add arg to adjust learning rate

This commit is contained in:
Starbeamrainbowlabs 2022-11-29 20:55:00 +00:00
parent 8e23e9d341
commit c384d55dff
Signed by: sbrl
GPG key ID: 1BE5172E637709C2
2 changed files with 18 additions and 8 deletions

View file

@ -8,15 +8,18 @@ from .components.convnext_inverse import do_convnext_inverse
from .components.LayerStack2Image import LayerStack2Image
from .components.LossCrossentropy import LossCrossentropy
def model_rainfallwater_mono(metadata, shape_water_out, model_arch_enc="convnext_xtiny", model_arch_dec="convnext_i_xtiny", feature_dim=512, batch_size=64, water_bins=2):
def model_rainfallwater_mono(metadata, shape_water_out, model_arch_enc="convnext_xtiny", model_arch_dec="convnext_i_xtiny", feature_dim=512, batch_size=64, water_bins=2, learning_rate=None):
"""Makes a new rainfall / waterdepth mono model.
Args:
metadata (dict): A dictionary of metadata about the dataset to use to build the model with.
shape_water_out (int[]): The width and height (in that order) that should dictate the output shape of the segmentation head. CURRENTLY NOT USED.
model_arch (str, optional): The architecture code for the underlying (inverted) ConvNeXt model. Defaults to "convnext_i_xtiny".
feature_dim (int, optiona): The size of the bottleneck. Defaults to 512.
model_arch_enc (str, optional): The architecture code for the underlying (inverted) ConvNeXt model for the encoder. Defaults to "convnext_xtiny".
model_arch_dec (str, optional): The architecture code for the underlying (inverted) ConvNeXt model for the decoder. Defaults to "convnext_i_xtiny".
batch_size (int, optional): The batch size. Reduce to save memory. Defaults to 64.
water_bins (int, optional): The number of classes that the water depth output oft he segmentation head should be binned into. Defaults to 2.
learning_rate (float, optional): The (initial) learning rate. YOU DO NOT USUALLY NEED TO CHANGE THIS. For experimental purposes only. Defaults to None, which means it will be determined automatically.
Returns:
tf.keras.Model: The new model, freshly compiled for your convenience! :D
@ -70,8 +73,11 @@ def model_rainfallwater_mono(metadata, shape_water_out, model_arch_enc="convnext
outputs = layer_next
)
optimizer = "Adam"
if learning_rate is not None:
optimizer = tf.keras.optimizers.Adam(learning_rate=learning_rate)
model.compile(
optimizer="Adam",
optimizer=optimizer,
loss=LossCrossentropy(batch_size=batch_size),
# loss=tf.keras.losses.CategoricalCrossentropy(),
metrics=[tf.keras.metrics.CategoricalAccuracy()]

View file

@ -20,6 +20,7 @@ def parse_args():
parser.add_argument("--bottleneck", help="The size of the bottleneck [default: 512].", type=int)
parser.add_argument("--arch-enc", help="Next of the underlying encoder convnext model to use [default: convnext_xtiny].")
parser.add_argument("--arch-dec", help="Next of the underlying decoder convnext model to use [default: convnext_i_xtiny].")
parser.add_argument("--learning-rate", help="The initial learning rate. YOU DO NOT USUALLY NEED TO CHANGE THIS. For experimental use only [default: determined automatically].", type=int)
return parser
@ -43,6 +44,8 @@ def run(args):
args.arch_enc = "convnext_xtiny"
if (not hasattr(args, "arch_dec")) or args.arch_dec == None:
args.arch_dec = "convnext_i_xtiny"
if (not hasattr(args, "learning_rate")) or args.learning_rate == None:
args.learning_rate = None
# TODO: Validate args here.
@ -72,6 +75,7 @@ def run(args):
feature_dim = args.bottleneck,
model_arch_enc = args.arch_enc,
model_arch_dec = args.arch_dec,
learning_rate = args.learning_rate,
metadata = read_metadata(args.input),
shape_water_out=[ args.water_size, args.water_size ], # The DESIRED output shape. the actual data will be cropped to match this.