Merge branch 'hotfix/hotfix-epsilonwarning'

.travis.yml

@@ -4,19 +4,29 @@ language: python
 
 cache: pip
 
-python:
-  - "2.7"
-  - "3.4"
-  - "3.5"
-  - "3.5-dev" # 3.5 development branch
-  - "3.6"
-  - "3.6-dev" # 3.6 development branch
+matrix:
+    include:
+        - python: 2.7
+        - python: 3.4
+        - python: 3.5
+        - python: 3.5-dev
+        - python: 3.6
+        - python: 3.6-dev
+        - python: 3.6
+        - python: 2.7
+          env: KERAS_BACKEND=theano THEANO_FLAGS=optimizer=fast_compile
+        - python: 3.6
+          env: KERAS_BACKEND=theano THEANO_FLAGS=optimizer=fast_compile
 
 # command to install dependencies
 install:
   - pip install --upgrade pip setuptools
   - cd python && pip install -r requirements-ci.txt
 
+  - if [[ "$KERAS_BACKEND" == "theano" ]]; then
+      pip install theano;
+    fi
+
 # command to run tests
 # (we are already in "python" directory here)
 script:

README.ko.md

@@ -30,7 +30,7 @@ WTTE-RNN은 이러한 문제가 어떻게 다루어져야 하는가에 대한 
 
 ![중도절단된 데이터](./readme_figs/data.gif)
 
-일종의 트릭으로, TTE 자체를 예측하는 대신 기계 학습 모델이 *확률 분포의 매개 변수* 를 출력하도록 해 봅시다. 어떤 확률 분포도 가능합니다만, 그 [굉장함](https://ragulpr.github.io/2016/12/22/WTTE-RNN-Hackless-churn-modeling/#embrace-the-Weibull-euphoria) 때문에 *Weibull 분포* 를 좋아하므로 사용했습니다. 기계 학습 알고리즘 또한 그래디언트 기반의 임의의 알고리즘이 될 수 있습니다만, 우리가 그 [굉장함](http://karpathy.github.io/2015/05/21/rnn-effectiveness/)때문에 RNN을 좋아하므로 RNN을 선택했습니다.
+일종의 트릭으로, TTE 자체를 예측하는 대신 기계 학습 모델이 *확률 분포의 매개 변수* 를 출력하도록 해 봅시다. 어떤 확률 분포도 가능합니다만, 그 [굉장함](https://ragulpr.github.io/2016/12/22/WTTE-RNN-Hackless-churn-modeling/#embrace-the-Weibull-euphoria) 을 좋아해서 *Weibull 분포* 를 사용했습니다. 기계 학습 알고리즘 또한 그래디언트 기반의 임의의 알고리즘이 될 수 있습니다만, 우리가 그 [굉장함](http://karpathy.github.io/2015/05/21/rnn-effectiveness/)을 좋아해서 RNN을 선택했습니다.
 
 
 ![WTTE-RNN 구조 예제](./readme_figs/fig_rnn_weibull.png)

README.md

@@ -12,7 +12,7 @@ WTTE-RNN is an algorithm and a philosophy about how this should be done.
 * [blog post](https://ragulpr.github.io/2016/12/22/WTTE-RNN-Hackless-churn-modeling/)
 * [master thesis](https://ragulpr.github.io/assets/draft_master_thesis_martinsson_egil_wtte_rnn_2016.pdf)
 * Quick visual intro to the [model](https://imgur.com/a/HX4KQ)
-* Jupyter notebook: [basics](examples/keras/simple_example.ipynb)
+* Jupyter notebooks: [basics](examples/keras/simple_example.ipynb), [more](https://github.com/ragulpr/wtte-rnn-examples)
 
 
 # Installation
@@ -93,28 +93,14 @@ Pull-requests, recommendations, comments and contributions very welcome.
 * Weibull functions (cdf, pdf, quantile, mean etc)
 * Objective functions written for:
   - Tensorflow
-  - Keras
+  - Keras (Tensorflow + Theano) ✔
 * Keras helpers
   - Weibull output layers
   - Loss functions
   - Callbacks
 * ~~ Lots of example-implementations ~~
     - [Basic notebook](examples/keras/simple_example.ipynb) will be kept here but to save space and encourage viz check out [the examples-repo](https://github.com/ragulpr/wtte-rnn-examples) or fork your notebooks there
 
-## Multi-framework support
-
-The core technology is the objective functions.
-We plan to make them to be reusable with any machine-learning framework.
-
-* TensorFlow ✔
-* Keras (TensorFlow wrapper) ✔
-* MXnet
-* Theano
-* Torch
-* h2o
-* scikitFlow
-* MLlib
-
 # Licensing
 
 * MIT license
@@ -131,7 +117,14 @@ We plan to make them to be reusable with any machine-learning framework.
 ```
 
 ## Contributing
+Contributions/PR/Comments etc are very welcome! Post an issue if you have any questions and feel free to reach out to egil.martinsson[at]gmail.com.
+
+### Contributors (by order of commit)
 
-Reach out to egil.martinsson[at]gmail.com if you have any questions.
-If you can, please open an issue so other can join in.
-Contributions/PR/Comments etc are very welcome!
+* Egil Martinsson
+* Clay Kim
+* Jannik Hoffjann
+* Daniel Klevebring
+* Jeongkyu Shin 
+* Joongi Kim 
+* Jonghyun Park

docs/conf.py

@@ -54,7 +54,7 @@
 
 # General information about the project.
 project = 'WTTE-RNN'
-copyright = '2016-2017. Egil Martinsson and contributors'
+copyright = '2016-2018. Egil Martinsson and contributors'
 author = 'Contributors.'
 
 # The version info for the project you're documenting, acts as replacement for

python/README.md

@@ -69,7 +69,14 @@ $ pip install -e .[build,test,dev,tf]
 ```
 
 ## Contributing
+Contributions/PR/Comments etc are very welcome! Post an issue if you have any questions and feel free to reach out to egil.martinsson[at]gmail.com.
 
-Reach out to egil.martinsson[at]gmail.com if you have any questions.
-Contributions/PR/Comments etc are very welcome!
+### Contributors (by order of commit)
 
+* Egil Martinsson
+* Clay Kim
+* Jannik Hoffjann
+* Daniel Klevebring
+* Jeongkyu Shin 
+* Joongi Kim 
+* Jonghyun Park

python/setup.py

@@ -30,7 +30,7 @@
 
 setup(
     name='wtte',
-    version='1.0.1',
+    version='1.0.2',
     description='Weibull Time To Event model. A Deep Learning model for churn- and failure prediction and everything else.',
     author='Egil Martinsson',
     author_email='[email protected]',

python/wtte/wtte.py

@@ -74,7 +74,7 @@ def output_lambda(x, init_alpha=1.0, max_beta_value=5.0,
 
     """
     if max_beta_value is None or max_beta_value > 3:
-        if K.epsilon() <= 1e-07 and K.backend() == 'tensorflow':
+        if K.epsilon() > 1e-07 and K.backend() == 'tensorflow':
             message = "\
             Using tensorflow backend and allowing high `max_beta_value` may lead to\n\
             gradient NaN during training unless `K.epsilon()` is small.\n\