Datasets:
add oracle features
Browse files
README.md
CHANGED
@@ -101,12 +101,6 @@ The dataset is based on the three commonly used datasets:
|
|
101 |
- [IMDB dataset](https://huggingface.co/datasets/imdb)
|
102 |
- [CIFAR-10 dataset](https://www.cs.toronto.edu/~kriz/cifar.html)
|
103 |
|
104 |
-
All datasets are subsampled to be of equal size (n=50,000). The CIFAR-10 data is based on the trainings dataset, whereas the IMDB data contains train and test data
|
105 |
-
to obtain 50,000 observations. The labels of the CIFAR-10 data are set to integer values 0 to 9.
|
106 |
-
The Diamonds dataset is cleaned (values with `x`, `y`, `z` equal to 0 are removed) and outliers are dropped (such that 45<`depth`<75, 40<`table`<80, `x`<30, `y`<30 and 2<`z`<30).
|
107 |
-
The remaining 53,907 observations are downsampled to the same size of 50,000 observations. Further `price` and `carat` are transformed with the natural logarithm and `cut`,
|
108 |
-
`color` and `clarity` are dummy coded (with baselines Fair, D and I1).
|
109 |
-
|
110 |
The versions to create this dataset can be found on Kaggle:
|
111 |
|
112 |
- [Diamonds dataset](https://www.kaggle.com/datasets/shivam2503/diamonds)
|
@@ -115,6 +109,14 @@ The versions to create this dataset can be found on Kaggle:
|
|
115 |
|
116 |
The original citations can be found below.
|
117 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
118 |
## Uses
|
119 |
|
120 |
The dataset should as a benchmark to compare different causal inference methods for observational data under multimodal confounding.
|
@@ -131,20 +133,22 @@ The data fields can be devided into several categories:
|
|
131 |
- `Y` (`float64`): Outcome of interest
|
132 |
- `D_1` (`float64`): Treatment value
|
133 |
|
134 |
-
- **Tabular Features**
|
135 |
-
- `price` (`float64`):
|
136 |
-
|
137 |
- **Text Features**
|
138 |
- `review` (`string`): IMDB review text
|
139 |
-
- `sentiment` (`string`): Corresponding
|
140 |
|
141 |
- **Image Features**
|
142 |
- `image` (`image`): Image
|
143 |
- `label` (`int64`): Corresponding label from `0` to `9`
|
144 |
|
|
|
|
|
|
|
145 |
- **Oracle Features**
|
146 |
-
- `cond_exp_y` (`float64`): Expected value `Y` conditional on `D_1`,
|
147 |
-
- `
|
|
|
|
|
148 |
|
149 |
## Limitations
|
150 |
|
|
|
101 |
- [IMDB dataset](https://huggingface.co/datasets/imdb)
|
102 |
- [CIFAR-10 dataset](https://www.cs.toronto.edu/~kriz/cifar.html)
|
103 |
|
|
|
|
|
|
|
|
|
|
|
|
|
104 |
The versions to create this dataset can be found on Kaggle:
|
105 |
|
106 |
- [Diamonds dataset](https://www.kaggle.com/datasets/shivam2503/diamonds)
|
|
|
109 |
|
110 |
The original citations can be found below.
|
111 |
|
112 |
+
### Dataset Preprocessing
|
113 |
+
|
114 |
+
All datasets are subsampled to be of equal size (`50,000`). The CIFAR-10 data is based on the trainings dataset, whereas the IMDB data contains train and test data
|
115 |
+
to obtain `50,000` observations. The labels of the CIFAR-10 data are set to integer values `0` to `9`.
|
116 |
+
The Diamonds dataset is cleaned (values with `x`, `y`, `z` equal to `0` are removed) and outliers are dropped (such that `45<depth<75`, `40<table<80`, `x<30`, `y<30` and `2<z<30`).
|
117 |
+
The remaining `53,907` observations are downsampled to the same size of `50,000` observations. Further `price` and `carat` are transformed with the natural logarithm and `cut`,
|
118 |
+
`color` and `clarity` are dummy coded (with baselines `Fair`, `D` and `I1`).
|
119 |
+
|
120 |
## Uses
|
121 |
|
122 |
The dataset should as a benchmark to compare different causal inference methods for observational data under multimodal confounding.
|
|
|
133 |
- `Y` (`float64`): Outcome of interest
|
134 |
- `D_1` (`float64`): Treatment value
|
135 |
|
|
|
|
|
|
|
136 |
- **Text Features**
|
137 |
- `review` (`string`): IMDB review text
|
138 |
+
- `sentiment` (`string`): Corresponding sentiment, either `positive` or `negative`
|
139 |
|
140 |
- **Image Features**
|
141 |
- `image` (`image`): Image
|
142 |
- `label` (`int64`): Corresponding label from `0` to `9`
|
143 |
|
144 |
+
- **Tabular Features**
|
145 |
+
- `price` (`float64`):
|
146 |
+
|
147 |
- **Oracle Features**
|
148 |
+
- `cond_exp_y` (`float64`): Expected value of `Y` conditional on `D_1`, `sentiment`, `label` and `price`
|
149 |
+
- `l1` (`float64`): Expected value of `Y` conditional on `sentiment`, `label` and `price`
|
150 |
+
- `m1` (`float64`): Expected value of `D_1` conditional on `sentiment`, `label` and `price`
|
151 |
+
- `g1` (`float64`): Additive component of `Y` based on `sentiment`, `label` and `price`
|
152 |
|
153 |
## Limitations
|
154 |
|