A time series (TS) is a series of data points indexed (or listed or graphed) in time order, with equally spaced time intervals.

Let me tell you a little about them, in the following thread 🧵

1/9

Currently, time series analysis is applied in many fields, such as:

🔹Economic forecasting
🔹Inventory analysis
🔹Stock price forecasting
🔹Medical supplies forecasting
🔹Census analysis

2/9

The 4 main components of a time series are:

▪️Level: the average value in the series if it were a straight line.
▪️Trend: the linear increasing or decreasing behavior of the series over time.

3/9

and,

▪️Seasonality: the repeating patterns or cycles of behavior over time.
▪️Noise: the random variation in the series that cannot be explained by the model.

4/9

A time series can be decomposed into its components using the following 2 methods:

▪️Additive decomposition: Level + Trend + Seasonality + Noise

▪️Multiplicative decomposition: Level * Trend * Seasonality * Noise

5/9

Google Research is proud to help give voices back to those battling ALS, like @SteveGleason, through the use of machine learning and speech recognition/synthesis.

You can learn more by watching the video below and also learn more about Project Euphonia at https://t.co/6gif7UqFtV

Graph Neural networks tutorial in Google Colab https://t.co/RwN1le0C7A

#AI #DeepLearning #MachineLearning #DataScience

Some pretty amazing slides and notes for this Data Structures course.

https://t.co/JF7FSFKFT6

20 more life lessons a father must teach his son

=THREAD=

1. Your freedom begins the day you realize nobody is coming to save you

Your life is completely in your own hands. And that’s what makes life beautiful. Internalize the fact that your success is your responsibility alone. Because independence is freedom, and freedom is power.

2. Self-discipline unlocks the world

To master the world, master yourself. Success requires consistent effort. And consistent effort is impossible without self-discipline. Discipline is your hedge against failed motivation. Do the hard thing, especially when you don't want to.

3. Leaders don’t ask for permission

Leadership isn’t negotiated. A leader who begs for compliance is a leader in name only. Assume responsibility. Do what must be done. Do it competently. People will follow your vision because you attract them. Not because you chase them.

4. Be wary of the herd

In nature, the herd is usually running from danger. In humans, the herd is usually running toward it. If you never depart from the herd, you will always be following someone else’s path. Leadership means having the courage to change direction.

DiffSVC: A Diffusion Probabilistic Model for Singing Voice Conversion
pdf: https://t.co/945wanF1up
abs: https://t.co/299tKHAzzV

project page:

Tensorflow implementation of Involution
https://t.co/UM0KCrDOXz

The main concept is to symmetrically switch the inherent properties of convolution namely spatial-agnosticism and channel-specificity.

In the notebook, one will find an involution `tf.keras.layers.Layer`. It can be picked up and used in other architectures too.

The paper:

video on Involution:

How many times you have heard that your residuals should be independent, normally distributed with mean zero, and with constant variance?

No matter if your answer is zero or a million times. This thread can help you

It doesn't even matter that you don't know about residuals👇

First of all: What is a residual?

It is a synonym for error. It is the difference between the expected output and the output of our model:

Y - Ym

Where Ym is the output of our model. So residuals can be positive or negative and we need them to stay close to zero

1: Independence

We don't want the error in for some input to be dependent on the error for another input

That would mean there is information about the relationship between inputs and outputs that our model is missing and that is present in the residuals2

2: Mean equals zero

Well, this is an intuitive one, isn't it?

We have said already that we need residuals close to zero. There should be both negative and positive residuals.

But we don't want a single huge positive error and then a lot of small negative ones

Let's continue

3: Normally distributed

The mean of the residuals is zero, but that's not enough

We need about 50% of the errors to be negative and the other half to be positive

Also, we need most of the errors to be close to zero

In other words, we need the errors to be normally distributed

✨ Space and Time Complexity ✨

One of the most important concepts when studying DSA is the complexity of the algorithm

In this thread I will dive deep into what that means and how you can measure it 👇🏻

When learning algorithms you can measure how efficient the algorithm is by knowing how many lines of code it executes to run an algorithm for an input of size n.

However, it is very hard to measure exactly how many lines of code the algorithm takes.

This is why big O exists.

what is big O? 🤔

Big O is the most used asymptotic notation. Asymptotic notations are a way to approximate how many lines of code the algorithm executes relative to the input size which we will refer to as "n" for the rest of the thread

To understand Big O better I will explain it through an algorithm: Selection Sort

This algorithm sorts an array by looping through the whole array and for each element in the array it iterates through the rest of the array to find the smallest element to put it in the beginning

This image visualizes how selection sort works:

@paperswithcode @huggingface Ready for some NLP on NLP paper action?
checkout the awesome 👻QASPER dataset by @allen_ai
https://t.co/315lv8AtlO nicely linked to the @huggingface dataset loader I had just added 🤯

Processing images digitally with Python is not only a fun task, but it's also very easy to do.

In the following thread🧵, let me explain how you can do it from scratch using simple operations on a matrix.

1/7

First of all, as a context, PNG (Portable Network Graphics) images can naturally be represented by a 2D matrix of pixels.

At the same time, the pixels are made up of 3 RGB components, representing the colors Red, Green Blue.

2/7

Therefore, after converting a PNG image into a matrix, 2 operations that can be performed very easily are:

▪️Flip vertically: invert the order of the columns of the matrix

▪️Rotate 90 degrees: invert the rows by columns, traversing the matrix inversely

3/7

In addition, the illumination of the image can be modified:

▪️Apply brightness: add a constant value to all pixels of the image (e.g. alpha 0.1)

▪️Apply contrast: subtract a constant value to all pixels of the image

4/7

You can also calculate the negative of the image. To do this, subtract 1 from each color (maximum value for each component) and save the absolute value of this subtraction.

The image will look like this (very cool, right?):

5/7

Getting lean 101.

THREAD ⬇️

1) Calorie deficit.

Obviously this is a no brainer and the most important out of all these tips.

A 250-1000 calorie deficit per day is a good range to be in depending on how much fat you've actually got to lose.

A good rule of thumb is to shoot for a loss of 0.5-1% of body weight per week.

You'll be on the higher end of this if you're on the higher end of the body fat percentage scale.

2) Lift weights 4-5x per week.

Lifting is the signal to keep that hard earned muscle mass hanging around when you're in a deficit.

You've still got to train as if you're trying to build muscle.

Train close to failure by taking sets to within 0-3 RIR (Reps in reserve)

Don't push overall volume too high though, as you haven't got the same raw materials for recovery.

Find a split/frequency that allows you to get your total weekly volume in and get to work.

Training is not about what you can do, but what you can recover from.

Choosing the right machine learning model for a given problem is a no-free lunch scenario - There is no model that is guaranteed to solve a problem before you try and evaluate different types of models.

Here are more things that you can consider while choosing a model: 👇

🔹The scope of the problem
🔹The size and the type of the dataset
🔹The level of interpretability in need
🔹And the available training resources

To learn more:

I'll blow your mind with a technique you aren't using yet.

Sometimes, you want your system to do exactly the opposite of what your machine learning model thinks you should do.

Let me convince you. ↓

I'm going to start with a nice problem:

Imagine a model that looks at a picture of an electrical transformer and predicts whether it's about to break or not.

Don't worry about how the model does this. We are going to focus on the results instead.

There are 4 possible results for this model:

1. It predicts a bad unit as bad.
2. It predicts a bad unit as good.
3. It predicts a good unit as bad.
4. It predicts a good unit as good.

#2 and #3 are the mistakes the model makes.

Assuming we run 100 units through the model, we can organize the results in a matrix:

• The rows represent the "actual" condition of the transformer.

• The columns represent the "prediction" of the model.

We call this a "Confusion Matrix."

This is how we can read this confusion matrix:

• 60 bad units were predicted as bad.
• 3 bad units were predicted as good.
• 7 good units were predicted as bad.
• 30 good units were predicted as good.

Why are graphs the future of biomedical research and what is the value of NLP here?

A small case study about:

How to speed up drug discovery with knowledge graphs and discover potential cures for diseases

In this case text mining is used to contextualize knowledge about:

- Genes
- Compounds
- Diseases
- Adverse drug effects
- Receptor bindings

Which text types are processed here? Medical literature, patient notes, electronic health records, clinical reports etc.

But how to start?

First you need to identify the different entities such as compounds, diseases, adverse drug effects and receptor bindings.

This is achieved through Natural Language Processing (NLP) and there are suitable pre-trained models for processing biomedical, scientific or clinical text like scispaCy

@spacy_io models for processing biomedical, scientific or clinical

Another library which is specialized in biomedical text is Spark NLP

@JohnSnowLabs