There will be assessment tests after every 5 assignments: you will get the notifications in your

classroom regarding this, we will be conducting these tests every month.

1. Python (without Numpy) ⇒ ​[Module_1, Module_2(upto 12.37)]

a. Matrix Multiplication
b. Proportional Sampling
c. Replace numbers with #,
i. Ex 1: A = 234 Output: ###
ii. Ex 2: A = a2b3c4 Output: ###
iii. Ex 3: A = abc Output: (empty string)
iv. Ex 5: A = #2a$#b%c%561# Output: ####
d. Print name of students
i. who got top 5 ranks
ii. Who got least 5 ranks
iii. Who got marks in IQR
e. Print 5 closest elements for a given point (p,q) based on the angle between (p,q)
and (x,y)
f. from the given set of hyperplanes find a hyperplane that will separate both red
and green points
g. Given two columns of data, the first column F will have 5 unique values and the
second column S will have 3 unique values (0,1,2).
i. Find P(F_1|S==0), P(F_1|S==1), P(F_1|S==2)
ii. Find P(F_2|S==0), P(F_2|S==1), P(F_2|S==2)
iii. Find P(F_3|S==0), P(F_3|S==1), P(F_3|S==2)
iv. Find P(F_4|S==0), P(F_4|S==1), P(F_4|S==2)
v. Find P(F_5|S==0), P(F_5|S==1), P(F_5|S==2)
h. Filling the missing values in the specified format
i. _,_,_,value ex: _,_,_,40 ⇒ 10,10,10,10
ii. value, _,_,_,value ex:60,_,_,_,40 ⇒ 20,20,20,20,20
iii. Value, _,_,_ ex: 40, _,_,_ ⇒ 10,10,10,10
i. Given two sentences S1, S2
i. Number of common words between S1, S2
ii. Words in S1 but not in S2
iii. Words in S2 but not in S1
j. Given two columns of data Y, Y_Score calculate the value of this function
f(Y,Y_score) = -1*\frac{1}{n}\Sigma(Y Log10(Y_score)+(1-Y)log10(1-Y_score))
2. EDA ⇒ ​[Module_1, Module_2(upto 12.37)]
a. Get insights from data, given set of questions we need to answer them by doing
a bit of analysis
b. Need two data sets of similar type one for reference, one for assignment
3. Implementing TFIDF vectorizer ⇒ ​[Module_1, Module_2, Module_3(upto 18.14)]
a. Build a TFIDF Vectorizer, given the reference for countvectorizer
b. Implement min_df and max_feautres attributes
4. Implement RandomSearchCV with k fold cross validation on KNN ⇒ ​[Module_1,
Module_2, Module_3(upto 19.31)]
a. For each hyper parameter select two disjoint set of indices and divide the data
into train and test
b. Train model on train data and find the performance metric value on test data
c. Calculate the average performance metric score for each hyper parameter

5. Compute Performance metrics without Sklearn ⇒ ​[Module_1, Module_2, Module_3(upto

22.8)]
Given original and predicted values (without sklearn)
a. #P >> #N
i. Calculate F1 Score
ii. Calculate AUC
iii. Calculate Accuracy
iv. Calculate confusion matrix
b. #P << #N
i. Calculate F1 Score
ii. Calculate AUC
iii. Calculate Accuracy
iv. Calculate confusion matrix
c. Find out the best threshold from given probability scores that will give the lowest
score f(y,y^predict) = a*fpr+b*fnr, here y^pred = [0 if y_score < y_threshold else
1]
d. Given y and y_predicted(both are real-valued features)
i. Calculate MSE
ii. MAPE
iii. R Squared Error
ASSESSMENT TEST: 1
6. Apply Multinomial NB on Donors Choose Dataset ⇒ ​[Module_1, Module_2,
Module_3(upto 24.20)]
7. Implement SGD Classifier with Log Loss and L2 regularization Using SGD: without using
sklearn ⇒ ​[Module_1, Module_2, Module_3(upto 27.11)]
8. How each model behaves ⇒ ​[We will be updating assignments soon, Module_1,
Module_2, Module_3, Module_4(upto 29.14)]
a. Given a highly imbalanced dataset (we will be incrementing imbalance in data
adding few data points) observe how each model behaves
i. Draw hyperplane in Logistic regression
ii. Draw hyperplane in SVM
iii. Draw decision boundary in KNN
b. Given 3d data points, such a way that var(3)>>(var2)>var(1)
i. How hyperplane changes before and after standardization of data in
Logistic Regression
ii. How hyperplane changes before and after normalization of data in
Logistic Regression
iii. How hyperplane changes before and after standardization of data in SVM
iv. How hyperplane changes before and after normalization of data in SVM
c. Elliptical data with one or two outliers linear regression
d. Create a dataset with features [X, X^2, 2*X, Y, Z, X+Y] and perform perturbation
test (iris.csv)
e. What happens in testing time of kernel SVM
f. Plot scaling and Isotonic Regression to find the probabilities from the outputs of
SVM [ ​Miscellaneous Topics in module 5​: ​35.1 ​Calibration of Models:Need for calibration
35.2 C ​ alibration Plots​. 35.3 Platt’s C
​ alibration/Scaling.​ 35.4 I​ sotonic Regression​]
9. Apply Decision Trees on Donors Choose Dataset: ​[Module_1, Module_2, Module_3,
Module_4(upto 31.14)]
10. Application of Bootstrap samples in Random Forest: ​[Module_1, Module_2, Module_3,
Module_4(upto 33.8)]
a. Choose any base model of your choice(either a Decision tree or Logistic
regression). You can choose 30 to 40 base models based on your RAM.
b. Do both Row and column sampling to train each of the base learners.
c. Find the confidence interval on AUC based on the results of base learners
ASSESSMENT TEST: 2
11. Apply GBDT/XGBOOST/LIGHT-GBM on Donors Choose Dataset: ​[Module_1,
Module_2, Module_3, Module_4(upto 33.18)]
12. Clustering on Graph Dataset: ​[Module_1, Module_2, Module_3, Module_4,
Module_7(upto 45.9)]
13. Recommendation Systems and Truncated SVD: Implement SGD algorithm to predict the
ratings that user is going to give to given movie.
Provided reference notebook ​[Module_1, Module_2, Module_3, Module_4,
Module_7(upto 46.14) and case study 9]
14. Microsoft Malware detection Case Study assignment: ​[Module_1, Module_2, Module_3,
Module_4, Module_5, Module_6(refer case study 6)]
15. Facebook Assignment: ​[Module_1, Module_2, Module_3, Module_4, Module_5,
Module_6(refer case study 3)]
16. SQL Assignment: ​[Module_1(upto 9.27)]
ASSESSMENT TEST: 3
17. Implement a backpropagation on a given computation graph: ​[Module_1, Module_2,
Module_3, Module_4, Module_5, Module_8(upto 50.14),
https://www.youtube.com/watch?v=i94OvYb6noo​, we will be providing reference videos
and notebooks]
a. Reference will be given for a couple of computational graphs
18. Tensorflow Assignment, working with callbacks and vanishing gradient problem: ​[We will
be updating assignments soon, Module_1, Module_2, Module_3, Module_4, Module_5,
Module_8(upto 50.14), we will be providing reference videos and notebooks]
19. Given an rvl-cdip dataset, classify the given document using transfer learning: ​[We will
be updating assignments soon, Module_1, Module_2, Module_3, Module_4, Module_5,
Module_8(upto 53.18), we will be providing reference videos and notebooks]
a. Model 1: ​INPUT --> VGG-16 without Top layers(FC) --> Conv Layer --> Max
pool Layer --> 2 FC layers --> Output Layer ​Train only new Conv block, FC
layers, output layer. Don't train the VGG-16 network.
b. Model 2: ​INPUT --> VGG-16 without Top layers(FC) --> 2 Conv Layers
identical to FC --> Output Layer ​Train only last 2 Conv layers identical to FC
layers, 1 output layer. Don't train the VGG-16 network.
c. Model 3: '​INPUT --> VGG-16 without Top layers(FC) --> 2 Conv Layers
identical to FC --> Output Layer​' and train only Last 6 Layers of VGG-16
network, 2 Conv layers identical to FC layers, 1 output layer.
20. Classifying CIFAR-10 dataset images with DenseNet and work with optimization: ​[We
will be updating assignments soon, Module_1, Module_2, Module_3, Module_4,
Module_5, Module_8(upto 53.18), we will be providing reference videos and notebooks]
a. Reference will be given for Dense-net architectures
ASSESSMENT TEST: 4
21. Object detection - YOLO pretrained model on image net dataset: ​[We will be updating
assignments soon, Module_1, Module_2, Module_3, Module_4, Module_5,
Module_8(upto 53.18), we will be providing reference videos and notebooks]
22. CNN with text dataset: ​[We will be updating assignments soon, Module_1, Module_2,
Module_3, Module_4, Module_5, Module_8(upto 53.18), we will be providing reference
videos and notebooks]
23. LSTM with Text and categorical data: ​[We will be updating assignments soon, Module_1,
Module_2, Module_3, Module_4, Module_5, Module_8(upto 54.10), we will be providing
reference videos and notebooks]
a. Model 1: Glove embedding on text data, embedding layers on categorical
features, dense layers for numerical features
b. Model 2: Glove embedding on text data(consider words with TF IDF values within
IQR), embedding layers on categorical features, dense layers for numerical
features
c. Model 3: Glove embedding on text data → LSTM, one hot encode the categorical
and merge all of them → CNN1D. Merge both the outputs of LSTM and CNN1D
24. LSTM with Time series data: ​[We will be updating assignments soon, Module_1,
Module_2, Module_3, Module_4, Module_5, Module_8(upto 54.10), we will be providing
reference videos and notebooks]
25. Encode-decoder Architecture for text abstraction, seq-seq: ​[We will be updating
assignments soon, Module_1, Module_2, Module_3, Module_4, Module_5,
Module_8(upto 54.10), we will be providing reference videos and notebooks]
ASSESSMENT TEST: 5
26. Personal Case study -1: ML/RS
27. Personal Case study -2: DL NLP
28. Personal Case study -3: DL CV
29. Blogs on Personal Case studies
30. Blog on given concept