From 980b3d1fb6b4b343e1607aff2877684ae5a53ae4 Mon Sep 17 00:00:00 2001
From: Michael <m.langguth@fz-juelich.de>
Date: Mon, 21 Jun 2021 11:37:56 +0200
Subject: [PATCH] Reduction of default hyperparameters to the relevant ones for
the dataset and corrected usage of Hparams-handling in era5_dataset.py.
Conflicts:
video_prediction_tools/model_modules/video_prediction/datasets/era5_dataset.py
---
.../video_prediction/datasets/era5_dataset.py | 14 +++++++-------
1 file changed, 7 insertions(+), 7 deletions(-)
diff --git a/video_prediction_tools/model_modules/video_prediction/datasets/era5_dataset.py b/video_prediction_tools/model_modules/video_prediction/datasets/era5_dataset.py
index ce62965a..664a817a 100644
--- a/video_prediction_tools/model_modules/video_prediction/datasets/era5_dataset.py
+++ b/video_prediction_tools/model_modules/video_prediction/datasets/era5_dataset.py
@@ -10,7 +10,7 @@ import tensorflow as tf
from collections import OrderedDict
from tensorflow.contrib.training import HParams
from google.protobuf.json_format import MessageToDict
-
+from general_utils import reduce_dict
class ERA5Dataset(object):
@@ -58,20 +58,18 @@ class ERA5Dataset(object):
def get_default_hparams_dict(self):
"""
- The function that contains default hparams
+ Provide dictionary containing default hyperparameters for the dataset
Returns:
- A dict with the following hyperparameters.
- context_frames : the number of ground-truth frames to pass in at start.
+ A dict with the following hyperparameters relevant for the dataset.
+ context_frames : the number of ground-truth frames to pass in at start
+ batch_size : number of training examples per mini-batch
max_epochs : the number of epochs to train model
- lr : learning rate
loss_fun : the loss function
"""
hparams = dict(
context_frames=10,
max_epochs = 20,
batch_size = 40,
- lr = 0.001,
- loss_fun = "rmse",
shuffle_on_val= True,
)
return hparams
@@ -89,7 +87,9 @@ class ERA5Dataset(object):
"""
Parse the hparams setting to ovoerride the default ones
"""
+ self.hparams_dict = reduce_dict(self.hparams_dict, self.get_default_hparams().values())
parsed_hparams = self.get_default_hparams().override_from_dict(self.hparams_dict or {})
+
return parsed_hparams
--
GitLab