deep_coders(sourav,nitin)
- 1. DEEP CODERS ECKOVATION MACHINE LEARNING members Nitin Khatkar :01711503116 Sourav Tiwari :03011503116 Gulshan :01211503116 Shrey Achreja :41311503116
- 2. What cuisine is this recipe?? Picture yourself strolling through your local, open-air market... What do you see? What do you smell? What will you make for dinner tonight? We want to thank Yummly for providing this unique dataset. 2
- 3. Data Description ▫ In the dataset, we include the recipe id, the type of cuisine, and the list of ingredients of each recipe (of variable length). The data is stored in JSON format. ▫ An example of a recipe node in train.json is given aside: 3
- 4. “We would predict the cuisine for each recipe in the test case.” 4
- 5. 5 STEPS FOLLOWED TO SOLVE THE GIVEN PROBLEM STEP 3 At last, we will apply the suitable algorithm to it and find the best suitable algorithm to it. STEP 1 First we will perform EDA and will remove all the redundant data from given dataset. STEP 2 Then , we will form our feature matrix as well target metrix.
- 6. PRE-PROCESSING TOP 10 INGREDIENTS ACCORDING TO THE CUSINE
- 7. ALGORTITHM FOR FINDING TOP 10 CUISINE GIVEN ON NEXT SLIDE -> 7
- 8. ALGORITHM FOR FINDING TOP 10 CUISINE ▫ First make a dictionary with keys as different cuisine’s and ingredients present in it as values.(dic) ▫ Then, with the help of above dictionary make a new dictionary containing the counts of the ingredients present in it.(count_dictionary) ▫ At last, make the pie chart of top 10 ingredients with the help of above two dictionaries.(code given on next slide) 8
- 9. CODE FOR PLOTTING TOP 10 INGREDIENTS 9
- 10. APPLYING ML TO IT FIRST GENERATING X AND Y FOR FURTHER APPLYING ANY ALGORITHM TO IT. 10
- 11. 11 GENERATING X AND Y ▫ Create an empty list y,total_ingredients. ▫ Append all the unique ingredients in the list total_ingredients. ▫ Create a zero matrix using numpy and name it as x.(number of rows equal to y and columns equal to total_ingredients) ▫ For every ingredient in y replace with 1. ▫ Our feature matrix x and target y is ready.
- 12. DIFFERENT ALGORITHM USED 12 FOURTH FINALLY WE WENT FOR FINSL TEST OF DEEP LEARNING BUT DUE COULD NOT PERFORM ON FULL DATA DUE TO LOW END SOECIFIACTIONS THIRD OUT OF CUROSITY, WE ALSO APPLIED NAÏVE BAYES TO IT, BUT AGAIN GOT A VERY LOW SCORE OF 0.36 SECOND THEN ,WE TESTED FOR RANDOM FOREST AND GOT A SATISFACTORY SCORE OF 0.72 FIRST WE STARTED WITH DECISION TREE. OUTCOME WAS NOT AT ALL FRUITFUL AS IT ACHIEVED SCORE OF 0.30
- 13. 13 ALGORITHM 1 ▫ APPLYING DECISION TREE TO IT. ▫ GOT A SCORE OF 0.26
- 14. 14 ALGORITHM 2 ▫ APPLYING RANDOM FOREST TO IT. ▫ GOT A SCORE OF 0.72
- 15. 15 ALGORITHM 3 ▫ APPLYING NAÏVE NAYES TO IT ▫ GOT A SCORE OF 0.36
- 16. 16 ALGORITHM 4 ▫ APPLYING DEEP LEARNING TO IT. ▫ GOT A SCORE OF 0.64 ON ONLY 10000 DATA
- 17. SHORTCOMINGS Deep learning could not be applied on whole dataset due to low end specifications and SVM could also not be applied due to memory error. RESULT Conclusion Random Forest is best algo , But if deep learning performed on full dataset then , conclusion may differ. 17
- 18. DATA COMPARISON SCORE DECISION TREE 0.30 NAÏVE BAYES 0.36 RANDOM FOREST 0.72 DEEP LEARNING 0.64 18
- 19. 0.72 Final score achieved Highest is 0.82 19
- 20. 20 Forest Cover Type Prediction ▫ The study area includes four wilderness areas located in the Roosevelt National Forest of northern Colorado. Each observation is a 30m x 30m patch. We are asked to predict an integer classification for the forest cover type.
- 21. 21 Data Description ▫ The seven types are: ▫ 1 - Spruce/Fir 2 - Lodgepole Pine 3 - Ponderosa Pine 4 - Cottonwood/Willow 5 - Aspen 6 - Douglas-fir 7 - Krummholz ▫ The training set (15120 observations) contains both features and the Cover_Type. The test set contains only the features. You must predict the Cover_Type for every row in the test set (565892 observations).
- 22. “We would predict the forest-cover type based upon the given value of parameters.” 22
- 23. 23 STEPS FOLLOWED TO SOLVE THE GIVEN PROBLEM 23 STEP 3 At last, we will apply the suitable algorithm to it and find the best suitable algorithm to it. STEP 1 First we will perform EDA and will remove all the redundant data from given dataset. STEP 2 Then , we will form our feature matrix as well target metrix.
- 24. 24 PRE-PROCESSING PLOTTING PARAMETER’S WITH RESPECT TO FOREST-COVER TYPE
- 25. 25
- 26. SAMPLE CODE FOR POTTING GRAPH 26
- 27. 27 APPLYING ML TO IT FIRST GENERATING X AND Y FOR FURTHER APPLYING ANY ALGORITHM TO IT. 27
- 28. 28 GENERATING X AND Y ▫ Assigning the value’s in cover_type column in target matrix(y). ▫ Then, after removing the cover_type from data-frame assigning the value to x. ▫ Removing the redundant columns.
- 29. 29 DIFFERENT ALGORITHM USED FOURTH Finally we went ahead for deep learning and got score 0.84 as the Random forest. THIRD OUT OF CUROSITY, WE ALSO APPLIED NAÏVE BAYES and SVM TO IT, BUT AGAIN GOT A VERY LOW SCORE OF 0.58 and 0.14 respectively. SECOND THEN ,WE TESTED FOR RANDOM FOREST AND GOT A SATISFACTORY SCORE OF 0.84 FIRST WE STARTED WITH DECISION TREE. OUTCOME WAS NOT AT ALL FRUITFUL AS IT ACHIEVED SCORE OF 0.60
- 30. 30 ALGORITHM 1 ▫ Applying decision tree to our problem ▫ We got a score of 0.30
- 31. 31 ALGORITHM 2 ▫ Applying random forest to our problem we get. ▫ We got a score of 0.84
- 32. 32 ALGORITHM 3 ▫ Applied naïve-bayes and SVM to it. ▫ But did not get fruitful result because it’s not a probability problem
- 33. 33 ALGORITHM 4 ▫ At last applying deep learning to our problem ▫ We got a score of 0.80
- 34. 34 DATA COMPARISON SCORE DECISION TREE 0.60 NAÏVE BAYES 0.58 RANDOM FOREST 0.84 SVM 0.14 DEEP LEARNING 0.80 34
- 36. 36 THANKS!