gistfile1.diff

commit 3d4383d88346b60d669ce72fa4e2a6d4a5e4861d
Author: Viswanath Sivakumar <viswanath@fb.com>
Date:   Tue Aug 25 10:49:59 2020 -0400

    TODO nocommit

diff --git a/platform/pybmi/pybmi/modeling/keystrokes/corpus.py b/platform/pybmi/pybmi/modeling/keystrokes/corpus.py
index 30b4b8fd9..9f1bd7431 100644
--- a/platform/pybmi/pybmi/modeling/keystrokes/corpus.py
+++ b/platform/pybmi/pybmi/modeling/keystrokes/corpus.py
@@ -89,6 +89,7 @@ def get_single_user_split(corpus: pd.DataFrame,
                           seed=None) -> Tuple[pd.DataFrame, pd.DataFrame]:
     """Get a train/validation split for a single-user experiment"""
     corpus = corpus[corpus["user"] == user]
+    # TODO slog corpus = corpus[corpus['condition'].str.contains('off_keyboard')]
     val = corpus \
         .pipe(get_possible_validation_sets) \
         .pipe(stratified_sample, n=1, seed=seed)
diff --git a/scripts/experiments/typing/common.py b/scripts/experiments/typing/common.py
index 81f6d0e2d..a42fc4016 100644
--- a/scripts/experiments/typing/common.py
+++ b/scripts/experiments/typing/common.py
@@ -7,4 +7,5 @@ def parse_args():
         '--arch', type=str, choices=['lstm', 'w2l', 'tds'], default='lstm')
     parser.add_argument('--batch-size', type=int, default=16)
     parser.add_argument('--seed', type=int, default=1701)
+    parser.add_argument('--pp-stride', type=int, default=80)
     return parser.parse_args()
diff --git a/scripts/experiments/typing/generate_cross_session.py b/scripts/experiments/typing/generate_cross_session.py
index f3d95d27e..bee8af472 100644
--- a/scripts/experiments/typing/generate_cross_session.py
+++ b/scripts/experiments/typing/generate_cross_session.py
@@ -19,7 +19,8 @@ from common import parse_args
 args = parse_args()
 
 corpus = k.get_typing_corpus().loc[lambda df: df["ok?"] == "Y"]
-loader_config = k.config_utils.get_cospectrum_loader_config()
+loader_config = k.config_utils.get_cospectrum_loader_config(
+    pp_stride=args.pp_stride)
 
 val_metrics = [
     "character_error_rate",
@@ -31,12 +32,13 @@ val_metrics = [
 # Grid search hyperparameter options
 training_fixed = {
     "batch_size": [args.batch_size],
-    "learning_rate": np.logspace(-3.75, -3.75, 1).tolist(),
-    "max_epochs": [200],
+    "learning_rate": np.logspace(-2.5, -2.5, 1).tolist(),
+    "max_epochs": [300],
     "patience": [30],
-    "scheduler_step_size": [200],
-    "scheduler_gamma": [1.0],
+    "scheduler_step_size": [100],
+    "scheduler_gamma": [0.1],
     "window_length": [200],
+    "scheduler_warmup_epochs": [10],
     "temporal_jitter": [3],
     "band_rotations": [(-1, 0, 1)],
     "val_metrics": [val_metrics],
@@ -45,6 +47,7 @@ training_fixed = {
     "n_f_masks": [2],
     "freq_mask_param": [2],
     "time_mask_param": [20],
+    "seed": [args.seed],
 }
 
 if args.arch == 'lstm':
@@ -57,7 +60,6 @@ elif args.arch == 'tds':
         "L1": [3],
         "L2": [1],
         "kw": [11],
-        "rpad": [-1]
     }
 
 hyperparameter_fixed = {