Delayed Feedback Model (DFM)

Overview

PyTorch implementation of the paper.
Modeling Delayed Feedback in Display Advertising, Olivier Chapelle, KDD2014

Dataset

columns

• feature1 ... feature_n: Categorical feautre column. (Assuming all variables are categorical.)
• elapsed_day: Days elapsed since click.
• cv_delay_day: Days delayed from click to conversion. (Only observable if conversions are observed)
• supervised: Conversion label. (If conversions are observed 1)

sample dataset

   feature1  feature2  feature3  elapsed_day  cv_delay_day  supervised
0         1         1         1           10           3.0           1
1         3         3         3            3           NaN           0
2         5         5         5           30           NaN           0
3         7         7         7            2           1.0           1
4         2         2         2            6           NaN           0
5         5         5         5            1           NaN           0
6         1         1         1           11           8.0           1
7         3         3         3           32           NaN           0

remarks

In the paper, feature hashing is used for vectorization of categorical variables.

All the features are mapped into a sparse binary feature vector of dimension 2^24 via the hashing trick [17].

In this implementation, embedding layer is used instead of feature hashing for vectorization.