jonasrothfuss / meta_learning_pacoh Goto Github PK
View Code? Open in Web Editor NEWMeta-learning Gaussian process (GP) priors via PAC-Bayes bounds
License: MIT License
Meta-learning Gaussian process (GP) priors via PAC-Bayes bounds
License: MIT License
Hi,
I am running your code to compare Test RMSE on meta-training and meta-testing tasks as a function of the number of meta-training tasks for PACOH and MLAP.
As you wrote in your code, I run:
meta_train_data, meta_test_data, _ = provide_data(dataset='sin_20')
NN_LAYERS = (32, 32, 32, 32)
gp_model_mlap = GPRegressionMetaLearnedPAC(meta_train_data, num_iter_fit=20000, task_kl_weight=1.0,
meta_kl_weight=1e-5, lr=1e-3, lr_decay=0.97, posterior_lr_multiplier=5.0,
svi_batch_size=5, task_batch_size=5,
covar_module='NN', mean_module='NN', mean_nn_layers=NN_LAYERS,
kernel_nn_layers=NN_LAYERS, cov_type='diag', normalize_data=True)
gp_model_mlap.meta_fit(valid_tuples=meta_test_data[:10], log_period=500, eval_period=10000, n_iter=10000)
The last line gives me the following error:
PACOH\meta_learn\models.py in set_parameters_as_vector(self, value)
288 # self.set_parameter(name, value[:, idx:idx_next])
289 else:
--> 290 raise AssertionError
Could you please help me to fix this problem?
In fact, for the given parameters, we have value.ndim=3. However, it should be either one or two.
Best regards
AR
Hi, I have a metalearning problem where all inputs and outputs are 2D images. I'd like to use the wonderful deep GP package you have developed. Based on the MNIST 2D example, it seems your code converts 2D images to 1D vector and then sample pixels as contexts. In my problem, I'd like to use past image frames as contexts. In other words, I'd like to exploit autocorrelation in the target variable. Is this possible? If so, how about scalability, my image is typically 360x180. I've done this using maml and CNN, but the results are not satisfactory.
Hi @jonasrothfuss @fortuin !!
I was wondering how I use the pac regression with a basic uni-variate example.
Specifically with the below dataset.
import numpy as np
import yfinance as yf
data = yf.download("SPY", start="2012-01-01", end="2017-04-30")['Adj Close']
y=data
X=np.linspace(1, len(data))
Would i be able to use the meta learning on the above data set or would this be a little over the top.
Kind regards and thanks,
Andrew
Hello, thank you for open sourcing the repo! May I know where to obtain the SwissFEL, physionet etc. data from?
I'm not sure where to download the h5 files from, and I get the following missing file errors when calling provide_data(dataset, DATA_SEED)
:
OSError: ``...meta_learning_pacoh/data/physionet2012/set_a_merged.h5`` does not exist
or meta_learning_pacoh/data/swissfel/2018_10_31/neldermead/data/evaluations.hdf5
does not exist.
Thank you very much! I tried also looking at the previous SwissFEL paper by Kirschner but could not find any download sources: https://arxiv.org/pdf/1902.03229.pdf
A declarative, efficient, and flexible JavaScript library for building user interfaces.
๐ Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. ๐๐๐
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google โค๏ธ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.