def read_dataset(config):
# Read metadata and split dataset in training and validation
metadata = deepprofiler.dataset.metadata.Metadata(config["paths"]["index"],
dtype=None)
if config["prepare"]["compression"]["implement"]:
metadata.data.replace({
'.tiff': '.png',
'.tif': '.png'
},
inplace=True,
regex=True)
# Add outlines if specified
outlines = None
if "outlines" in config["prepare"].keys(
) and config["prepare"]["outlines"] != "":
df = pd.read_csv(config["paths"]["metadata"] + "/outlines.csv")
metadata.mergeOutlines(df)
outlines = config["paths"]["root"] + "inputs/outlines/"
print(metadata.data.info())
# Split training data
split_field = config["train"]["partition"]["split_field"]
trainingFilter = lambda df: df[split_field].isin(config["train"][
"partition"]["training_values"])
validationFilter = lambda df: df[split_field].isin(config["train"][
"partition"]["validation_values"])
metadata.splitMetadata(trainingFilter, validationFilter)
# Create a dataset
keyGen = lambda r: "{}/{}-{}".format(r["Metadata_Plate"], r[
"Metadata_Well"], r["Metadata_Site"])
dset = ImageDataset(metadata, config["dataset"]["metadata"]["label_field"],
config["dataset"]["images"]["channels"],
config["paths"]["images"], keyGen, config)
# Add training targets
for t in config["train"]["partition"]["targets"]:
new_target = deepprofiler.dataset.target.MetadataColumnTarget(
t, metadata.data[t].unique())
dset.add_target(new_target)
# Activate outlines for masking if needed
if config["dataset"]["locations"]["mask_objects"]:
dset.outlines = outlines
dset.prepare_training_locations()
return dset
def read_dataset(config):
# Read metadata and split dataset in training and validation
metadata = deepprofiler.dataset.metadata.Metadata(config["paths"]["index"], dtype=None)
# Add outlines if specified
outlines = None
if "outlines" in config["prepare"].keys() and config["prepare"]["outlines"] != "":
df = pd.read_csv(config["paths"]["locations"] + "/outlines.csv")
metadata.mergeOutlines(df)
outlines = config["paths"]["locations"]
print(metadata.data.info())
# Split training data
split_field = config["train"]['dset']["split_field"]
trainingFilter = lambda df: df[split_field].isin(config['train']["dset"]["training_values"])
validationFilter = lambda df: df[split_field].isin(config['train']["dset"]["validation_values"])
metadata.splitMetadata(trainingFilter, validationFilter)
# Create a dataset
keyGen = lambda r: "{}/{}-{}".format(r["Metadata_Plate"], r["Metadata_Well"], r["Metadata_Site"])
dset = ImageDataset(
metadata,
config['train']["sampling"]["field"],
config['prepare']["images"]["channels"],
config["paths"]["images"],
keyGen
)
# Add training targets
for t in config['train']["dset"]["targets"]:
new_target = deepprofiler.dataset.target.MetadataColumnTarget(t, metadata.data[t].unique())
dset.add_target(new_target)
# Activate outlines for masking if needed
if config['train']["dset"]["mask_objects"]:
dset.outlines = outlines
return dset
def test_split_metadata(metadata, dataframe, out_dir):
metadata.loadSingle(os.path.join(out_dir, 'test.csv'), 'default', int)
train_rule = lambda data: data['Metadata_Plate'] < 50
val_rule = lambda data: data['Metadata_Plate'] >= 50
metadata.splitMetadata(train_rule, val_rule)
assert len(metadata.train) + len(metadata.val) == len(metadata.data)
