def infer(input: str, model: CharCNN) -> str:
"""
Simple prediction function that takes in a string input, and makes
a prediction on its sentiment using an existing model.
"""
out_classes = {0:"Poor", 1:"Average", 2:"Good"}
input_prep = preprocessing(input, infer=True)
input_quant = TextDataset.quant_text(input_prep)
input_quant = Tensor(input_quant.to_numpy()).unsqueeze(0) # give it batch dimension
prediction = predict(model(input_quant)).item()
output = out_classes[prediction]
return f"Predicted Sentiment: {output}\n"
work_dir = config['work_dir']
mismatch = config['mismatch']
primer_file = config['primer_file']
vector_file = config['vector_file']
logger_level = config['logger_level']
# 生成工作目录，初始化日志，读取样本的barcode和标识
make_output_dir(work_dir)
init_logger(work_dir, logger_level)
barcodes, idents = load_barcode(bc_file)
logging.debug('Loaded barcodes: ' + barcodes.__str__())
try:
for step in step_list:
if step == 1:
# 样本分箱
preprocessing(fw_file, rc_file, work_dir, barcodes, mismatch)
idents = check_pre_file(work_dir, idents)
# 样本read数筛选和格式转换
fastq2fasta(work_dir, idents, is_gzip, maxsize)
logging.info('STEP 1: PREPROCESSING COMPLETED')
elif step == 2:
# 去除引物
suffix = 'fasta.gz' if is_gzip else 'fasta'
idents = check_output_files(work_dir, idents, step, suffix)
remove_primer(work_dir, primer_file, idents, is_gzip)
logging.info('STEP 2: PRIMER SEQUENCES REMOVED')
elif step == 3:
# 去除载体
suffix = 'clean.fasta.gz' if is_gzip else 'clean.fasta'
idents = check_output_files(work_dir, idents, step, suffix)
remove_vector(work_dir, vector_file, idents, is_gzip)
([
'top_' + str(p) + '_building'
for p in [1, 2, 5, 10, 15, 20, 25, 30, 50]
], None), #Leak on CV
('time_stamp', None) #Magic feature
]
# Currently LightGBM core dumps on categorical data, deactivate in the transformer
################ Preprocessing #####################
cache_file = './cache.db'
cv = StratifiedKFold(n_splits=7, shuffle=True, random_state=1337)
folds = list(cv.split(X, y))
# Sorry for the spaghetti code
x_trn, x_val, y_trn, y_val, labelencoder, X_train, X_test, y_train = preprocessing(
X, X_test, y, tr_pipeline, select_feat, folds, cache_file)
############ Train and Validate ####################
print("############ Final Classifier ######################")
clf, metric, n_stop = training_step(x_trn, x_val, y_trn, y_val, X_train,
y_train, folds)
################## Predict #########################
output(X_test, idx_test, clf, labelencoder, n_stop, metric)
with open(
'./out/' + time.strftime("%Y-%m-%d_%H%M-") + '-valid' + str(metric) +
'-features.txt', "w") as text_file:
for item in select_feat:
text_file.write("{}\n".format(item))
if __name__ == "__main__":
all_params = get_default_params()
data_paths = all_params[0]
prep_params = all_params[1]
quant_params = all_params[2]
data_params = all_params[3]
model_params = all_params[4]
pipe_params = all_params[5]
# Setup tensorboard
writer = SummaryWriter(config['log_path'])
print("Preparing Data")
train_df, test_df = preprocessing(data=None, **data_paths, **prep_params)
# Stratify shuffle the test set to ensure each batch during validation will
# contain all of the classes as per proportion in the original test set
test_df = stratified_shuffle(test_df, model_params['batch_size'],
prep_params['label_col'], 10)
# Data loading
train_dl, test_dl = data_loading(train_df, test_df,
prep_params['text_col'],
prep_params['label_col'], **quant_params,
**data_params)
print("Start Training")
# Model instantiation
device = get_default_device()
METADATA_DIR = 'data/csv/'
METADATA = 'df_crop_' + OBJECT_NAME
df_metadata = pd.read_csv(path.join(sys.path[1], METADATA_DIR, METADATA))
image_names = df_metadata['filename'].tolist()
result = []
with ProgressBar(max_value=len(image_names)) as bar:
bar.update(0)
for i, image_name in enumerate(image_names):
image = cv2.imread(path.join(sys.path[1], IMAGES_DIR, image_name))
image = preprocessing(image=image)
predicted_text = recognize_text(image=image)
target_text = df_metadata[df_metadata['filename'] ==
image_name]['attribute'].tolist()[0]
score = text_recognition_score(target_text=target_text,
predicted_text=predicted_text)
result.append(
[image_name, target_text, predicted_text, score[0], score[1]])
bar.update(i)
df_result = pd.DataFrame(data=result,
columns=[