def generate_testdata(self):
run_names = ["run_1", "run_2"]
ground_truth = {
"run_1": [
["Description", "A", "B"],
["name_1", 1.0, -1.0],
["name_2", -0.5, 0.5],
],
"run_2": [
["Description", "A", "B"],
["name_1", 1.0, -1.0],
["name_2", -0.5, np.nan],
],
}
embedding_ground_truth = {
"run_1": [[1.0, 0.5], [-0.5, 0.5]],
"run_2": [[1.0, 0.5], [-0.5, 0.5]],
}
for run_name in run_names:
subdir = os.path.join(self.logdir, run_name)
writer = tf.compat.v2.summary.create_file_writer(subdir)
data = ground_truth[run_name]
python_result = []
python_annotations = []
python_classes = []
for row_index, row in enumerate(data):
if row_index > 0:
python_result.append([])
for col_index, column in enumerate(row):
if row_index == 0:
if col_index > 0:
python_classes.append(column)
else:
if col_index == 0:
python_annotations.append(column)
else:
python_result[len(python_result) -
1].append(column)
with writer.as_default():
tensor_result = tf.convert_to_tensor(python_result)
tensor_annotations = tf.convert_to_tensor(python_annotations)
tensor_classes = tf.convert_to_tensor(python_classes)
tensor_embeddings = tf.convert_to_tensor(
embedding_ground_truth[run_name])
summary.npmi_values(tensor_result, 1)
summary.npmi_annotations(tensor_annotations, 1)
summary.npmi_metrics(tensor_classes, 1)
summary.npmi_embeddings(tensor_embeddings, 1)
writer.close()
def setup_run(logdir, run_name):
# Getting a list of dummy annotations from a separate file.
python_annotations = npmi_demo_data.MOUNTAINS
# Writing out random nPMI values.
python_classes = ["nPMI@SKIABLE", "nPMI@NOT_SKIABLE"]
python_result = np.random.rand(len(python_annotations), len(python_classes))
python_result = python_result * 2.0 - 1.0
logdir = os.path.join(logdir, run_name)
os.makedirs(logdir)
writer = tf.summary.create_file_writer(logdir)
with writer.as_default():
tensor_result = tf.convert_to_tensor(python_result)
tensor_annotations = tf.convert_to_tensor(python_annotations)
tensor_metrics = tf.convert_to_tensor(python_classes)
summary.npmi_values(tensor_result, 1)
summary.npmi_annotations(tensor_annotations, 1)
summary.npmi_metrics(tensor_metrics, 1)
writer.close()
def convert_file(file_path):
"""Converts a csv file to a logfile readable by the nPMI plugin.
Args:
file_path: the path to the csv file to be converted
"""
if not os.path.exists(file_path):
print("No such file found. Conversion failed.")
return
metrics = []
annotations = []
values = []
with open(file_path) as csv_file:
csv_reader = csv.reader(csv_file)
metrics = next(csv_reader)[1:]
if not contains_npmi_metric(metrics):
print(
"No metric is prefixed with nPMI@ or nPMI_diff@. No export generated."
)
return
for row in csv_reader:
annotations.append(row[0])
values.append(row[1:])
values = np.array(values).astype(np.float)
writer = tf.summary.create_file_writer(os.path.dirname(file_path))
with writer.as_default():
tensor_result = tf.convert_to_tensor(values)
tensor_annotations = tf.convert_to_tensor(annotations)
tensor_metrics = tf.convert_to_tensor(metrics)
summary.npmi_values(tensor_result, 1)
summary.npmi_annotations(tensor_annotations, 1)
summary.npmi_metrics(tensor_metrics, 1)
writer.close()
print("Successfuly saved converted output to %s" %
os.path.dirname(file_path))
def write_results(self, tensor):
writer = tf2.summary.create_file_writer(self.get_temp_dir())
with writer.as_default():
summary.npmi_annotations(tensor, 1)
writer.close()
