50. Homework-7#

50.1. Total number of points: 40#

51. Due date: Nov 21, 2022#

Before you submit this homework, make sure everything runs as expected. First, restart the kernel (in the menu, select Kernel → Restart) and then run all cells (in the menubar, select Cell → Run All). You can discuss with others regarding the homework but all work must be your own.

This homework will test your knowledge on random forests (including feature importances), and neural networks. The Python notebooks shared will be helpful to solve these problems.

Steps to evaluate your solutions:

Step-1: Try on Colab or Anaconda (Windows: https://docs.anaconda.com/anaconda/install/windows/ ; Mac:https://docs.anaconda.com/anaconda/install/mac-os/ ; Linux: https://docs.anaconda.com/anaconda/install/linux/)

Step-2: Open the Jupyter Notebook by first launching the anaconda software console

Step-3: Open the homework’s .ipynb file and write your solutions at the appropriate location “# YOUR CODE HERE”

Step-4: You can restart the kernel and click run all (in the menubar, select Cell → Run All) on the center-right on the top of this window.

Step-5: Now go to “File” then click on “Download as” then click on “Notebook (.ipynb)” Please DO NOT change the file name and just keep it as “.ipynb”

Step-6: Go to lms.rpi.edu and upload your homework at the appropriate link to submit this homework.

52. Please note that for any question in this assignment you will receive points ONLY if your solution passes all the test cases including hidden testcases as well. So please make sure you try to think all possible scenarios before submitting your answers.#

  • Note that hidden tests are present to ensure you are not hardcoding.

  • If caught cheating:

    • you will receive a score of 0 for the 1st violation.

    • for repeated incidents, you will receive an automatic ‘F’ grade and will be reported to the dean of Lally School of Management.

#Please run this cell to load the dataset. 
url = 'https://raw.githubusercontent.com/lmanikon/lmanikon.github.io/master/teaching/datasets/dataset_Facebook.csv'

import pandas as pd
import math
df  = pd.read_csv(url)
df = df.dropna()
df.columns

df

53. 1. Create a new dataframe df_sub2 that contains only these feature columns in df :#

  • Page total likes, Type, Category, Post Month, Post Weekday, Post Hour, Paid, Total Interactions

54. 2. Transform the categorical attribute Type in df_sub2 to numerical attribute this way:#

  • Link:1, Photo:2, Status:3, Video:4

55. 3. Perform Standardization (using this formula https://en.wikipedia.org/wiki/Standard_score) only on Page total likes column in df_sub2#

  • Please use <dataframe>['<column>'].mean() <dataframe>['<column>'].std() if you are using the mean and std values to manipulate the column.

56. 4. Using df_sub2 perform train_test_split operation to build training (X_train, y_train) and testing data (X_test, y_test).#

  • Use test_size=0.3, random_state=42 as the parameters for train_test_split() function.

  • Feature columns (X): Page total likes, Type, Category, Post Month, Post Weekday, Post Hour, Paid

  • CLass label column (y): Total Interactions

57. 5. Train the randomforest regressor (initialized as variable rf) using these parameters: max_depth=3, random_state=0.#

  • Using the trained model rf compute the impurity-based feature importances.

  • Append the names of these top-3 features to list impFeatures. Please make sure you type the feature names exactly as in df_sub2

# YOUR CODE HERE
raise NotImplementedError()

##Cell-1 -- Do not modify this cell
assert set(df_sub2.columns)=={'Post Hour', 'Total Interactions', 'Post Weekday', 'Category', 'Paid', 'Post Month', 'Page total likes', 'Type'}
assert len(df_sub2)==495
assert len(df_sub2.iloc[0,:])==8

##Cell-2 -- Do not modify this cell
assert round(df_sub2['Page total likes'].mean(),0)==0
assert len(X_train)==346
assert 'Post Hour' in impFeatures

##Cell-3 -- Do not modify this cell
#assert rf.n_classes_==230  #THIS WAS AN INCORRECT TEST.
assert len(impFeatures)==3
assert math.ceil(df_sub2['Page total likes'].std())==1

##Cell-4 -- Do not modify this cell
assert (rf.n_features_)==7
assert len(y_test)==149
assert set(impFeatures)=={'Page total likes', 'Post Hour', 'Post Weekday'}

58. Part-2 [16 points]: We will now use neural networks to model a regression problem.#

59. 1. Create a new dataframe df_sub4 that contains only these feature columns in df :#

  • Page total likes, Type, Category, Post Month, Post Weekday, Post Hour, Paid, Total Interactions

60. 2. Perform one-hot encoding on these features below in the dataframe df_sub4 to create a new dataframe df_OHE#

  • Type, Category, Post Month, Post Weekday, Post Hour

61. 3. Perform normalization only on Page total likes and Total Interactions column in df_OHE using MinMaxScaler or minmax_scale#

  • note that both these functions will output the same result

62. 4. Using df_OHE perform train_test_split operation to build training (X_train, y_train) and testing data (X_test, y_test).#

  • Use test_size=0.3, random_state=42 as the parameters for train_test_split() function.

  • Feature columns (X): Everything except Total Interactions

  • CLass label column (y): Total Interactions

63. 5. Build a 3-layer neural network with relu as the 2 hidden layers’ activation function and sigmoid as the final layer’s activation function.#

  • Use the same adam optimizer and mean_squared_error as the loss function.

  • epochs=150, batch_size=10

  • since the class label is continuous, we use metrics.MeanSquaredError() as the metric (from keras import metrics) to measure the performance of our classifier captured as variable mse

#Please run this cell to install keras
#!pip install keras
#!pip install tensorflow

# YOUR CODE HERE
raise NotImplementedError()

##Cell-5 -- Do not modify this cell
assert (math.ceil(df_OHE['Page total likes'].sum())) == 357
assert (math.ceil(df_OHE['Total Interactions'].sum())) == 17
assert (math.ceil(df_OHE['Page total likes'].mean())) == 1
assert (math.ceil(df_OHE['Total Interactions'].mean())) == 1

##Cell-6 -- Do not modify this cell
assert len(X_train)==346
assert len(y_test)==149
assert mse<0.005
assert len(model.weights)==6