vivienzou1 / homedepotkaggle

Home Depot Product Relevance Model - A Kaggle task

HomeDepotKaggle

Home Depot Product Relevance Model - A Kaggle task

Home Depot Product Relevance Model Competition

Getting Started

This is a minor study(play around) for the Home Depot Product Relevance Model Kaggle Task. Our final score is not optimized to hit the top leaderboard, but our performance is considerable for only 6 features.

Prerequisites

Python 3.6
Jupyter notebook

Running the tests

Just simply download and run the script at your choice (python script or jupyter notebook)

Features

Only 6 features extracted from the dataset to achieve the study goal. They are:

1. Levenshtein distance between search term and product title
2. Levenshtein distance between search term and product description
3. Cosine similarity between search term and product title (TF-IDF)
4. Cosine similarity between search term and product description (TF-IDF)
5. Cosine similarity between search term and product title (Word2Vec)
6. Cosine similarity between search term and product description (Word2Vec)

Pipeline

1. Fill up all missing data with empty string value or default value
2. Merge the training data set and testing data set with description data set
3. Extract and parse different features .. * Apply the following steps on each description: to_lowercase -> split -> apply Snowball stemmer -> aggregate
4. Refine training data set and testing data set
5. Test data with different models

Models and score

Random Forest

Parameters	Private Score	Public Score
n_estimators=30, max_depth=16	0.48779	0.48956
n_estimators=100, max_depth=80	0.4943	0.49739
n_estimators=128, max_depth=8	0.4856	0.48769
n_estimators=128, max_depth=5	0.49004	0.4913
n_estimators=128, max_depth=15	0.48734	0.4899
n_estimators=500, max_depth=5, min_samples_leaf=6, max_features=0.9, min_samples_split=1.0, n_jobs= -1, random_state=2014	0.48971	0.4911

Ada Boost

Parameters	Private Score	Public Score
n_estimators=30	0.50011	0.50039
n_estimators=128	0.49972	0.49978
n_estimators=300	0.50021	0.50048

Boosting Regression

Parameters	Private Score	Public Score
n_estimators=30	0.50007	0.5034
n_estimators=128	0.49383	0.49643
n_estimators=300	0.49295	0.49592

Extra Tree Regression

Parameters	Private Score	Public Score
n_estimators=30, max_depth=16	0.49062	0.4918
n_estimators=128, max_depth=8	0.48808	0.48949
n_estimators=128, max_depth=16	0.48659	0.48817
n_estimators=300, max_depth=16	0.48604	0.48784

Gradient Boosting Regression

Parameters	Private Score	Public Score
n_estimators=30, max_depth=6	0.48533	0.48728
n_estimators=128, max_depth=6	0.48505	0.48733
n_estimators=128, max_depth=16	0.50459	0.50796
n_estimators=500, max_depth=6, min_samples_split=1.0, min_samples_leaf=15, learning_rate=0.035, loss='ls',random_state=10	0.48748	0.4897
n_estimators=500, max_depth=6, min_samples_split=2, min_samples_leaf=15, learning_rate=0.035, loss='ls',random_state=10	0.48476	0.48694

Linear Regression

Parameters	Private Score	Public Score
	0.49104	0.49301

Ridge Regression

Parameters	Private Score	Public Score
	0.49104	0.49302

MLP Regression

Parameters	Private Score	Public Score
Solver=lbfgs, alpha=1e-5	0.74715	0.74326
Solver=adam,alpha=1e-3	0.81787	0.8206

eXtreme Gradient Boosting Regression (XGB)

Parameters Private Score Public Score
colsample_bytree: 1	silent: 1	nthread: 8	min_child_weight: 10	'n_estimators': 300	subsample: 1	learning_rate: 0.09	objective: 'reg:linear'	seed: 10	max_depth: 7	gamma: 0. 0.48663 0.48888
colsample_bytree: 0.77	silent: 1	nthread: 8	min_child_weight: 15	0.48419 0.48659
n_estimators: 500	subsample: 0.77	learning_rate: 0.035	objective: 'reg:linear'
seed: 11	max_depth: 6	gamma: 0.2

Bagging Regression + XGB

Parameters Private Score Public Score
colsample_bytree: 1	silent: 1	nthread: 8	min_child_weight: 10	'n_estimators': 300	subsample: 1	learning_rate: 0.09	objective: 'reg:linear'	seed: 10	max_depth: 7	gamma: 0. n_estimators=10	random_state=np.random.RandomState(2018) 0.48459 0.4871
colsample_bytree: 0.77	silent: 1	nthread: 8	min_child_weight: 15	0.48379 0.48613
n_estimators: 500	subsample: 0.77	learning_rate: 0.035	objective: 'reg:linear'
seed: 11	max_depth: 6	gamma: 0.2 n_estimators=10	random_state=np.random.RandomState(2018)

KNN

Parameters	Private Score	Public Score
128, weights="uniform", leaf_size=5	0.52871	0.52994

Decision Tree

Parameters	Private Score	Public Score
	0.52605	0.52666

With Attribute (compared with same parameters)

Models	Public Score without Attribute	Public Score with Attribute
Random Forest	0.4911	0.53559
AdaBoost	0.49592	0.52592
Bagging	0.48613	0.51251
Gradient Boosting	0.48694	0.51322
Linear Regression	0.49301	0.5276
Ridge	0.49302	0.5276
XGB	0.48659	0.51272
KNN	0.52994	0.51984
Decision Tree	0.52666	0.52998