Bastian Haase is an alum from the Insight Data Engineering program in Silicon Valley, now working as a Program Director at Insight Data Science for the Data Engineering and the DevOps Engineering programs. In this blog post, he shares his experiences on how to get started working on open source software.

One area that Deep Learning has not explored extensively is the uncertainty in estimates. Most Deep Learning frameworks currently focus on giving a best estimate as defined by a loss function. Occasionally something beyond a point estimate is required to make a decision. This is where a distribution would be useful. Bayesian statistics lends itself to this problem really well since a distribution over the dataset is inferred. However, Bayesian methods so far have been rather slow and would be expensive to apply to large datasets. As far as decision making goes, most people actually require quantiles as opposed to true uncertainty in an estimate. For instance when measuring a child´s weight for a given age, the weight of an individual will vary. What would be interesting is (for arguments sake) the 10th and 90th percentile. Note that the uncertainty is different to quantiles in that I could request for a confidence interval on the 90th quantile. This article will purely focus on inferring quantiles.

**50 Most Popular Python Projects in 2018**

1) TensorFlow Models

2) Keras

3) Flask

4) scikit-learn

5) Zulip

6) Django

7) Rebound

8) Google Images Download

9) YouTube-dl

10) System Design Primer

11) Mask R-CNN

12) Face Recognition

13) snallygaster

14) Ansible

15) Detectron

16) asciinema

17) HTTPie

18) You-Get

19) Sentry

20) Tornado

21) Magenta

22) ZeroNet

23) Gym

24) Pandas

25) Luigi

26) spaCy (by Explosion AI)

27) Theano

28) TFlearn

29) Kivy

30) Mailpile

31) Matplotlib

32) YAPF (by Google)

33) Cookiecutter

34) HTTP Prompt

35) speedtest-cli

36) Pattern

37) Gooey (Beta)

38) Wagtail CMS

39) Bottle

40) Prophet (by Facebook)

41) Falcon

42) Mopidy

43) Hug

44) SymPy

45) Dash

46) Visdom

47) LUMINOTH

48) Pygame

49) Requests

50) Statsmodels

2) Keras

3) Flask

4) scikit-learn

5) Zulip

6) Django

7) Rebound

8) Google Images Download

9) YouTube-dl

10) System Design Primer

11) Mask R-CNN

12) Face Recognition

13) snallygaster

14) Ansible

15) Detectron

16) asciinema

17) HTTPie

18) You-Get

19) Sentry

20) Tornado

21) Magenta

22) ZeroNet

23) Gym

24) Pandas

25) Luigi

26) spaCy (by Explosion AI)

27) Theano

28) TFlearn

29) Kivy

30) Mailpile

31) Matplotlib

32) YAPF (by Google)

33) Cookiecutter

34) HTTP Prompt

35) speedtest-cli

36) Pattern

37) Gooey (Beta)

38) Wagtail CMS

39) Bottle

40) Prophet (by Facebook)

41) Falcon

42) Mopidy

43) Hug

44) SymPy

45) Dash

46) Visdom

47) LUMINOTH

48) Pygame

49) Requests

50) Statsmodels

Below is a distilled collection of conversations, messages, and debates I´ve had with peers and students on how to optimize deep models. If you have tricks you´ve found impactful, please share them!!

**Overview and benchmark of traditional and deep learning models in text classification**

This article is an extension of a previous one I wrote when I was experimenting sentiment analysis on twitter data. Back in the time, I explored a simple model: a two-layer feed-forward neural network trained on keras. The input tweets were represented as document vectors resulting from a weighted average of the embeddings of the words composing the tweet. The embedding I used was a word2vec model I trained from scratch on the corpus using gensim. The task was a binary classification and I was able with this setting to achieve 79% accuracy. The goal of this post is to explore other NLP models trained on the same dataset and then benchmark their respective performance on a given test set. We’ll go through different models: from simple ones relying on a bag-of-word representation to a heavy machinery deploying convolutional/recurrent networks: We’ll see if we’ll score more than 79% accuracy!

**Marginal Effects for Regression Models in R**

Regression coefficients are typically presented as tables that are easy to understand. Sometimes, estimates are difficult to interpret. This is especially true for interaction or transformed terms (quadratic or cubic terms, polynomials, splines), in particular for more complex models. In such cases, coefficients are no longer interpretable in a direct way and marginal effects are far easier to understand. Specifically, the visualization of marginal effects makes it possible to intuitively get the idea of how predictors and outcome are associated, even for complex models.

**Predict Customer Churn with Gradient Boosting**

Customer churn is a key predictor of the long term success or failure of a business. But when it comes to all this data, what´s the best model to use This post shows that gradient boosting is the most accurate way of predicting customer attrition. I´ll show you how you can create your own data analysis using gradient boosting to identify and save those at risk customers!

Advertisements