hydrogo/rainnet

RainNet: a convolutional neural network for radar-based precipitation nowcasting

rainnet's Issues

I want to know how to visualize the trained model, is there any way?

I want to know how to visualize the trained model, is there any way?

model.predict(X) -- runs the model with specified input data X to obtain the corresponding nowcast.

can you execute the sample

issues defining input size/shape

Hello,

I don't understand how you define the input size in rainnet. Actually when we look at the RYDL data, they are 900 x 900 arrays, but you set the argument input_shape=(928, 928, 4) so I don't understand.
Besides I try to make forecasts from other rain maps of size 224x224 and I have errors when I execute model.fit

Can you help me to understand how to define the size of the data or tell me what elements to modify in the algorithm when changing the size of the data please?

Thank you

no check for consecutive times lots when training

I've noticed that the get_item method in the dataset class does not proof if the 5 np.arrays used for training are actually 5 consecutive timeslots. I'm referring to this line:

arr = np.array(self.dataset.get(self.keys[i+j])) and
y = np.squeeze(processor.data_preprocessing(np.array(self.dataset[self.keys[i+4]])[np.newaxis,:,:])) #corrected after github issue

E.g. the first 3 arrays can be the 01.01.2020 5:00, 01.01.2020 5:05, 01.01.2020 5:10 but the next is 01.01.2020 5:20 and should not be used.

Training RainNet from scratch

Hi Georgy,

I am a student at Indian Institute of tropical meteorology, Pune, India. I found your Rainnet project very interesting and wanted to do similar work by using radar datasets from our project. As a beginner in machine learning, I am struggling to train the rainnet model from scratch . I want to use our archived radar datasets for this work. Is it possible to share the code that relates to the training of the model?
It will be helpful for me to train the rainnet model. Also, training code can be helpful for me to crate images as per model requirements and to pre-process them. So that can be used as input to the model for training.

Looking forward for your positive reply very soon.

Thanks with best regards,

Prasad

Nowcast array maximal value

Dear Georgy,
I've been playing recently with provided example and found out, that I always end with nwcst prediction array
that has maximal value around 1.219. No matter what RY_latest maximum precipitation was provided.
Is tere something that I'm missing?
Some kind of 'descaling' or 'denormalization' possibly?

Access to the image names txt file

Hi,
As I click on the link
"The image names are present in the keys txt file which can be downloaded from here."

It shows the data is not existing. Could you provide how to access the file with image names?

Thank you in advance!

Regarding RYproduct dataset

Can you please explain a bit about how did you use the RYproduct dataset which you have mentioned in the readme? The zenodo link in the readme points to an hdf5 file which just contains the images. I would be grateful if could tell me how did you divided those images for training , into inputs and targets .
Thanks.

Training Rainnet

Hi Georgy, I am a student at Stanford taking course CS231 on CNNs. Found your Rainnet project very interesting and decided to replicate it as part of my course project.
I have trouble making it train from scratch, using my data (initially moving MNIST). Is it possible to share the remaining code that relates to the training of the model?
Many thanks, Georgios.
gs7776 at stanford.edu

how to get the weights once the model is trained

Hello,
first of all, thank you for sharing your work
I try to use the jupyter notebook developed with @PradyumnaGupta to train the RainNet model, once the model is trained with model.fit can you tell me how to get the files with weights (like rainnet_weights.h5) used for prediction, please ?

Thank you

Could you please tell me how to decompress the RYDL.hdf5 file to get the picture and put it into the model for prediction?

label and color

Your data contains many labels and how is it divided by color?

Datasheet

Sir ..I was unable to view the datasheet ..it showing that the file can't open..which software I need to open the file sir

Increment in the i in the Dataset class

In the next line in the Dataset class:

y = np.squeeze(data_preprocessing(np.array(self.dataset[self.keys[i+3]])[np.newaxis,:,:]))

I wonder if if instead of i+3 it wouldn't be i+4. Because, if I understand it correctly, RainNet takes the last 4 timesteps and predicts the next one (the 5th). So, with i+3 we are still in one of the data points in the x array, that runs in range(4), so from 0 to 3. Am I right or am I misunderstanding something? Thanks, @PradyumnaGupta

Problems Decoding radar data from radarlan/ry

Hello I am a Grad student trying to run your example code for rainnet. I have installed all the packages as instructed, but when I run the code it reaches this line and this exception is thrown out:
code:
data, attr = wrl.io.read_radolan_composite( io.BytesIO( data_binary ), missing=0 )
exception:
'utf-8' codec can't decode byte 0x82 in position 0: invalid start byte
How could I resolve this? Thanks for the help.

Syntax error in wradlib install during running the rainnet code in colab