if not categorize:
profile.categorize = None
profile.inputdir(jobfile.rootdir)
if graph:
for cpu in cpus:
profile.cpu = cpu
if funcdata:
name = 'funcstacks%d' % cpu
else:
name = 'stacks%d' % cpu
output = StatOutput(jobfile, info=profile)
output.xlabel = 'System Configuration'
output.ylabel = '% CPU utilization'
output.stat = name
output.graph(name, graph)
if dodot:
for cpu in cpus:
profile.cpu = cpu
profile.write_dot(jobfile=jobfile, threshold=threshold)
if textout:
for cpu in cpus:
profile.cpu = cpu
profile.write_txt(jobfile=jobfile)
if not graph and not textout and not dodot:
for cpu in cpus:
if not categorize:
profile.categorize = None
def run_evaluation(embeddings, report = False, graphs = False, classifier_args = [],
classifier = 'knn', selected_feature_areas = None,
selected_features = None, folder = 'output'):
'''
This is the main function. It runs the evaluation on the embeddings, and
returns the results and averages.
Parameters
----------
embeddings : dictionary (string: list (int/float))
A dictionary with the embeddings to evaluate. The keys must be the
language identifier, and the values are the language embeddings
report : bool
Whether to write text reports with the results and averages in the
output folder
graphs : bool
Whether to create graphs of the results in the output folder. This
includes bar graphs, maps, and count graphs. This can take several
minutes to complete
classifier_args : list (int)
A list of arguments to the classifier. If using k-nearest neighbors,
it's k, if using multilayer perceprton, it's the layer sizes, where
the length of the list indicates the number of layers
classifier : string
Which classifier to use. Possible values are knn for k-nearest neighbors,
mlp for multilayer perceptron, and svm for support vector machine.
selected_feature_areas : list (int)
List of feature areas to evalute for. Leave out for all, or use ant of the
following integers
0 - None (add individual features with selected features)
1 - Phonology
2 - Morphology
3 - Nominal Categories
4 - Nominal Syntax
5 - Verbal Categories
6 - Word Order
7 - Simple Clauses
8 - Complex Sentences
9 - Lexicon
10 - Sign Languages
11 - Other
12 - Word Order
selected_features : list (string)
Add individual features not included by selected_feature_areas
folder : string
The output folder where to place the text reports and graphs.
Returns
--------
Tuple (True, (dictionary, dictionary)) or
Tuple (False, string)
If success, it results a tuple where the first value is true, and the
second value is a tuple with a the results and the averages.
If failes, the first value is false, and the second value is an error
message.
'''
if not verify_embeddings(embeddings):
return (False, "Wrong embeddings format. Format must be a dictionary where the keys are language IDs (ISO 639-3) and values are the language embeddings")
if folder and not folder.endswith("/"):
folder += "/"
if folder and not os.path.isdir(folder):
return (False, "Could not find the path {}".format(folder))
print("Starting")
with open('language.csv', 'rt', encoding='utf8') as file:
reader = csv.reader(file)
languages = list(reader)
headers = languages.pop(0)
# Remove languages we do not have embeddings for
languages = [lang for lang in languages if lang[1] in embeddings]
with open('feature_areas.csv', 'rt', encoding='utf8') as file:
reader = csv.reader(file)
feature_areas = {rows[0]:rows[1] for rows in reader}
included_features = get_included_features(feature_areas, selected_feature_areas,
selected_features)
if classifier == 'knn':
classifier = neighbors.KNeighborsClassifier(classifier_args[0]) if classifier_args \
else neighbors.KNeighborsClassifier(10)
elif classifier == 'mlp':
from sklearn.neural_network import MLPClassifier
classifier = MLPClassifier(hidden_layer_sizes=tuple(classifier_args)) if classifier_args \
else MLPClassifier(hidden_layer_sizes=(50, 50, 50))
else:
classifier = svm.SVC()
print("Evaluating embeddings")
results = evaluate(languages, headers, embeddings, included_features, classifier)
print("Calculating averages")
averages = calculate_averages(results)
if report:
print("Writing text reports")
write_report(folder, results, averages)
if graphs:
print("Creating bar graphs")
graph(folder, results, averages)
print("Creating maps")
maps(folder, averages, languages)
print("Creating count graphs")
count_score_graph(folder, averages, languages)
print("Finished\n")
return (True, (results, averages))
if not categorize:
profile.categorize = None
profile.inputdir(jobfile.rootdir)
if graph:
for cpu in cpus:
profile.cpu = cpu
if funcdata:
name = 'funcstacks%d' % cpu
else:
name = 'stacks%d' % cpu
output = StatOutput(jobfile, info=profile)
output.xlabel = 'System Configuration'
output.ylabel = '% CPU utilization'
output.stat = name
output.graph(name, graph)
if dodot:
for cpu in cpus:
profile.cpu = cpu
profile.write_dot(jobfile=jobfile, threshold=threshold)
if textout:
for cpu in cpus:
profile.cpu = cpu
profile.write_txt(jobfile=jobfile)
if not graph and not textout and not dodot:
for cpu in cpus:
if not categorize:
profile.categorize = None
import input
import process
import output
#main
process.value(x)
process.constant(x)
process.Numpy(x)
output.result(f)
if n == 0:
output.graph(f)
elif n == 1:
output.approximation(f)