WEBVTT 00:00.080 --> 00:01.080 Welcome back. 00:01.120 --> 00:04.240 Let's import the necessary libraries. 00:04.240 --> 00:11.080 So here create a new cell and let's import our necessary libraries. 00:11.080 --> 00:13.960 So this is the first section importing libraries. 00:14.000 --> 00:26.320 Let's start with import TensorFlow as tf import pandas as pd, import numpy as np and import matplotlib. 00:26.800 --> 00:28.680 Pyplot as plt. 00:28.880 --> 00:35.320 For better data visualization, we're going to use a new library called Seaborn. 00:35.560 --> 00:43.280 Seaborn is a Python data visualization library built on top of matplotlib library. 00:43.320 --> 00:49.240 It provides a high level interface for creating attractive statistical graphics. 00:49.440 --> 00:57.480 We're going to use like, uh, heatmaps and other complex and advanced graphics. 00:57.480 --> 01:03.920 So for this we're going to use our lovely library called Seaborn. 01:04.040 --> 01:07.070 Import Seaborn as SNES. 01:07.350 --> 01:17.590 If you are using the IDE's Python IDE's, you should start with, for example pip install TensorFlow, 01:17.990 --> 01:24.790 pip install seaborn, pip install matplotlib, pandas and numpy. 01:25.150 --> 01:34.030 And if you are using like IDE's, PyCharm, Jupyter, you should execute those Python package commands. 01:34.030 --> 01:43.590 Also, in order to check that they are already installed by Google Colab, you can run those commands, 01:43.630 --> 01:50.150 adding the exclamation mark to execute the Python installer package from a cell. 01:50.150 --> 01:56.190 So you notice that Google Colab requirement already satisfied. 01:56.190 --> 02:03.750 So all of those libraries are found directly and without using the pip install with. 02:03.910 --> 02:07.150 They are built in in the notebooks. 02:07.190 --> 02:15.470 Okay, so there is no need to to install any of those libraries because they are already installed and 02:15.470 --> 02:16.270 satisfied. 02:16.430 --> 02:18.110 So post this. 02:18.150 --> 02:20.910 Remove this cell and here we go. 02:21.190 --> 02:24.910 Okay, so we imported this new library. 02:24.950 --> 02:29.070 Seaborn, as I told you, it's used for data visualization. 02:29.270 --> 02:36.070 It's built on built on top of Matplot library for advanced visualization. 02:36.070 --> 02:43.190 We use Seaborn because it turns complex statistical relationships into beautiful, intuitive visualization 02:43.390 --> 02:48.910 that helps students and professionals understand their data at a glance. 02:48.950 --> 02:54.910 Also, we're going to use our old friend called scikit learn. 02:55.070 --> 03:03.550 We talked about Sky scikit learn library in the previous sections and in the introductory sections with 03:03.550 --> 03:06.950 data visualization and creating the models. 03:06.990 --> 03:13.110 We're going in this model to use scikit learn with TensorFlow. 03:13.110 --> 03:20.860 So TensorFlow is great for building and training neural networks, deep learning architectures, GPU 03:20.900 --> 03:24.700 acceleration, production, deployment and sky kit. 03:24.740 --> 03:31.460 Learn is great for data pre-processing and feature engineering, traditional machine learning algorithms, 03:31.500 --> 03:37.260 model evaluation metrics, data splitting and cross-validation. 03:37.260 --> 03:44.500 So we're going to use the scikit learn with with this model because we have nine parameters. 03:44.500 --> 03:47.820 We need a simple API one line to split data. 03:47.860 --> 03:54.540 Also, we're going to split the data between training and testing, flexible test size, random state 03:54.540 --> 04:02.340 and and used in thousands of projects and perfect for TensorFlow returning numpy arrays. 04:02.340 --> 04:09.980 So we use scikit learn for data pre-processing because preventing data leakage is critical for valid 04:09.980 --> 04:10.500 results. 04:10.540 --> 04:19.220 Simpler code industry standards, error resistance, and model agnostic same pre-processing works with 04:19.220 --> 04:20.860 any ML library. 04:20.980 --> 04:26.580 TensorFlow excels at what it's designed for building and training neural networks. 04:26.620 --> 04:30.980 Let each library do what it does best. 04:31.020 --> 04:31.460 Okay. 04:31.820 --> 04:39.780 So we're going to use it with visualizing the standard metrics additional metrics and clear interpretations 04:39.780 --> 04:41.420 from Sky. 04:42.260 --> 04:53.940 Learn model selection import train test split Skyler metrics, import mean squared error, MSE mean 04:53.980 --> 04:58.580 absolute error, and r squared score. 04:58.620 --> 05:03.900 Also we're going to import our friend standard scatter. 05:04.100 --> 05:08.540 And this is the main reason I used the sky layer. 05:08.580 --> 05:14.980 So sky learn dot pre-processing import standard scalar. 05:15.020 --> 05:25.100 Standard scalar is a data pre-processing tool that transforms your features to have a mean of zero and 05:25.140 --> 05:28.290 standard deviation of one. 05:28.330 --> 05:35.010 This is also called standardization or Z-score normalization. 05:35.010 --> 05:43.050 And we're going to use the skyline pre-processing and import the Standardscaler from this package in 05:43.050 --> 05:45.930 order to use it with our data. 05:45.970 --> 05:56.730 The general formula of standardization, or Z-score is Z equals to x minus m over D, X is the original 05:56.730 --> 06:06.250 feature value minus mean of the feature over the standard deviation of the feature. 06:06.290 --> 06:09.210 Z is the scaled value Z-score. 06:09.330 --> 06:16.130 Okay, so I want from you to understand this and why we're going to use the standard scaler in order 06:16.130 --> 06:22.410 to transform your features, to have a mean of zero and standard deviation of one. 06:22.450 --> 06:22.890 Okay. 06:23.090 --> 06:28.090 Use standard scaler when features have different units and scales in. 06:28.090 --> 06:29.610 In this example we have. 06:29.650 --> 06:38.730 We have nine features and every feature has a different scale or unit, and using algorithms sensitive 06:38.730 --> 06:47.810 to feature scales like neural networks, SVM, KNN, you assume features are roughly normally distributed. 06:47.810 --> 06:58.610 So data is very important and the type of the data is very crucial in order to, uh, to determine what 06:58.650 --> 07:03.770 algorithms, what libraries, what scalars you should use. 07:03.770 --> 07:12.290 In this case, we're going to use the standard scalar that it is used that makes features compatible 07:12.290 --> 07:14.850 by giving them same scale. 07:14.850 --> 07:23.210 So we have nine different scales with nine different inputs, nine different variables and features. 07:23.290 --> 07:26.050 We need to normalize them. 07:26.050 --> 07:31.130 So standard scalar makes features compatible by giving the.