changelog.md

Changelog

1.1.0

Cross validation

• params distribution: RandomizedSearchCV
• scorer: CAPE (from Anasse library)

1.0.3

Feature extraction

1. cyclical time encoding:

   2. remove half_hour_of_day
   3. remove day_of_week

2. lags:

   • Features: wind features only
   • range: [1]

3. rollmean:

   • Features: [wind_speed, wind_vector_azimuth, meteorological_wind_direction]
   • period: [3H]

1.0.2

Refactoring

• model to models

Fitting

• Estimator = RandomForestRegressor

• Parameters:

  • set random_state=42 instead of None

• Fill NAs strategy:

  2. Second fill: By self.y_median, i.e the median Production median (instead of 0)

1.0.1

• add the submission file submission.csv to the repo

1.0.0 -- First submission

Preprocessing

1. df_to_ts

Feature extraction

1. cyclical time encoding:

   1. hour_of_day
   2. half_hour_of_day
   3. week_of_year
   4. month_of_year
   5. day_of_week

2. numerical weather prediction merging by median

3. wind speed

4. wind vector azimuth

5. meteorological wind direction

6. lags:

   • Features: all
   • range: [1, 2]

Feature selection

1. remove all NWP features
2. variance threshold: threshold==0.8
3. variance inflation factor: threshold==5

Data cleaning

1. remove remaining NAs
2. Get X and y from X_df and y_df

Fitting

• Estimator = RandomForestRegressor

• Parameters = Default

• Fill NAs strategy:

  1. With last valid observation (done within _clean_data method)
  2. By 0 (if first n lines are NAs (done within _clean_data method))