Update README.md

README.md

@@ -1,3 +1,17 @@
 ---
 license: mit
 ---
+
+# Data Format
+The dataset has been split into `train`, `val`, and `test` jsonl files. 
+Each line of each jsonl file has the following fields:
+* `object_id`: unique identifier for the object
+* `times_wv`: an array of shape _(T, 2)_, where the i-th element contains (t_i, w_i), the time and wavelength of the i-th measured flux value. If t_i = 0, then this is a padded value and the associated flux f_i will also be 0.
+* `lightcurve`: an array of shape _(T, 2)_, where the i-th element contains (f_i, f_err_i), the i-th flux and uncertainty on that flux measurement (flux error).
+* `label`: an integer label corresponding to an astronomical object type. This is the value to be predicted.
+
+# Usage
+The directory includes a custom dataset loading script (`raw_train_with_labels.py`), which will be used when calling 
+```
+load_dataset("/path/to/current/dir")
+```