From d12bb23c5e1a9b8a6bcca44ccc89fd42ab4bc1e2 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Fran=C3=A7ois=20Laurent?= <francois.laurent@posteo.net>
Date: Mon, 1 Aug 2022 09:48:08 +0200
Subject: [PATCH] pretrained_model_instance arg
---
.../default}/autoencoder_config.json | 0
.../default}/best_validated_encoder.pt | Bin
src/maggotuba/models/train_model.py | 10 +++++-----
3 files changed, 5 insertions(+), 5 deletions(-)
rename {models => pretrained_models/default}/autoencoder_config.json (100%)
rename {models => pretrained_models/default}/best_validated_encoder.pt (100%)
diff --git a/models/autoencoder_config.json b/pretrained_models/default/autoencoder_config.json
similarity index 100%
rename from models/autoencoder_config.json
rename to pretrained_models/default/autoencoder_config.json
diff --git a/models/best_validated_encoder.pt b/pretrained_models/default/best_validated_encoder.pt
similarity index 100%
rename from models/best_validated_encoder.pt
rename to pretrained_models/default/best_validated_encoder.pt
diff --git a/src/maggotuba/models/train_model.py b/src/maggotuba/models/train_model.py
index 0401cd7..5527b97 100644
--- a/src/maggotuba/models/train_model.py
+++ b/src/maggotuba/models/train_model.py
@@ -7,7 +7,7 @@ import torch
import os
import glob
-def train_model(backend):
+def train_model(backend, pretrained_model_instance="default"):
# make_dataset generated or moved the larva_dataset file into data/interim/{instance}/
#larva_dataset_file = backend.list_interim_files("larva_dataset_*.hdf5") # recursive
larva_dataset_file = glob.glob(str(backend.interim_data_dir() / "larva_dataset_*.hdf5")) # not recursive (faster)
@@ -16,7 +16,7 @@ def train_model(backend):
nlabels = len(dataset.labels)
assert 0 < nlabels
# copy the pretrained model into the model instance directory
- pretrained_autoencoder_dir = backend.model_dir() / ".."
+ pretrained_autoencoder_dir = backend.model_dir() / "pretrained_models" / pretrained_model_instance
config_file = None
for file in pretrained_autoencoder_dir.iterdir():
if not file.is_file():
@@ -28,9 +28,9 @@ def train_model(backend):
dir = backend.model_dir().relative_to(backend.project_dir)
config["log_dir"] = str(dir)
# optional updates?
- config["project_dir"] = config["exp_folder"] = str(dir)
- config["exp_name"] = backend.model_instance
- config["config"] = str(dir / os.path.basename(config["config"]))
+ #config["project_dir"] = config["exp_folder"] = str(dir)
+ #config["exp_name"] = backend.model_instance
+ #config["config"] = str(dir / os.path.basename(config["config"]))
with open(str(dst), "w") as f:
json.dump(config, f, indent=2)
assert config_file is None
--
GitLab