Spaces:
Runtime error
Runtime error
Delete app
Browse files- app/__init__.py +0 -0
- app/app.py +0 -23
- app/predict.py +0 -154
app/__init__.py
DELETED
File without changes
|
app/app.py
DELETED
@@ -1,23 +0,0 @@
|
|
1 |
-
import gradio as gr
|
2 |
-
import pandas as pd
|
3 |
-
import json
|
4 |
-
from tqdm import tqdm
|
5 |
-
from predict import Preprocess, Facility_Model, obj_Facility_Model, processor
|
6 |
-
|
7 |
-
def predict_facility(data):
|
8 |
-
pred_data = processor.process_tokenizer(data)
|
9 |
-
predictions = obj_Facility_Model.inference(pred_data)
|
10 |
-
return json.loads(predictions)
|
11 |
-
|
12 |
-
|
13 |
-
iface = gr.Interface(
|
14 |
-
fn=predict_facility,
|
15 |
-
inputs="text",
|
16 |
-
outputs="json",
|
17 |
-
title=" Single Facility Prediction",
|
18 |
-
description="Predict the facility based on input data.",
|
19 |
-
#examples=[["kilifi"], ["mombasa"], ["nairobi"]],
|
20 |
-
)
|
21 |
-
|
22 |
-
if __name__ == "__main__":
|
23 |
-
iface.launch()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
app/predict.py
DELETED
@@ -1,154 +0,0 @@
|
|
1 |
-
import os
|
2 |
-
import random
|
3 |
-
import json
|
4 |
-
import numpy as np
|
5 |
-
import torch
|
6 |
-
import heapq
|
7 |
-
import pandas as pd
|
8 |
-
from tqdm import tqdm
|
9 |
-
from transformers import AutoModelForSequenceClassification, AutoTokenizer
|
10 |
-
from torch.utils.data import TensorDataset, DataLoader
|
11 |
-
|
12 |
-
|
13 |
-
class Preprocess:
|
14 |
-
def __init__(self, tokenizer_vocab_path, tokenizer_max_len):
|
15 |
-
self.tokenizer = AutoTokenizer.from_pretrained(tokenizer_vocab_path,
|
16 |
-
use_auth_token='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')
|
17 |
-
self.max_len = tokenizer_max_len
|
18 |
-
|
19 |
-
def clean_text(self, text):
|
20 |
-
text = text.lower()
|
21 |
-
stopwords = ["i", "was", "transferred",
|
22 |
-
"from", "to", "nilienda", "kituo",
|
23 |
-
"cha", "lakini", "saa", "hii", "niko",
|
24 |
-
"at", "nilienda", "nikahudumiwa", "pole",
|
25 |
-
"deliver", "na", "ni", "baada", "ya",
|
26 |
-
"kutumwa", "kutoka", "nilienda",
|
27 |
-
"ndipo", "nikapewa", "hiyo", "lindam ama", "nikawa",
|
28 |
-
"mgonjwa", "nikatibiwa", "in", "had", "a",
|
29 |
-
"visit", "gynaecologist", "ndio",
|
30 |
-
"karibu", "mimi", "niko", "sehemu", "hospitali",
|
31 |
-
"serikali", "delivered", "katika", "kaunti", "kujifungua",
|
32 |
-
"katika", "huko", "nilipoenda", "kwa", "bado", "naedelea",
|
33 |
-
"sija", "maliza", "mwisho",
|
34 |
-
"nilianza", "kliniki", "yangu",
|
35 |
-
"nilianzia", "nilijifungua"]
|
36 |
-
text_single = ' '.join(word for word in text.split() if word not in stopwords)
|
37 |
-
return text_single
|
38 |
-
|
39 |
-
def encode_fn(self, text_single):
|
40 |
-
"""
|
41 |
-
Using tokenizer to preprocess the text
|
42 |
-
example of text_single:'Nairobi Hospital'
|
43 |
-
"""
|
44 |
-
tokenizer = self.tokenizer(text_single,
|
45 |
-
padding=True,
|
46 |
-
truncation=True,
|
47 |
-
max_length=self.max_len,
|
48 |
-
return_tensors='pt'
|
49 |
-
)
|
50 |
-
input_ids = tokenizer['input_ids']
|
51 |
-
attention_mask = tokenizer['attention_mask']
|
52 |
-
return input_ids, attention_mask
|
53 |
-
|
54 |
-
def process_tokenizer(self, text_single):
|
55 |
-
"""
|
56 |
-
Preprocess text and prepare dataloader for a single new sentence
|
57 |
-
"""
|
58 |
-
input_ids, attention_mask = self.encode_fn(text_single)
|
59 |
-
data = TensorDataset(input_ids, attention_mask)
|
60 |
-
return data
|
61 |
-
|
62 |
-
|
63 |
-
class Facility_Model:
|
64 |
-
def __init__(self, facility_model_path: any,
|
65 |
-
max_len: int):
|
66 |
-
self.max_len = max_len
|
67 |
-
self.softmax = torch.nn.Softmax(dim=1)
|
68 |
-
self.gpu = False
|
69 |
-
self.model = AutoModelForSequenceClassification.from_pretrained(facility_model_path,
|
70 |
-
use_auth_token='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')
|
71 |
-
self.model.eval() # set pytorch model for inference mode
|
72 |
-
|
73 |
-
if torch.cuda.device_count() > 1:
|
74 |
-
self.model = torch.nn.DataParallel(self.model)
|
75 |
-
|
76 |
-
if self.gpu:
|
77 |
-
seed = 42
|
78 |
-
random.seed(seed)
|
79 |
-
np.random.seed(seed)
|
80 |
-
torch.manual_seed(seed)
|
81 |
-
torch.cuda.manual_seed_all(seed)
|
82 |
-
torch.backends.cudnn.deterministic = True
|
83 |
-
self.device = torch.device('cuda')
|
84 |
-
else:
|
85 |
-
self.device = 'cpu'
|
86 |
-
|
87 |
-
self.model = self.model.to(self.device)
|
88 |
-
|
89 |
-
def predict_single(self, model, pred_data):
|
90 |
-
"""
|
91 |
-
Model inference for new single sentence
|
92 |
-
"""
|
93 |
-
pred_dataloader = DataLoader(pred_data, batch_size=10, shuffle=False)
|
94 |
-
for i, batch in enumerate(pred_dataloader):
|
95 |
-
with torch.no_grad():
|
96 |
-
outputs = model(input_ids=batch[0].to(self.device),
|
97 |
-
attention_mask=batch[1].to(self.device)
|
98 |
-
)
|
99 |
-
loss, logits = outputs.loss, outputs.logits
|
100 |
-
probability = self.softmax(logits)
|
101 |
-
probability_list = probability.detach().cpu().numpy()
|
102 |
-
return probability_list
|
103 |
-
|
104 |
-
def output_intent_probability(self, pred: any) -> dict:
|
105 |
-
"""
|
106 |
-
convert the model output into a dictionary with all intents and its probability
|
107 |
-
"""
|
108 |
-
output_dict = {}
|
109 |
-
# transform the relation table(between label and intent)
|
110 |
-
path_table = pd.read_csv('/content/drive/MyDrive/dhis14000/dhis_label_relation_14357.csv')
|
111 |
-
|
112 |
-
label_intent_dict = path_table[["label", "corresponding_label"]].set_index("corresponding_label").to_dict()[
|
113 |
-
'label']
|
114 |
-
|
115 |
-
# transform the output into dictionary(between intent and probability)
|
116 |
-
for intent in range(pred.shape[1]):
|
117 |
-
output_dict[label_intent_dict[intent]] = pred[0][intent]
|
118 |
-
|
119 |
-
return output_dict
|
120 |
-
|
121 |
-
def inference(self, prepared_data):
|
122 |
-
"""
|
123 |
-
Make predictions on one new sentence and output a JSON format variable
|
124 |
-
"""
|
125 |
-
temp = []
|
126 |
-
prob_distribution = self.predict_single(self.model, prepared_data)
|
127 |
-
prediction_results = self.output_intent_probability(prob_distribution.astype(float))
|
128 |
-
|
129 |
-
# Filter out predictions containing "dental" or "optical" keywords
|
130 |
-
filtered_results = {intent: prob for intent, prob in prediction_results.items()
|
131 |
-
if
|
132 |
-
"dental" not in intent.lower() and "optical" not in intent.lower() and "eye" not in intent.lower()}
|
133 |
-
|
134 |
-
sorted_pred_intent_results = sorted(filtered_results.items(), key=lambda x: x[1], reverse=True)
|
135 |
-
sorted_pred_intent_results_dict = dict(sorted_pred_intent_results)
|
136 |
-
# Return the top result
|
137 |
-
top_results = dict(list(sorted_pred_intent_results)[:4])
|
138 |
-
temp.append(top_results)
|
139 |
-
final_preds = json.dumps(temp)
|
140 |
-
#final_preds = ', '.join(top_results.keys())
|
141 |
-
#final_preds = ', '.join(top_results)
|
142 |
-
# final_preds = final_preds.replace("'", "")
|
143 |
-
return final_preds
|
144 |
-
|
145 |
-
|
146 |
-
jacaranda_hugging_face_model = "Jacaranda/dhis_14000_600k_Test_Model"
|
147 |
-
|
148 |
-
obj_Facility_Model = Facility_Model(facility_model_path=jacaranda_hugging_face_model,
|
149 |
-
max_len=128
|
150 |
-
)
|
151 |
-
|
152 |
-
processor = Preprocess(tokenizer_vocab_path=jacaranda_hugging_face_model,
|
153 |
-
tokenizer_max_len=128
|
154 |
-
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|