emanuele/kaggle_ops

My best submission to the Kaggle competition "Online Product Sales", ranked 21th over 366 teams.

kaggle_ops

My best submission to the Kaggle competition "Online Product Sales", ranked 21th over 366 teams (score: 0.57885).

http://www.kaggle.com/c/online-sales/leaderboard

---

Requirements:

0. NumPy, scikit-learn

1. Pandas, http://pandas.pydata.org/ , just to create the initial dataset or to explore it.

2. joblib, http://packages.python.org/joblib/ , if you want to run blender_parallel.py , i.e. to use all your cores with GradientBoostingRegressor().

3. The "Online Product Sales" trainset/testset files to be put in the subdirectory "data/".

---

Usage:

• "python explore.py" to have a look to the quantitative variables and to decided which of them to put in the logscale.

• "python create_dataset.py" , creates and save the dataset from the initial trainset/testset files.

• "python blender.py" computes the actual submission (simple blending of GradientBoosting).

• "python blender_parallel.py" computes the actual submission splitting the computation on as many cores as you like.

kaggle_ops's Issues

Quan Logs

1. This is regarding the the explorer.py file. Is there a reason by you left out Quan_1-3, 20?

quantitative_columns = list(filter(lambda s: s.startswith("Quan"), df.columns))

This is the list of variables to show in logscale:

to_log = ["Quan_4", "Quan_5", "Quan_6", "Quan_7", "Quan_8", "Quan_9", "Quan_10", "Quan_11", "Quan_12", "Quan_13", "Quan_14", "Quan_15", "Quan_16", "Quan_17", "Quan_18", "Quan_19", "Quan_21", "Quan_22", "Quan_27", "Quan_28", "Quan_29", "Quant_22", "Quant_24", "Quant_25"]

explorer.py

Hi Emanuele,

I'm getting this error with this piece of code:
"ValueError: num must be 1 <= num <= 25, not 0"

for i, col in enumerate(quantitative_columns):
a = df[col]
print(col, pandas.isnull(a).sum())
plt.subplot(5,5,i)
if col in to_log:
a = np.log(a)
plt.hist(a[pandas.notnull(a)], bins=30, label=col)
plt.legend()
print(len(quantitative_columns))