Data Exploratory in ML


Data Exploratory

 Single data point or single axis 

1. Mean - influence by extreme value but good part it consider all of them equally

2. Median - not influence by extreme value however distribution is not consider it doesn't matter what is at extreme and start

3. Percentile - good to show data but multiple measure get involve 

4. Standard deviation - consider all of them and distribution but hard to calculate 


bivariant/ two data point / two axis

1. Correlation - positive support each other, negative opposite of each other, no correlation doesn't impact 


Garbage in garbage out 

Overfit or underfit 

Insufficient data - simple model or data augmentation , synthetic data , use ensemble learning 

Transfer learning - similar model which is already available pre trained 

Data augmentation - change some data like change something like scaling , rotate image 

Synthetic data - artificial generated data 


Curse of dimensionality - too much column 

Outdated date - too much rows 

Concept drift - due to outdated data . As things change with time 











Comments

Post a Comment

Popular posts from this blog

How to Run Anaconda3 Code from PowerShell ?

 While executing PowerShell script if you get  + FullyQualifiedErrorId : UnauthorizedAccess error.  Please follow below steps. 1. Execute >Get-ExecutionPolicy -Scope CurrentUser  If this returns Undefined then we have to give access to current user to execute PowerShell command.  2. Execute >Set-ExecutionPolicy -ExecutionPolicy RemoteSigned -Scope CurrentUser 3. Now you should be able to run Powershell Command.  4. Once you have done with work then again revoke access as best practice.  5. Execute >Set-ExecutionPolicy -ExecutionPolicy Undefined -Scope CurrentUser Activate Virtual Environment using below command. 
Read more

What is Flask?

Flask is web application framework in python. It's built on the Werkzeug WSGI toolkit and Jinja2 template engine.  Web server gateway interface (WSGI) helps in requests, response object and other utility functions. Jinja2 : templating engine helps in rendering dynamic web pages.  Pip install Flask  ML model deployment using flask Step 1 : Pickling the ML model  import pickle  pickle.dump(model1, open('myfirst_model.pkl','wb')) Pickle module help in serializing ML model and save to a file.  Later we can load model from file and deserialize to make predictions. model = pickle.load(open('myfirst_model.pkl','rb')) 
Read more

What is MLflow?

We will not get lost in more details however it's good to know that MLflow help in managing ML models and we can deploy it to cloud services like Amazon sagemaker , Azure and kubernetes etc. pip install mlflow  or pip install mlflow[extras] Components of MLflow: 1. Tracking 2. Model Registry  3. Model 4. Projects 
Read more