def _load_state_vars(state, gtruth_df, median_win=None):
pred_df = gtruth_df.copy()
# Define dataloader
many_hot_encoder = ManyHotEncoder.load_state_dict(state["many_hot_encoder"])
scaler = _load_scaler(state)
crnn = _load_crnn(state)
# Note, need to unsqueeze axis 1
transforms_valid = get_transforms(cfg.max_frames, scaler=scaler, add_axis=1)
# Note, no dataloader here
strong_dataload = DataLoadDf(pred_df, many_hot_encoder.encode_strong_df, transforms_valid, return_indexes=True)
pooling_time_ratio = state["pooling_time_ratio"]
many_hot_encoder = ManyHotEncoder.load_state_dict(state["many_hot_encoder"])
if median_win is None:
median_win = state["median_window"]
return {
"model": crnn,
"dataload": strong_dataload,
"pooling_time_ratio": pooling_time_ratio,
"many_hot_encoder": many_hot_encoder,
"median_window": median_win
}
if save_best_cb.apply(valid_synth_f1):
model_fname = os.path.join(saved_model_dir, "baseline_best")
torch.save(state, model_fname)
results.loc[epoch, "global_valid"] = valid_synth_f1
results.loc[epoch, "loss"] = loss_value.item()
results.loc[epoch, "valid_synth_f1"] = valid_synth_f1
if cfg.early_stopping:
if early_stopping_call.apply(valid_synth_f1):
logger.warn("EARLY STOPPING")
break
if cfg.save_best:
model_fname = os.path.join(saved_model_dir, "baseline_best")
state = torch.load(model_fname)
crnn = _load_crnn(state)
logger.info(f"testing model: {model_fname}, epoch: {state['epoch']}")
else:
logger.info("testing model of last epoch: {}".format(cfg.n_epoch))
results_df = pd.DataFrame(results).to_csv(os.path.join(
saved_pred_dir, "results.tsv"),
sep="\t",
index=False,
float_format="%.4f")
# ##############
# Validation
# ##############
crnn.eval()
transforms_valid = get_transforms(cfg.max_frames, scaler, add_axis_conv)
predicitons_fname = os.path.join(saved_pred_dir, "baseline_validation.tsv")