TODO

• date_time split (YY / MM / DD?)
• prop location aggregation
• competitor aggregation (min/max, #num of competitors)

Code

Task 0: Preparation

• Download dataset
• Load in the data-set using a load_expedia_data function.

Task 1: Business Understanding

Related work research, etc.

Task 2: Data Understanding

• Transform variables for pandas

  • Impute null / missing variables:
    • visitor_hist_starrating,
    • visitor_hist_adr_usd,
    • prop_review_score,
    • prop_location_score2,
    • srch_query_affinity_score,
    • orig_destination_distance,
    • comp1_rate,
    • comp1_inv,
    • comp1_rate_percent_diff,
    • comp2_rate,
    • comp2_inv,
    • comp2_rate_percent_diff,
    • comp3_rate,
    • comp3_inv,
    • comp3_rate_percent_diff,
    • comp4_rate,
    • comp4_inv,
    • comp4_rate_percent_diff,
    • comp5_rate,
    • comp5_inv,
    • comp5_rate_percent_diff,
    • comp6_rate,
    • comp6_inv,
    • comp6_rate_percent_diff,
    • comp7_rate,
    • comp7_inv,
    • comp7_rate_percent_diff,
    • comp8_rate,
    • comp8_inv,
    • comp8_rate_percent_diff,
    • gross_bookings_usd

• Generate statistics of the data (mean, spread, etc)

• Generate meaningful plots

• Mark a number of manual features as most interesting (high correlations, etc)?

• Maybe do some more in-depth analysis, such as Chi-square test for categorical variables?

Task 3: Data Preparation

• Impute missing values
• Apply feature selection
• Encode to feature encodings (SK-learn)

Task 4: Modeling and Evaluation

• Test various models (motivate your choices here)

Paper

• Setup a paper format
  ...

General

• Deal with categories in categorical variables that might not exist in the test-set
• Add csv results with DCG score, for each config setup.

Feature Engineering [Sami]

• Combine competitor
• Average location
• IsInSameCountry
• IsInCountryAsDestination
• DateTime

Feature Impute [Thomas]

• Groupby hotel or user, and fill in their nans

Feature Selection [Thomas]

• Explore which models are better (SVC)

• Optional: apply PCA on the data -> throw this into the algorithm as well.

Business Understanding [Thomas / Sami / Jonathan]

• Find concrete inspirations (Tips for each)

Todo's part 1

Coding

• Finish data transformations [Sami]
• Encode unfinished data transformations (sklearn stuff) [Jonathan]
• Move all classes below 5 to an 'others' class

Writing

• Write about preprocessing: feature engineering, explanations

• Statistics and correlations

• Plots about featues and resulting models

• Discussion about dataset / observations/ models etc

• Algorithms and parameters descriptions

• Which features were selected (algorithm-wise)

• Evaluation (performances)

• Compare results

Todo's part 2

Coding

• Finish plots [Thomas]
• Get some basic statistics (maybe variance / correlations/ etc)
• Finish data encoding [Jonathan]
• Train 2 models [jonathan]
• Gather a few metrics (Recall / Precision)

Writings

• Write about plots and maybe a few statistics of the features [thomas]

• Write about our feature transformations (which one we left out, etc) [thomas/jona]

• Write about classifiers, their results and why we used them [jonathan]

Todo's part 3

3a - Kaggle [sami]

• Describe competition
• Describe some winning technique
• Do some analysis

3b - MAE vs MSE [Thomas]

Done

3c - SmsCollective [jonathan]

• Describe techniques used
• Describe what to do for transformations
• Describe which models to use
• Analysis

General

Resources

• Kaggle process with nice data preprocessing including Pipelines
  ❗New

• 32 Sklearn tips on Kaggle

Learn-to-rank General

Resources

• Dataquest Kaggle tutorial
  Slightly different dataset but still relevant

• Lucidworks ABC to Learning to Rank

Our dataset

Resources

• Kaggle: Expedia Personalized Sort Discussion