(Day 21) (Almost) finished prep for TensorFlow developer certificate

Ivan Ivanov · January 22, 2024

Hello :) Today is Day 21!

A quick summary of today:

However, I don’t think I can take the actual test yet. Because I think I need more practice. From tomorrow, I will apply what I learned in the course as the data I see in kaggle.

Today I finished the last part - Sequences, Time series & Prediction

I found the code for a very useful function

image

When predicting time series with tensorflow, the dataset must be divided into windows

image image

You can train with this simple model, but it might be too simple

I learned about the Huber loss - In statistics, the Huber loss is a loss function used in robust regression, that is less sensitive to outliers in data than the squared error loss

image image

You have to change the data to the lambda layer before dividing it into windows and putting it into the Bidirectional LSTM layer. In addition, x*100.0 is applied to change it to a more understandable type in the last layer

In the last piece of homework, I made the below model

image image

results: mse: 5.34, mae: 1.80

  • Certificate

image

I also found a recent talk by Andrew Ng on opporunities in AI

image

The full color is in the current state, but the semi-transparent color is in the state of AI after 3 years.


That is all for today!

See you tomorrow :)

Original post in Korean