would be appreciated
if I have two shapes in a picture - one Gaussian and one non-Gaussian -- what would be the best method to tell them apart if we have the raw data and there are millions of them
readonlyden
@readonlyden
Hello everyone! I'm trying to use ML.NET in my research about recommendation systems and I can't find a good article about building a content-based recommendation system using ML.NET. Can you help me on this? Is it possible to use Field-Aware approach from samples without user id to build a content-based filtering system?
SUDALV
@SUDALV92
I have a Data labeling project with 50 labeled pictures (its enough for test). I have exported table dataset from this data labeling project. Now i'm trying to create an Auto ML run, but it throws an error: "Number of usable training rows (0) (i.e. excluding rows with NaN or invalid target values) is less than total requested CV splits (10). Please reduce the number of splits requested, or increase the size of the input dataset with valid prediction targets."
alt
HashGrammer
@HashGrammer
Image Classification is receiving Received a label value of 3 which is outside the valid range of [0, 3) on my new sample set. Old sample set still works fine. Folder structure is identical and names of images should matter since the label is created by folder name. Not sure what I should look into first...
It's happening on my training pipeline method Fit(trainSet)
HashGrammer
@HashGrammer
Figured it out, the workspace had old files that were causing confusion. I emptied the workspace and re-ran the training and it went through just fine.
Sarosh Wadia
@saroshwadia
Any examples/best practices for using ML.NET to classify text into specific categories?
Bogdan Iulian
@IulianBogdan
Hey everyone! I ran into an issue that baffles me for days and I hope I can get some help here :) I've built a CNN in python using tensorflow and keras for a multiclass classification case. I managed to convert it to onnx and load it using MLContext.Transforms.ApplyOnnxModel. I am able to make predictions, but the results are completely different from what I get in in python. I used netron to inspect the graph and mapped the type (float32[unk, 86]) to a float[] in c#. But, for the same prediction, when I inspect the values in the array in python and c# they are completely different. For sure, I'm doing something wrong but I am new to the field and can't really figure it out. Anybody could shed some light on me, please?
Adnan Salahuddin
@AdnanSalah84
Hi folks,
I am looking for Word2Vec.
Is it available in ML.net?
Syed M. Sawaid
@SyedMSawaid
How can I use data from PostgreSQL in ML.NET Recommendation System?
1 reply
I have tried my best but still it sucks.
carrilloan
@carrilloan
Hi everyone, can you help me? can we load a Tensorflow lite model in ML.NET?
Gregor Beyerle
@WalternativE
If you can port it to ONNX you should be good to go. I wouldn't be aware of ML.NET being able to load TF lite out of the box but I've also never tried to do that.
Taras Novak
@RandomFractals
hi
I am new to ML.net. What's the best place to start and explore your ML stack?
Eric Erhardt
@eerhardt
@RandomFractals - check out https://dotnet.microsoft.com/apps/machinelearning-ai/ml-dotnet for the best place to start
1 reply
Taras Novak
@RandomFractals
@eerhardt have you considered moving to discord like .net interactive did? https://discord.com/channels/732297728826277939/760936842073210944
Shakerlicious
@Shakerlicious
Hi guys! Has anyone faced any issues in loading a faster-rcnn onnx model? I have this issue "InvalidProtocolBufferException: Protocol message had too many levels of nesting. May be malicious. Use CodedInputStream.SetRecursionLimit() to increase the depth limit." Did some search and found that the recursion limit in OnnxTransformer is capped at 100. Is there a way to increase it for larger models?
Michael Sharp
@michaelgsharp
@Shakerlicious are you using the Nuget package for ML.NET or are you building from source? The released version of ML.NET has a limit of 10. That was changed after the last release to 100 and so the next release will have that value. If you are building from source and still hitting that issue, it means we will need to create/modify the API so that you can specify that value if needed.
Eric Erhardt
@eerhardt

have you considered moving to discord like .net interactive did?

We are discussing it internally. If people here have feedback on the idea (for or against it), can you share it? If we don't hear anything, I assume that means people don't care if it is moved, so we should move it.

1 reply
Shakerlicious
@Shakerlicious
@michaelgsharp I managed to load the model. Built the OnnxTransformer from source and it worked! But yeah i agree that having an option to specify the recursion limit would be really useful!
Praveen Raghuvanshi
@praveenraghuvanshi
How do we do GroupBy on multiple columns in .net DataFrame, something similar to link
Prashanth Govindarajan
@pgovind
@praveenraghuvanshi : Unfortunately, the .NET DataFrame doesn't allow multiple GroupBys at the moment. You are limited to a single level GroupBy. If you really need the multi-level GroupBy, can you comment on dotnet/machinelearning#5758 with your scenario please?
Praveen Raghuvanshi
@praveenraghuvanshi
@pgovind Thanks for the response. Sure, I'll update the above mentioned github issue in couple of days.
Shakerlicious
@Shakerlicious
Hi again! I'm currently stuck with a problem where my Onnx model are giving incorrect outputs when ran on ML.NET as compared to in a python environment (using onnxruntime pip package). Full issue on dotnet/machinelearning#5842
Shakerlicious
@Shakerlicious

Hi again! I'm currently stuck with a problem where my Onnx model are giving incorrect outputs when ran on ML.NET as compared to in a python environment (using onnxruntime pip package). Full issue on dotnet/machinelearning#5842

Don't mind this. I found my mistake already.

Jessie Houghton
@houghj16
The .NET Team is conducting research into model explainability and fairness in ML.NET. We would love to hear from you. Please take this 10-minute survey.
alimhabidi
@alimhabidi
Hi @eerhardt
Crypties
@Crypties
@houghj16 or anyone else knowledgeable enough to answer: Could Open AI GPT2 be ported to C# using ML.NET? Any interest from any of you in doing that?
*with me .
qundeng97
@qundeng97
Hi there, I am new to ML.NET, am very impressed with ML.NET's model training performance. However compared to python Scikit-learn, the API is less flexible, e.g. no way to choose loss function in fast forest (always use mean squared error, while mean absolute error is desired) even though both are calculated in regression metrics as evaluation measures. Any work around for this? Also fast tree or fast forest does not provide individual tree level predict which handicap us from getting the predict distribution.
acrigney
@acrigney
Guys it would be really great if you could add more optimisers, say for regression I would like to use a Negative LogLikelihood loss function
acrigney
@acrigney
acrigney
@acrigney
Or I can use a NormalizeLogMeanVariance transformer