diff --git a/mlair/run_modules/experiment_setup.py b/mlair/run_modules/experiment_setup.py
index 8bbcfddf741c657af0d0e26f7c3937058e83f821..c12664e356d30bfc23333c3888d750bf245f4564 100644
--- a/mlair/run_modules/experiment_setup.py
+++ b/mlair/run_modules/experiment_setup.py
@@ -244,7 +244,7 @@ class ExperimentSetup(RunEnvironment):
do_uncertainty_estimate: bool = None, do_bias_free_evaluation: bool = None,
model_display_name: str = None, transformation_file: str = None,
calculate_fresh_transformation: bool = None, snapshot_load_path: str = None,
- create_snapshot: bool = None, snapshot_path: str = None, **kwargs):
+ create_snapshot: bool = None, snapshot_path: str = None, model_path: str = None, **kwargs):
# create run framework
super().__init__()
@@ -299,6 +299,7 @@ class ExperimentSetup(RunEnvironment):
self._set_param("batch_path", batch_path, default=os.path.join(experiment_path, "batch_data"))
# set model path
+ self._set_param("model_load_path", model_path)
self._set_param("model_path", None, os.path.join(experiment_path, "model"))
path_config.check_path_and_create(self.data_store.get("model_path"))
diff --git a/mlair/run_modules/model_setup.py b/mlair/run_modules/model_setup.py
index efeff06260d1df41d4a83592c225f316574e77eb..3164db789e79a92364b300b5d8724861031afeb8 100644
--- a/mlair/run_modules/model_setup.py
+++ b/mlair/run_modules/model_setup.py
@@ -6,6 +6,8 @@ __date__ = '2019-12-02'
import logging
import os
import re
+import shutil
+
from dill.source import getsource
import tensorflow.keras as keras
@@ -58,15 +60,16 @@ class ModelSetup(RunEnvironment):
"""Initialise and run model setup."""
super().__init__()
self.model = None
- exp_name = self.data_store.get("experiment_name")
- self.path = self.data_store.get("model_path")
self.scope = "model"
- path = os.path.join(self.path, f"{exp_name}_%s")
- self.model_name = path % "%s.h5"
- self.checkpoint_name = path % "model-best.h5"
- self.callbacks_name = path % "model-best-callbacks-%s.pickle"
self._train_model = self.data_store.get("train_model")
self._create_new_model = self.data_store.get("create_new_model")
+ self.path = self.data_store.get("model_path")
+ self.model_display_name = self.data_store.get_default("model_display_name", default=None)
+ self.model_load_path = None
+ path = self._set_model_path()
+ self.model_path = path % "%s.h5"
+ self.checkpoint_name = path % "_model-best.h5"
+ self.callbacks_name = path % "_model-best-callbacks-%s.pickle"
self._run()
def _run(self):
@@ -96,6 +99,20 @@ class ModelSetup(RunEnvironment):
# report settings
self.report_model()
+ def _set_model_path(self):
+ exp_name = self.data_store.get("experiment_name")
+ self.model_load_path = self.data_store.get_default("model_load_path", default=None)
+ if self.model_load_path is not None:
+ if not self.model_load_path.endswith(".h5"):
+ raise FileNotFoundError(f"When providing external models, you need to provide full path including the "
+ f".h5 file. Given path is not valid: {self.model_load_path}")
+ if any([self._train_model, self._create_new_model]):
+ raise ValueError(f"Providing `model_path` along with parameters train_model={self._train_model} and "
+ f"create_new_model={self._create_new_model} is not possible. Either set both "
+ f"parameters to False or remove `model_path` parameter. Given was: model_path = "
+ f"{self.model_load_path}")
+ return os.path.join(self.path, f"{self.model_display_name or exp_name}%s")
+
def _set_shapes(self):
"""Set input and output shapes from train collection."""
shape = list(map(lambda x: x.shape[1:], self.data_store.get("data_collection", "train")[0].get_X()))
@@ -159,11 +176,18 @@ class ModelSetup(RunEnvironment):
# store callbacks
self.data_store.set("callbacks", callbacks, self.scope)
+ def copy_model(self):
+ """Copy external model to internal experiment structure."""
+ if self.model_load_path is not None:
+ logging.info(f"Copy external model file: {self.model_load_path} -> {self.model_path}")
+ shutil.copyfile(self.model_load_path, self.model_path)
+
def load_model(self):
"""Try to load model from disk or skip if not possible."""
+ self.copy_model()
try:
- self.model.load_model(self.model_name)
- logging.info(f"reload model {self.model_name} from disk ...")
+ self.model.load_model(self.model_path)
+ logging.info(f"reload model {self.model_path} from disk ...")
except OSError:
logging.info('no local model to load...')
@@ -195,14 +219,16 @@ class ModelSetup(RunEnvironment):
"""Load all model settings and store in data store."""
model_settings = self.model.get_settings()
self.data_store.set_from_dict(model_settings, self.scope, log=True)
- self.model_name = self.model_name % self.data_store.get_default("model_name", self.scope, "my_model")
- self.data_store.set("model_name", self.model_name, self.scope)
+ generic_model_name = self.data_store.get_default("model_name", self.scope, "my_model")
+ model_annotation = generic_model_name if self.model_display_name is None else ""
+ self.model_path = self.model_path % model_annotation
+ self.data_store.set("model_name", self.model_path, self.scope)
def plot_model(self): # pragma: no cover
"""Plot model architecture as `<model_name>.pdf`."""
try:
with tf.device("/cpu:0"):
- file_name = f"{self.model_name.rsplit('.', 1)[0]}.pdf"
+ file_name = f"{self.model_path.rsplit('.', 1)[0]}.pdf"
keras.utils.plot_model(self.model, to_file=file_name, show_shapes=True, show_layer_names=True)
except Exception as e:
logging.info(f"Can not plot model due to: {e}")