1597068900

The article aims to outline the concept of the Generalized Mixture Transition Distribution (MTDg) Model and introduce the `mtd-learn`

Python package developed by me. You can find a broader introduction to the models here (it’s not fully featured in this post because Medium does not support mathematical notation) and the package repository here.

The GeneralizedMixture Transition Distribution (MTDg) model was proposed by Raftery in 1985[1]. Its initial intent was to approximate high-order Markov Chains (MC), but it can serve as an independent model too. The main advantage of the MTDg model is that its number of independent parameters grows linearly with the order in contrast to the exponential growth of Markov Chains models.

The MTDg model is a sequence of random variables *(Xn)* such that:

where *it…i0 ∈ N*, *lambda* is a weight vector and *Q(g)* is an *m ⨯ m* matrix representing the association between g-thlag and the current state.

Following conditions has to be met for the model to produce probabilities:

The log-likelihood function of the MTDg model is given by:

where *n* means the number of transitions in a dataset.

You can think about the MTDg model as a weighted average of transition probabilities from subsequent lags. The example below shows how to calculate a probability of transition B->C->A->B from an order 3 MTDg model:

According to [1] the number of independent parameters of the MTDg model equals *lm(m-1) + l — 1*. In [2] Lebre and Bourguignon proved that the true number of independent parameters equals *(ml — m + 1)(l — 1)*. Since the `mtd-learn`

package uses the estimation method proposed in [2] the number of parameters is calculated with the latest formula.

#statistics #probability #machine-learning #mtd #markov-chains #deep learning

1617331277

The Association of Data Scientists (AdaSci), the premier global professional body of data science and ML practitioners, has announced a hands-on workshop on deep learning model deployment on February 6, Saturday.

Over the last few years, the applications of deep learning models have increased exponentially, with use cases ranging from automated driving, fraud detection, healthcare, voice assistants, machine translation and text generation.

Typically, when data scientists start machine learning model development, they mostly focus on the algorithms to use, feature engineering process, and hyperparameters to make the model more accurate. However, model deployment is the most critical step in the machine learning pipeline. As a matter of fact, models can only be beneficial to a business if deployed and managed correctly. Model deployment or management is probably the most under discussed topic.

In this workshop, the attendees get to learn about ML lifecycle, from gathering data to the deployment of models. Researchers and data scientists can build a pipeline to log and deploy machine learning models. Alongside, they will be able to learn about the challenges associated with machine learning models in production and handling different toolkits to track and monitor these models once deployed.

#hands on deep learning #machine learning model deployment #machine learning models #model deployment #model deployment workshop

1583299685

At **SqlDBM** BI database modeling tool help organizations to improve their decision and Analyze billions of records in seconds. Currently " Data Warehouse” is currently trending topic in the data area. We will covering what a Data Warehouse is and how it is created from a SQL script. Visit us to get know more about BI modeling Tools and how it work with SQL.

#export data model #SQL Server BI Modeling #BI modeling Tools #SQL Server Business Intelligence Modeling Tool

1624298520

In a series of weekly articles, I will be covering some important topics of statistics with a twist.

The goal is to use Python to help us get intuition on complex concepts, empirically test theoretical proofs, or build algorithms from scratch. In this series, you will find articles covering topics such as random variables, sampling distributions, confidence intervals, significance tests, and more.

At the end of each article, you can find exercises to test your knowledge. The solutions will be shared in the article of the following week.

Articles published so far:

- Bernoulli and Binomial Random Variables with Python
- From Binomial to Geometric and Poisson Random Variables with Python
- Sampling Distributions with Python

As usual, the code is available on my GitHub.

#statistics #distribution #python #machine-learning #sampling distributions with python #sampling distributions

1623263280

This blog is an abridged version of the talk that I gave at the Apache Ignite community meetup. You can download the slides that I presented at the meetup here. In the talk, I explain how data in Apache Ignite is distributed.

Inevitably, the evolution of a system that requires data storage and processing reaches a threshold. Either too much data is accumulated, so the data simply does not fit into the storage device, or the load increases so rapidly that a single server cannot manage the number of queries. Both scenarios happen frequently.

Usually, in such situations, two solutions come in handy—sharding the data storage or migrating to a distributed database. The solutions have features in common. The most frequently used feature uses a set of nodes to manage data. Throughout this post, I will refer to the set of nodes as “topology.”

The problem of data distribution among the nodes of the topology can be described in regard to the set of requirements that the distribution must comply with:

- Algorithm. The algorithm allows the topology nodes and front-end applications to discover unambiguously on which node or nodes an object (or key) is located.
- Distribution uniformity. The more uniform the data distribution is among the nodes, the more uniform the workloads on the nodes is. Here, I assume that the nodes have approximately equal resources.
- Minimal disruption. If the topology is changed because of a node failure, the changes in distribution should affect only the data that is on the failed node. It should also be noted that, if a node is added to the topology, no data swap should occur among the nodes that are already present in the topology.

#tutorial #big data #distributed systems #apache ignite #distributed storage #data distribution #consistent hashing

1597773960

NLP Models have shown tremendous advancements in syntactic, semantic and linguistic knowledge for downstream tasks. However, that raises an interesting research question — is it possible for them to go beyond pattern recognition and apply common sense for word-sense disambiguation?

Thus, to identify if BERT, a large pre-trained NLP model developed by Google, can solve common sense tasks, researchers took a closer look. The researchers from Westlake University and Fudan University, in collaboration with Microsoft Research Asia, discovered how the model computes the structured, common sense knowledge for downstream NLP tasks.

According to the researchers, it has been a long-standing debate as to whether pre-trained language models can solve tasks leveraging only a few shallow clues and their common sense of knowledge. To figure that out, researchers used a CommonsenseQA dataset for BERT to solve multiple-choice problems.

#opinions #ai common sense #bert #bert model #common sense #nlp model #nlp models