Motivation and Machine Learning - Part1
Been a long while here. Been up and running as usual, working my butt off to make a living and see how I can contribute to making the world a better place the little way I can. So a lot has happened since the last time I wrote here and i am grateful for everything.
I got selected for the Phase 1 of a Machine Learning challenge course sponsored by Microsoft in collaboration
Machine Learning is really awesome stuff that is poised to create a lot of opportunities in today's world while topping over a lot of traditional/manual processes. And I think you should pay attention it. That's what i have been doing. And I ave been thinking of a way to connect ML/AI to Motivating and inspiring the best they can be. How can ML/AI help us get people to be highly motivated and get their minds made up for the best?
That question will be answered soon. I am on the quest to also find out. To get everyone on same pane, I will be sharing the keypoints I have learnt from the challenge course modules here weekly. I am hoping that my summaries will help you get an idea of Machine Learning. And maybe you can figure out the answer to this question pretty soon. So I'll share a summary of 5 days:
Learnt about what is Machine learning again. It's a tool used to extract patterns from data and gives computers the ability to make predictions and identify patterns in data.
The difference between traditional Programming and machine learning is that while traditional takes in Data + Rules to give answers, Machine learning only takes in Data and Answers to gives us Rules.. How exciting.:blush:
ML is best suited for problems involving recognizing patterns, anomaly detection, time series forecasting and recommendation systems.
The revision on this helped the concept of ML stick deeper in my in my head. Wake me up tomorrow and ask me what ML is and I'll answer you without batting an eyelid.
lesson 2.3 - Applications of Machine Learning.
Covered 3 Techniques and 4 areas of applications:
Techniques - Statistical Machine Learning, Deep Learning and Reinforcement Learning
Application areas with examples - Natural Language Processing (Text and Speech translation and similarity), Computer Vision(Self Driving Cars, Object Identification/detection, LIDAR), analytics(regression, classification, clustering), Decision Making (Sequential Decision Making, Recommenders)
let me summarize the history of ML to you :blush:
1950s - 1970s - the focus was on building programs that copy the way humans think and act and machines outsmarting humans.
1980s-1990s : the focus shifted towards specific ML areas. ML algorithms were developed and neural networks.. but limitation on computing power.
2010s - explosion of computing power led to Advent of powerful GPUs that accelerated training of neural networks with lots of layers and nodes.
It's a day well spent.
2.5 - The Data Science Process and 2.6 Common Types of Data
Here's my summary:
First, Big data refers to the huge amount of data that is currently available and keeps growing as a result of an explosion of devices. The idea is to leverage this data to drive informed business decisions. Raw data refers to noisy and unreliable data which can produce misleading results if not worked up. It needs cleaning up.
Data science process:
• Collect Data: This is the step where you gather data relevant to your problem from different devices and sources.
•Prepare Data: Here you set up the data collected in a suitable format, and create or select features that you need for training. This is where manipulation, replacing missing values, aggregation is done.
•Training: This is the point where we try out different ML algorithms on prepared data. We split data into train, test and validation sets here and we proceed to evaluation.
• Evaluate: Here we evaluate our trained model, select the best algorithm that works best and keep checking model performance
•Deploy: At this point we are pretty satisfied with the evaluation results of our model, we package our model for use as a web service or API. Versioning and integration of training/evaluation scripts takes place here.
• Retrain: for our model to be sustainable and work well on new data, we need to retrain it on new data. This helps the model keep improving and adjusting to changes.
For the Common data types:
It's important we know this because the form and structure fed into an ML algorithm goes a long way to determine our approach, choice of algorithms and hyperparameters needed for our solution. We have 5 data types:
•Numerical: example Numbers, integers, floats
•Time Series: data collected over equally spaced period of time. Ex: real time stick data, forecasting data.
•Categorical: data values that show categories which do not clearly show a relationship unless a value is assigned to them. Eg: Ethnicity, Gender, Country
•Text: what you're reading right now:wink:
•Image data: Videos and Pics
So that's it. So far so good. Will try to update bi-weekly my learning.