Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Activity
Haiping
@Oceania2018
As I know the latest version fixed the issue.
uzfm
@uzfm
image.png
image.png
Haiping
@Oceania2018
Try to upgrade ImageClassfication to v0.4.0
uzfm
@uzfm
Update 0.4.0 did not help
"task[ID]" is a new model.
I want to make simultaneous predictions on two different models.
Haiping
@Oceania2018
Can you raise a issue in our issue page?
uzfm
@uzfm
can you give a link please?
rdthree
@rdthree
Hi, I have a question on NumSharp. I'm trying to get started and learn the API a bit. The DocFX documents search feature doesn't seem to work though.
so as far as a guide of sorts, this is all I've found: https://3rdman.de/2020/08/numsharp-cheat-sheet/
is there anything else out there in terms of examples, guides, tutorials?
rdthree
@rdthree
(sorry mean to say that the search doesn't go as deep as I'd like)
rdthree
@rdthree
hi again, just following on my own question here. it looks like much of the docs Numpy.NET, also applies to NumSharp. For example, the suggestion to reshape a 1D array, and then just trial and error with numpy methods.

for example
input

using NumSharp;
var m = np.array(new double[]{1,2,3,4,5}).reshape(5,1);
m.ToString().Display();

var n = np.array(new double[,]{{1},{2},{3},{4},{5}});
n.ToString().Display();

output

[[1], 
[2], 
[3], 
[4], 
[5]]

[[1], 
[2], 
[3], 
[4], 
[5]]
rdthree
@rdthree
another one...
//this only works with a [,] array, a [] array doesn't have axes to swap
//benefit here is not having to type the size of elements in the reshape
var o = np.array(new double[,]{{1,2,3,4,5}}).swapaxes(0,1);
o.ToString().Display();
[[1], 
[2], 
[3], 
[4], 
[5]]
Vincent B.
@evilz
Hello, I'm a big noob on this lib.
I'm looking for somethig to do the same that
form sklearn.datasets import make_blobs
X, y make_blobs(n_samples=100, n_features=2, centers=2, random_state=0)
napgab
@napgab

Hi, i am trying to translate this model to sci-sharp:

model = Sequential()
model.add(Dense(512, input_shape=(1000,), activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(512, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(num_classes))
model.add(Activation('softmax'))

Anyone can help me to find how to add the last activation layer? Cannot find it under keras.layers.

Larry Jones
@mrwizard82d1
Anyone interested in / thinking about porting PyMC?
Chad Williamson
@yellowfever13
Hello. Using SciSharp.TensorFlow.Redist-Windows-GPU (2.7.0) - is this compatible with CUDA 11.6? Not sure how to test if TF is using GPU, just getting started.
Chad Williamson
@yellowfever13
Interwebs say to use tf.test.xxxx to see GPU support, but the test namespace does not exist
Sean Hankins
@scarabdesign
I'm trying to convert this python script to Numpy.NET, but it seems to be giving me mass arrays of NaNs. Has anyone attempted this? I can post a version of my code if asked:
https://makersportal.com/blog/2018/11/25/goes-r-satellite-latitude-and-longitude-grid-projection-algorithm
Sean Hankins
@scarabdesign

Hey, I got through it! There seems to be an issue, however with the data I'm digging through. I was able to install an cli app that converts .nc files to JSON. Looking at the JSON dump, I see that the scale_factor and add_offset variables are listed as floats, but the number is too long for C# to parse without truncating:

                {
                    "name": "scale_factor",
                    "shape": {
                        "class": "H5S_SIMPLE",
                        "dims": [
                            1
                        ]
                    },
                    "type": {
                        "base": "H5T_IEEE_F32LE",
                        "class": "H5T_FLOAT"
                    },
                    "value": [
                        0.054936669766902924
                    ]
                },

Here's what it looks like from Microsoft.Research.Science.Data object:
ds["y"].Metadata["scale_factor"] = 0.05493667
So, I'm wondering what my options are. I hate to just rely on the JSON dump because it makes the original file at 500k to a 500 meg json file

Sean Hankins
@scarabdesign
Then again, doing a regular ncdump -h <filename> is showing the shorter value. Is it possible that the json dump added extra numbers?? I think it was written in C:
https://github.com/jllodra/ncdump-json
Sean Hankins
@scarabdesign
Also tried this json dump lib and it showed the same results
https://github.com/HDFGroup/hdf5-json
but in this example,I cloud't find how to predict my model
I tried var tensor = tf.image.resize_images_v2(tf.image.decode_image(tf.io.read_file(test),3,TF_DataType.TF_FLOAT), img_dim)/255;
    var tensors = model.predict(tensor,batch_size);
but get a error about InvalidArgumentError : input must be 4-dimensional
var tensor = tf.image.resize_images_v2(tf.image.decode_image(tf.io.read_file(test),3,TF_DataType.TF_FLOAT), img_dim)/255;
var tensors = model.predict(tensor,batch_size); //  InvalidArgumentError : input must be 4-dimensional
SebastienES
@SebastienES
Hello, I don't manage to use 2 functions in TensorFLOW.net, that seems to be missing.
tf.tensor_scatter_nd_update
or
tf. scatter_nd
That is used to update 1 or several entry in a Tensor.
Or is there another way to do that ?
I am doing some RL, and I have a Tensor QValues , 2D matrix, and in Q-Learning I need to update 1 entry QValues[state][action] = QValues[state][action] + learning_rate * error.
I managed to do the code for the rest, but I don't find how to do this in TensorFLOW.net.
I used Visual Studio 2022 and installed TensorFLOW.net 0.7 nuget and CPU Redist 2.7
bbhxwl
@bbhxwl
qq 群被封了么?今天刚注意。
Alizer
@AlizerUncaged
how do you load a tensorflow model from tensorflow hub? i have this model for example https://tfhub.dev/tensorflow/centernet/resnet50v1_fpn_512x512/1
AllenP
@longdrive01923
Do any of the SciSharp libraries implement the SciPy routines signal.fftconvolve, scipy.ndimage.zoom and scipy.ndimage.rotate?
小麦兜
@wangjunjx8868
@Oceania2018 新的群还有吗