A cheatsheet for numpy (python)

## Theory

```python
import numpy as np

np.array([1, 3, 5])
```

| command | purpose |
|---|---|
| `numpy.ones` | matrix of ones|

## References




A* search algorithm  (TypeScript implementation)

## Theory

$$
f(n) = g(n) + h(n)
$$

## Implementation

## References
- https://en.wikipedia.org/wiki/A*_search_algorithm   




Mozilla's DeepSpeech an open source speech to text engine

In this article, we will be trying a transcriber made using Mozilla's DeepSpeech.

## Installation

Let's start with an example.

```bash
# Create and activate a virtualenv
virtualenv -p python3 $HOME/tmp/deepspeech-venv/
source $HOME/tmp/deepspeech-venv/bin/activate

# Install DeepSpeech
pip3 install deepspeech

# Download pre-trained English model files
curl -LO https://github.com/mozilla/DeepSpeech/releases/download/v0.7.3/deepspeech-0.7.3-models.pbmm
curl -LO https://github.com/mozilla/DeepSpeech/releases/download/v0.7.3/deepspeech-0.7.3-models.scorer
```

## Demo - transcribing an audio file

Now let's transcribe an audio file.

```bash
# Download example audio files
curl -LO https://github.com/mozilla/DeepSpeech/releases/download/v0.7.3/audio-0.7.3.tar.gz
tar xvf audio-0.7.3.tar.gz

# Transcribe an audio file
deepspeech --model deepspeech-0.7.3-models.pbmm --scorer deepspeech-0.7.3-models.scorer --audio audio/2830-3980-0043.wav
```

The output should look similar to the verbose below. Notice the line
"experience proves this" which shows this is working.

```
Loading model from file deepspeech-0.7.3-models.pbmm
TensorFlow: v1.15.0-24-gceb46aa
DeepSpeech: v0.7.3-0-g8858494
Loaded model in 0.00997s.
Loading scorer from files deepspeech-0.7.3-models.scorer
Loaded scorer in 0.000207s.
Running inference.
experience proves this
Inference took 1.007s for 1.975s audio file.
```

The last line shows something note worthy - inference took less
time than the audio file length.

## Streaming to DeepSpeech

Make sure you have the [prerequisites](https://github.com/mozilla/DeepSpeech-examples/tree/r0.7/nodejs_mic_vad_streaming#prerequisites) before continuing.

```bash
# Prerequisites
wget https://github.com/mozilla/DeepSpeech/releases/download/v0.7.0/deepspeech-0.7.0-models.pbmm
wget https://github.com/mozilla/DeepSpeech/releases/download/v0.7.0/deepspeech-0.7.0-models.scorer

# Dependancies
sudo apt-get install libasound2-dev

# Clone and move into working directory
git clone https://github.com/mozilla/DeepSpeech-examples moz-ds-examples
cd moz-ds-examples/nodejs_mic_vad_streaming

# Install npm packages
npm install

# Start the server
node start.js
```

I had an issue where no verboise was given. This is because it uses the default
mic which may not exist.

So update `microphone` variable to the following - notice device property was
added to the [options](https://www.npmjs.com/package/mic#micoptions). Run
`arecord -l` to see the list of microphones installed.

```javascript
var microphone = mic({
  rate: "16000",
  channels: "1",
  debug: false,
  fileType: "wav",
  device: "plughw:2,0",
});
```

Now start streaming.

## Next steps

Mozilla's DeepSpeech has an API in C,.NET, Java, JavaScript (NodeJS/ElectronJS)
and Python so i'm sure you could integrate this the next time you need
speech-to-text - I know I will be. You may want to train your own models. You
could use this for an IoT device.


Getting started with GNU Octave (or MATLAB) for machine learning

82% of Fortune 100 companies use MATLAB; Octave allows you to prototype your
Machine Learning experiments faster, saving you time and money. Since it's a
high level programming language you can do things really fast. Some people
prefer to get their algorithms working in MATLAB and then migrate to another
language such as Python, R, Java or C++. So learning to use MATLAB is a good
investment.

> MathWorks provides a free online self-paced [MATLAB
> Course](https://matlabacademy.mathworks.com/). You also get a certificate which
> you can share to show others you have completed the course. The course is
> really well made.

## Installation

Since MATLAB has a price tag, we will use an open source alternative called
[Octave](https://octave.org). Octave is largely compatible with MATLAB - so you
can still follow along if your're a MATLAB user.

On Ubuntu 20.04

```bash
sudo apt update
sudo apt install octave
```

For other distributions, please consult [GNU Octave
download](https://www.gnu.org/software/octave/#install) website.

## Basics

Instead of replicating a tutorial, below are commands you will find helpful
when running your machine learning experiments.

| Command(s)                            | Usage                                                                                         |
| :------------------------------------ | :-------------------------------------------------------------------------------------------- |
| `+, -, \\, *`                         | Elementary math operations                                                                    |
| `==`                                  | Logical equal to                                                                              |
| `~=`                                  | Logical not equal to                                                                          |
| `&`                                   | Logical AND                                                                                   |
| `|`                                   | Logical OR                                                                                    |
| `xor(a, b)`                           | Logical XOR                                                                                   |
| `eye(n)`                              | n x n identity matrix                                                                         |
| `ones(n, m)`                          | n x m matrix of ones                                                                          |
| `rand(n, m)`                          | n x m matrix of randoms                                                                       |
| `randi(n, m)`                         | n x m matrix of random integers                                                               |
| `size(A)`                             | Dimensions of matrix $A$                                                                      |
| `clear A`                             | Remove variable $A$                                                                           |
| `clc`                                 | Clear command window                                                                          |
| `A'`                                  | Transpose of matrix $A$                                                                       |
| `A(1,2)`                              | Get element $A_{1,2}$                                                                         |
| `A(2,:)`                              | Get the 2nd row of $A$                                                                        |
| `A(end, :)`                           | Get the last row using the end keyword                                                        |
| `pinv(A)`                             | Pseudo inverse of matrix $A$                                                                  |
| `a = 3`                               | Variable assignment                                                                           |
| `a = 3;`                              | Variable assignment - semicolon suppresses output                                             |
| `pi`                                  | Constant PI                                                                                   |
| `A = [1 2; 3 4; 5 6]`                 | A 3x2 matrix with denoted elements                                                            |
| `V = [1; 2; 3]`                       | A 3x1 column vector i.e $V \in \mathbb{R}^3$                                                  |
| `1:4`                                 | Creates a matrix $\begin{bmatrix}1 & 2 & 3 & 4\end{bmatrix}$                                  |
| `0:2:4`                               | Creates a matrix $\begin{bmatrix}0 & 2 & 4\end{bmatrix}$ values from 0 to 4 with spacing of 2 |
| `linspace(first, last, num_elements)` | Useful when creating a range of values (no need to calculate spacing yourself)                |
| `help eye`                            | Shows man page for eye                                                                        |
| `doc eye`                             | Open doc page for eye                                                                         |
| `pwd`                                 | Print working directory                                                                       |
| `cd`                                  | Change directory                                                                              |
| `ls`                                  | List files                                                                                    |
| `load someData.dat`                   | Load data                                                                                     |
| `save filename variable`              | Save variables                                                                                |
| `save workspace.mat`                  | Save variables in workspace to a MAT-file                                                     |
| `load workspace.mat`                  | Load saved workspace variables                                                                |
| `format long`                         | Increase precision (default is short)                                                         |
| `who`                                 | Show variables in the current scope                                                           |
| `whos`                                | Show variables in the current scope with more info                                            |

Watch out: in Octave/MATLAB vectors are not zero-indexed i.e. they start at one

Note: All MATLAB variables are arrays. Understand the difference between
matrix, column vector, row vectror and scalar. Fun fact: MATLAB is an
abbreviation for MATrix LABoratory - makes sense.

Tip: If you only use one index with a matrix, it will traverse down each column
in order. Using one index, try extracting the eighth element of data.

## Manipulating data (matrices/vectors)

| Command       | Usage                                                                                                                                      |
| :------------ | :----------------------------------------------------------------------------------------------------------------------------------------- |
| `A * B`       | Matrix multiplication                                                                                                                      |
| `A .* B`      | Element-wise multiplication                                                                                                                |
| `abs(V)`      | Element-wise absolute                                                                                                                      |
| `exp(V)`      | Element-wise exponentiation                                                                                                                |
| `log(V)`      | Element-wise log                                                                                                                           |
| `max(A)`      | Maximum element in $A$                                                                                                                     |
| `round(A)`    | Round each element to the nearest integer                                                                                                  |
| `a < 2`       | Element-wise comparison                                                                                                                    |
| `find(A < 3)` | Elements that are less than 3                                                                                                              |
| `sum(A)`      | Sum of $A$                                                                                                                                 |
| `magic(n)`    | Generate a n x n magic matrix                                                                                                              |
| `A .* eye(n)` | Assuming $A$ is an n x n matrix then this yields the diagonal and all other elements are zero                                              |
| `flipud(A)`   | Flip matrix up-down                                                                                                                        |
| `A(A > 2)`    | Logical indexing. You can use a logical array as an array index, in which case MATLAB extracts the array elements where the index is true. |

## Plotting data

Plot the $\sin$ function.

```matlab
x = [0:0.01:0.98];
y = sin(2 * pi * 4 * x);

plot(x, y);
```

Plot the $\sin$ and $\cos$ function on the same graph.

```matlab
% Create an array from 0 to 0.98 in steps of 0.01
x = [0:0.01:0.98];

% Sine function
y_sin = sin(2 * pi * 4 * x); % notice the use of x

% Cosine function
y_cos = cos(2 * pi * 4 * x);

% Multple plots using 'hold on'
plot(x, y_sin);
hold on;
plot(x, y_cos, 'r'); % cos function in red

% Tip: use xlim([xmin xmax]) to zoom in on an area of interest

% Add title, axis labels and legend
title('Plot title')
xlabel('x-value')
ylabel('y-value')
legend('sin', 'cos')

% Save/export the plot as png
print -dpng 'plot.png'

% See plot gallery at https://uk.mathworks.com/products/matlab/plot-gallery.html
```

Use `figure` to create multiple figures. Tip: use `help plot` for more
information on how to use the `plot` command. Use `subplot` to plot side by
side.

Visualise a matrix.

```matlab
% Plot an 8x8 matrix with a color map (using command chaining)
imagesc(magic(8)), colorbar

% Close the plot
close
```

## Control statements

```matlab
% for loop
for i = 1:10
    % double each element of matrix A
    A(i) = 2 * A(i);
end

% while loop
i = 0;
while i <= 10
    i = i + 1;
end

% if, elseif, else
if i == 5
    disp('Found a 5');
elseif i == 4
    disp('Found a 4');
else
    disp('Keep searching');
end
```

## Functions

Functions are defined in files such as `addTwoNumbers.m`. Make sure your
function is in your octave search path or current working directory.

```matlab
% File: addTwoNumbers.m

function y = addTwoNumbers(a, b)
y = a + b;
```

Pro tip: Octave can natively return multiple values.

## Vectorisation (advance topic)

By using the vectorised implementation we can take advantage of matrix multiplication provided by the linear algebra library which will be faster than implementing a summation yourself (usually with a for loop). Below is an example.

$$
h_{\theta}(x) = \sum^n_{j=0} \theta_j x_j = \theta^T x
$$

where

$$
\theta =
\begin{bmatrix}
\theta_0 \\
\theta_1 \\
\vdots \\
\theta_n
\end{bmatrix},
x =
\begin{bmatrix}
x_0 \\
x_1 \\
\vdots \\
x_n
\end{bmatrix}
$$

It is easy to see how vectorisation works.

## Further reading

- [GNU Octave Documentation](https://octave.org/doc/v5.2.0/)


Getting started with linear regression using the ordinary least squares method

You often want to find parameters that minimise some cost function. One
approach could be using gradient descent. In this article, we will be talking
about another approach called the normal equation.

## Prerequisites

- Linear algebra basics
- Vector basics

## Getting started

$$
\theta = (X^T X)^{-1} X^T y
$$

Wow! One line. Also by using this method, you don't need to do feature scaling.

### Let's verify dimensions

## Octave implementation

```matlab
pinv(X'*X)*X'*y
```

## How it works? (optional)

Please feel free to skip this section. This is quite maths heavy. Here is the
outline of the mathematical proof. For brevity, we omit the full proof and only
provide an outline.

## Cautions

- Some of you may have noticed what if $X^T X$ is non invertible (singular) -
  this is the reason we use `pinv` as it always gives us a value for $\theta$
- Need to compute $(X^T X)^{-1}$ which is slow if $n$ is very large
  since most implementations have $O(n^3)$ time complexity ($nxn$ is the dimensions)
