Comments (1)
Since we are using log-likelihood as the evaluation metric, applying transformations to inter-event times (e.g. scaling / logartihm) will change the results. The log-likelihood is defined as \sum_i \log p^*(\tau_i)
, where p^*(\tau_i)
is the conditional density at point \tau_i
- we are summing log-densities for all the samples in the dataset. If we transform all the inter-event times \tau_i
(e.g. scale / apply log), the densities will also get changed according to the change of variables formula.
Instead, we do the following. All the models considered in our paper are defined as normalizing flows (i.e. a sequence of transformations of a base density).
z_0 ~ p(z_0)
z_1 = g_1(z_0)
...
z_M = g_M(z_{M-1})
tau = f(z_M)
As the final transformation (f
in the pseudocode above) we apply scaling / exponentiation (see decoders.py). This way
- we obtain a distribution over original inter-event times
- the models are easier to train, since the distribution over
z_M
should have 0 mean and unit variance.
Let me know if this explanation doesn't make sense to you and I will try to clarify it.
from ifl-tpp.
Related Issues (20)
- LogNorm curiosity HOT 4
- Code for using context vector in the models HOT 2
- on log likelihood misunderstanding HOT 4
- Loss with NLL of mark and MAE of inter-event time HOT 6
- history HOT 5
- Hyperparameters for reproducibility HOT 6
- Sampling points of a specific mark HOT 3
- Implementation on missing data imputation HOT 1
- Understanding given datasets HOT 5
- NLL results HOT 5
- Sampling with additional conditional information
- use other dataset HOT 1
- Missing data imputation HOT 1
- Calculate the mean of the entire distribution. HOT 7
- all evaluation expriments code HOT 1
- Calculate the mean of the entire distribution. HOT 23
- How could I get the predicted results? HOT 20
- ATM dataset testing HOT 2
- Learning with Marks HOT 9
- How to get the expression of the distribution of inter-event time HOT 6
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from ifl-tpp.