minor...

modules/icing/pirep_goes.py

@@ -1717,26 +1717,62 @@ def split_data(times):
                    [get_timestamp('2022-01-01_00:00'), get_timestamp('2022-01-07_23:59')],
                    [get_timestamp('2022-03-01_00:00'), get_timestamp('2022-03-07_23:59')]]
 
-    test_time_idxs = []
+    keep_out = 10800  # 3 hrs
+
+    vld_time_idxs = []
+    for t_rng in time_ranges:
+        t_rng[0] -= keep_out
+        t_rng[1] += keep_out
+        tidxs = np.searchsorted(times, t_rng)
+        vld_time_idxs.append(np.arange(tidxs[0], tidxs[1], 1))
+    vld_time_idxs = np.concatenate(vld_time_idxs, axis=None)
+    # train_time_idxs = time_idxs[np.in1d(time_idxs, vld_time_idxs, invert=True)]
+
+    # Save this just in case.
+    # # Keep out
+    # out_idxs = []
+    # for k, t_rng in enumerate(time_ranges):
+    #     t_a = time_ranges[k][0]
+    #     t_b = time_ranges[k][1]
+    #     tidxs = np.searchsorted(times, [t_a - 10800, t_a])
+    #     out_idxs.append(np.arange(tidxs[0], tidxs[1], 1))
+    #     tidxs = np.searchsorted(times, [t_b, t_b + 10800])
+    #     out_idxs.append(np.arange(tidxs[0], tidxs[1], 1))
+    # out_idxs = np.concatenate(out_idxs, axis=None)
+    # train_time_idxs = train_time_idxs[np.in1d(train_time_idxs, out_idxs, invert=True)]
+
+    time_ranges = [[get_timestamp('2018-02-01_00:00'), get_timestamp('2018-02-05_23:59')],
+                   [get_timestamp('2018-04-01_00:00'), get_timestamp('2018-04-05_23:59')],
+                   [get_timestamp('2018-06-01_00:00'), get_timestamp('2018-06-05_23:59')],
+                   [get_timestamp('2018-08-01_00:00'), get_timestamp('2018-08-05_23:59')],
+                   [get_timestamp('2018-10-01_00:00'), get_timestamp('2018-10-05_23:59')],
+                   [get_timestamp('2018-12-01_00:00'), get_timestamp('2018-12-05_23:59')],
+                   [get_timestamp('2019-02-01_00:00'), get_timestamp('2019-02-05_23:59')],
+                   [get_timestamp('2019-04-01_00:00'), get_timestamp('2019-04-05_23:59')],
+                   [get_timestamp('2019-06-01_00:00'), get_timestamp('2019-06-05_23:59')],
+                   [get_timestamp('2019-08-01_00:00'), get_timestamp('2019-08-05_23:59')],
+                   [get_timestamp('2019-10-01_00:00'), get_timestamp('2019-10-05_23:59')],
+                   [get_timestamp('2019-12-01_00:00'), get_timestamp('2019-12-05_23:59')],
+
+                   [get_timestamp('2021-10-05_00:00'), get_timestamp('2021-10-10_23:59')],
+                   [get_timestamp('2021-12-01_00:00'), get_timestamp('2021-12-05_23:59')],
+                   [get_timestamp('2022-02-01_00:00'), get_timestamp('2022-02-05_23:59')],
+                   [get_timestamp('2022-03-25_00:00'), get_timestamp('2022-03-30_23:59')]]
+
+    tst_time_idxs = []
     for t_rng in time_ranges:
+        t_rng[0] -= keep_out
+        t_rng[1] += keep_out
         tidxs = np.searchsorted(times, t_rng)
-        test_time_idxs.append(np.arange(tidxs[0], tidxs[1], 1))
-    test_time_idxs = np.concatenate(test_time_idxs, axis=None)
-    train_time_idxs = time_idxs[np.in1d(time_idxs, test_time_idxs, invert=True)]
-
-    # Keep out
-    out_idxs = []
-    for k, t_rng in enumerate(time_ranges):
-        t_a = time_ranges[k][0]
-        t_b = time_ranges[k][1]
-        tidxs = np.searchsorted(times, [t_a - 10800, t_a])
-        out_idxs.append(np.arange(tidxs[0], tidxs[1], 1))
-        tidxs = np.searchsorted(times, [t_b, t_b + 10800])
-        out_idxs.append(np.arange(tidxs[0], tidxs[1], 1))
-    out_idxs = np.concatenate(out_idxs, axis=None)
-    train_time_idxs = train_time_idxs[np.in1d(train_time_idxs, out_idxs, invert=True)]
-
-    return train_time_idxs, test_time_idxs
+        tst_time_idxs.append(np.arange(tidxs[0], tidxs[1], 1))
+    tst_time_idxs = np.concatenate(tst_time_idxs, axis=None)
+
+    vld_tst_time_idxs = np.concatenate([vld_time_idxs, tst_time_idxs])
+    vld_tst_time_idxs = np.sort(vld_tst_time_idxs)
+
+    train_time_idxs = time_idxs[np.in1d(time_idxs, vld_tst_time_idxs, invert=True)]
+
+    return train_time_idxs, vld_time_idxs, tst_time_idxs
 
 
 def normalize(data, param, mean_std_dict, add_noise=False, noise_scale=1.0, seed=None):