link script for rover data; print model before training; smooth GPS a…

…nd time before trying to merge

spf/dataset/v4_link.sh

@@ -0,0 +1,15 @@
+#!/bin/bash
+
+if [ $# -ne 2 ]; then
+	echo $0 target_link input
+	exit
+fi
+
+target=$1
+input=$2
+input_basename="${input%.*}"
+
+for x in ${input_basename}.*; do 
+	ext="${x##*.}"
+	ln -s $x ${target}.${ext} 
+done

spf/dataset/v4_tx_rx_to_v5.py

@@ -19,6 +19,19 @@ def lat_lon_to_xy(lat, lon, center_lat, center_lon):
     return proj_centered(lon, lat)
 
 
+def smooth_out_timestamps_and_gps(timestamps_and_gps):
+    for prop in ["times", "gps_lats", "gps_longs"]:
+        arr = getattr(timestamps_and_gps, prop)
+        for missing_idx in np.where(arr == 0)[0]:
+            neighbors = []
+            if missing_idx > 0 and arr[missing_idx - 1] != 0:
+                neighbors.append(arr[missing_idx - 1])
+            if missing_idx < len(arr) and arr[missing_idx + 1] != 0:
+                neighbors.append(arr[missing_idx + 1])
+            arr[missing_idx] = sum(neighbors) / len(neighbors)
+    return timestamps_and_gps
+
+
 @dataclass
 class TimestampAndGPS:
     times: np.ndarray  # Array to store timestamps
@@ -167,10 +180,12 @@ def merge_v4rx_v4tx_into_v5(tx_fn, rx_fn, zarr_out_fn, gps_center_long_lat, fix_
     rx_zarr = zarr_open_from_lmdb_store(rx_fn, readahead=True, mode="r")
 
     rx_time_and_gpses = {
-        rx_idx: TimestampAndGPS(
-            times=rx_zarr["receivers"][rx_idx]["gps_timestamp"][:],
-            gps_lats=rx_zarr["receivers"][rx_idx]["gps_lat"][:],
-            gps_longs=rx_zarr["receivers"][rx_idx]["gps_long"][:],
+        rx_idx: smooth_out_timestamps_and_gps(
+            TimestampAndGPS(
+                times=rx_zarr["receivers"][rx_idx]["gps_timestamp"][:],
+                gps_lats=rx_zarr["receivers"][rx_idx]["gps_lat"][:],
+                gps_longs=rx_zarr["receivers"][rx_idx]["gps_long"][:],
+            )
         )
         for rx_idx in ["r0", "r1"]
     }

spf/scripts/train_single_point.py

@@ -1009,6 +1009,9 @@ def train_single_point(args):
 
     m = load_model(config["model"], config["global"]).to(config["optim"]["device"])
 
+    logging.info("MODEL:")
+    logging.info(m)
+
     model_checksum("load_model:", m)
     optimizer, scheduler = load_optimizer(config["optim"], m.parameters())