Spaces:
Paused
Paused
Commit
β’
22ad617
1
Parent(s):
32ab113
ported everything into dictionary postgres
Browse files- category_mapper.py +9 -7
- chatgpt_audit.py +48 -49
- db/db_utils.py +13 -0
- dictionary/additions.csv +0 -32
- dictionary/dictionary.csv +0 -0
- {multi-item-experiments β old_experiments/multi-item-experiments}/classification_results2.csv +0 -0
- {multi-item-experiments β old_experiments/multi-item-experiments}/multifood2.py +0 -0
- {multi-item-experiments β old_experiments/multi-item-experiments}/multifood_viz.py +0 -0
- preseed.py β old_experiments/preseed.py +0 -0
- playground.py +1 -108
- similarity_fast.py +3 -6
- similarity_slow.py +3 -7
category_mapper.py
CHANGED
@@ -5,6 +5,7 @@ import pandas as pd
|
|
5 |
from tqdm import tqdm
|
6 |
from openai import OpenAI
|
7 |
from dotenv import load_dotenv
|
|
|
8 |
|
9 |
|
10 |
load_dotenv()
|
@@ -12,6 +13,9 @@ load_dotenv()
|
|
12 |
api_key = os.getenv("OPENAI_API_KEY")
|
13 |
client = OpenAI(api_key=api_key)
|
14 |
|
|
|
|
|
|
|
15 |
# Load your Excel file
|
16 |
file_path = './dictionary/final_corrected_wweia_food_category_complete - final_corrected_wweia_food_category_complete.csv'
|
17 |
spreadsheet = pd.read_csv(file_path)
|
@@ -48,11 +52,12 @@ def parse_response(response):
|
|
48 |
return None
|
49 |
|
50 |
# open up the current dictionary csv file
|
51 |
-
|
52 |
-
|
53 |
|
54 |
-
for
|
55 |
# Get the food item and category
|
|
|
56 |
food_item = row['description']
|
57 |
category = row['food_category']
|
58 |
|
@@ -68,8 +73,5 @@ for index, row in tqdm(df_dictionary.iterrows(), desc="Processing input words"):
|
|
68 |
print(f"A: '{best_category}'")
|
69 |
print()
|
70 |
|
71 |
-
# Update the dictionary.csv file by adding the best category to the wweia_category column
|
72 |
if best_category:
|
73 |
-
|
74 |
-
|
75 |
-
df_dictionary.to_csv(csv_file_path, index=False)
|
|
|
5 |
from tqdm import tqdm
|
6 |
from openai import OpenAI
|
7 |
from dotenv import load_dotenv
|
8 |
+
from db.db_utils import get_connection
|
9 |
|
10 |
|
11 |
load_dotenv()
|
|
|
13 |
api_key = os.getenv("OPENAI_API_KEY")
|
14 |
client = OpenAI(api_key=api_key)
|
15 |
|
16 |
+
db_conn = get_connection()
|
17 |
+
db_cursor = db_conn.cursor()
|
18 |
+
|
19 |
# Load your Excel file
|
20 |
file_path = './dictionary/final_corrected_wweia_food_category_complete - final_corrected_wweia_food_category_complete.csv'
|
21 |
spreadsheet = pd.read_csv(file_path)
|
|
|
52 |
return None
|
53 |
|
54 |
# open up the current dictionary csv file
|
55 |
+
db_cursor.execute('SELECT * FROM dictionary where wweia_category is null')
|
56 |
+
rows = db_cursor.fetchall()
|
57 |
|
58 |
+
for row in tqdm(rows, desc="Processing"):
|
59 |
# Get the food item and category
|
60 |
+
fdc_id = row['fdc_id']
|
61 |
food_item = row['description']
|
62 |
category = row['food_category']
|
63 |
|
|
|
73 |
print(f"A: '{best_category}'")
|
74 |
print()
|
75 |
|
|
|
76 |
if best_category:
|
77 |
+
db_cursor.execute('UPDATE dictionary SET wweia_category = %s WHERE fdc_id = %s', (best_category, fdc_id))
|
|
|
|
chatgpt_audit.py
CHANGED
@@ -83,17 +83,14 @@ def parse_response(response):
|
|
83 |
return None
|
84 |
|
85 |
|
86 |
-
csv_file_paths = ['./dictionary/dictionary.csv','./dictionary/additions.csv']
|
87 |
-
dictionary = []
|
88 |
-
for csv_file_path in csv_file_paths:
|
89 |
-
df_dictionary = pd.read_csv(csv_file_path)
|
90 |
-
_dictionary = df_dictionary['description'].astype(str).tolist()
|
91 |
-
stripped_dictionary = [word.strip() for word in _dictionary]
|
92 |
-
dictionary.extend(stripped_dictionary)
|
93 |
-
|
94 |
db_conn = get_connection()
|
95 |
db_cursor = db_conn.cursor()
|
96 |
|
|
|
|
|
|
|
|
|
|
|
97 |
# select all mappings that have not been reviewed
|
98 |
db_cursor.execute("SELECT input_word, dictionary_word, similar_words FROM mappings WHERE reviewed = 0")
|
99 |
results = db_cursor.fetchall()
|
@@ -101,45 +98,47 @@ results = db_cursor.fetchall()
|
|
101 |
# iterate through each row, grab the input_word and ask chatgpt to compare it to the dictionary_word
|
102 |
|
103 |
print("Soft drink, NFS" in dictionary)
|
104 |
-
|
105 |
-
|
106 |
-
|
107 |
-
|
108 |
-
|
109 |
-
|
|
|
|
|
110 |
|
111 |
-
|
112 |
-
|
113 |
-
|
114 |
-
|
115 |
-
|
116 |
-
|
117 |
-
|
118 |
-
|
119 |
-
|
120 |
-
|
121 |
-
|
122 |
-
|
123 |
-
|
124 |
-
|
125 |
-
|
126 |
-
|
127 |
-
|
128 |
-
|
129 |
-
|
130 |
-
|
131 |
-
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
|
136 |
-
|
137 |
-
|
138 |
-
|
139 |
-
|
140 |
-
|
141 |
-
|
142 |
-
|
143 |
-
|
144 |
-
|
145 |
-
db_conn.close()
|
|
|
83 |
return None
|
84 |
|
85 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
86 |
db_conn = get_connection()
|
87 |
db_cursor = db_conn.cursor()
|
88 |
|
89 |
+
# Load the dictionary
|
90 |
+
db_cursor.execute("SELECT description FROM dictionary")
|
91 |
+
dictionary = db_cursor.fetchall()
|
92 |
+
dictionary = [item[0] for item in dictionary]
|
93 |
+
|
94 |
# select all mappings that have not been reviewed
|
95 |
db_cursor.execute("SELECT input_word, dictionary_word, similar_words FROM mappings WHERE reviewed = 0")
|
96 |
results = db_cursor.fetchall()
|
|
|
98 |
# iterate through each row, grab the input_word and ask chatgpt to compare it to the dictionary_word
|
99 |
|
100 |
print("Soft drink, NFS" in dictionary)
|
101 |
+
print(dictionary)
|
102 |
+
print("ensure dictionary works before we start")
|
103 |
+
|
104 |
+
# csv_data = []
|
105 |
+
# for row in results:
|
106 |
+
# input_word = row[0]
|
107 |
+
# dictionary_word = row[1]
|
108 |
+
# similar_words = [item.strip() for item in row[2].split('|')]
|
109 |
|
110 |
+
# # find words from the dictionary list based on small levenstein distance between input_word and each word in the dictionary
|
111 |
+
# levenshtein_words = find_close_levenshtein_words(input_word, dictionary)
|
112 |
+
# print(f"Input: {input_word}")
|
113 |
+
# print(f" - dictionary_word: {dictionary_word}")
|
114 |
+
# print(f" - similar_words: {similar_words}")
|
115 |
+
# print(f" - levenshtein_words: {levenshtein_words}")
|
116 |
+
|
117 |
+
# # concatenate the similar_words and levenshtein_words
|
118 |
+
# all_words = similar_words + levenshtein_words
|
119 |
+
# all_words = list(set(all_words)) # remove duplicates
|
120 |
+
# response = query_gpt(input_word, dictionary_word, all_words)
|
121 |
+
# if response:
|
122 |
+
# csv_data.append({
|
123 |
+
# 'input_word': input_word,
|
124 |
+
# 'original_dictionary_word': dictionary_word,
|
125 |
+
# 'new_dictionary_word': response
|
126 |
+
# })
|
127 |
+
# if response == dictionary_word and response in dictionary:
|
128 |
+
# print(f" - Mapping is correct")
|
129 |
+
# db_cursor.execute("UPDATE mappings SET reviewed = 1 WHERE input_word = ?", (input_word,))
|
130 |
+
# else:
|
131 |
+
# # We should update the mapping in the database
|
132 |
+
# # We should replace dictionary_word with response
|
133 |
+
# # We should set reviewed to 1
|
134 |
+
# # first confirm that the response is in the dictionary
|
135 |
+
# if response in dictionary:
|
136 |
+
# print(f" - Updating mapping with {response}")
|
137 |
+
# db_cursor.execute("UPDATE mappings SET dictionary_word = ?, reviewed = 1 WHERE input_word = ?", (response, input_word))
|
138 |
+
# db_conn.commit()
|
139 |
+
# else:
|
140 |
+
# print(f" - Response {response} is not in the dictionary")
|
141 |
+
|
142 |
+
# update_csv(csv_data)
|
143 |
+
|
144 |
+
# db_conn.close()
|
db/db_utils.py
CHANGED
@@ -37,6 +37,19 @@ def initialize_db(conn):
|
|
37 |
updated_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
|
38 |
)
|
39 |
''')
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
40 |
conn.commit()
|
41 |
|
42 |
def get_mapping_from_db(cursor, cleaned_word):
|
|
|
37 |
updated_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
|
38 |
)
|
39 |
''')
|
40 |
+
cursor.execute('''
|
41 |
+
CREATE TABLE IF NOT EXISTS dictionary (
|
42 |
+
fdc_id INTEGER PRIMARY KEY,
|
43 |
+
description TEXT,
|
44 |
+
food_category TEXT,
|
45 |
+
wweia_category TEXT,
|
46 |
+
water_content REAL,
|
47 |
+
dry_matter_content REAL,
|
48 |
+
leakage REAL,
|
49 |
+
created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
|
50 |
+
updated_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
|
51 |
+
)
|
52 |
+
''')
|
53 |
conn.commit()
|
54 |
|
55 |
def get_mapping_from_db(cursor, cleaned_word):
|
dictionary/additions.csv
DELETED
@@ -1,32 +0,0 @@
|
|
1 |
-
description,food_category
|
2 |
-
"Mixed Produce", "Heterogeneous Mixture"
|
3 |
-
"Miscellaneous", "Heterogeneous Mixture"
|
4 |
-
"Assorted Vegetables", "Heterogeneous Mixture"
|
5 |
-
"Various Fruits", "Heterogeneous Mixture"
|
6 |
-
"Assorted Produce", "Heterogeneous Mixture"
|
7 |
-
"Mixed Vegetables", "Heterogeneous Mixture"
|
8 |
-
"Mixed Vegetables, canned", "Heterogeneous Mixture"
|
9 |
-
"Miscellaneous Items", "Heterogeneous Mixture"
|
10 |
-
"Mixed Goods", "Heterogeneous Mixture"
|
11 |
-
"Assorted Groceries", "Heterogeneous Mixture"
|
12 |
-
"Various Groceries", "Heterogeneous Mixture"
|
13 |
-
"Mixed Food Items", "Heterogeneous Mixture"
|
14 |
-
"Assorted Foods", "Heterogeneous Mixture"
|
15 |
-
"Varied Produce", "Heterogeneous Mixture"
|
16 |
-
"Assorted Fruit and Veg", "Heterogeneous Mixture"
|
17 |
-
"Mixed Fruits and Vegetables", "Heterogeneous Mixture"
|
18 |
-
"Miscellaneous Produce", "Heterogeneous Mixture"
|
19 |
-
"Assorted Consumables", "Heterogeneous Mixture"
|
20 |
-
"Various Edibles", "Heterogeneous Mixture"
|
21 |
-
"Mixed Edibles", "Heterogeneous Mixture"
|
22 |
-
"Assorted Edible Items", "Heterogeneous Mixture"
|
23 |
-
"Mixed Fresh Produce", "Heterogeneous Mixture"
|
24 |
-
"Various Produce Items", "Heterogeneous Mixture"
|
25 |
-
"Misc Grocery", "Heterogeneous Mixture"
|
26 |
-
"Misc Meat", "Heterogeneous Mixture"
|
27 |
-
"Misc Produce", "Heterogeneous Mixture"
|
28 |
-
"Misc Vegetables", "Heterogeneous Mixture"
|
29 |
-
"Grocery Items", "Heterogeneous Mixture"
|
30 |
-
"Grocery", "Heterogeneous Mixture"
|
31 |
-
"Misc Items", "Non-Food Item"
|
32 |
-
"Non-Food Item", "Non-Food Item"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
dictionary/dictionary.csv
DELETED
The diff for this file is too large to render.
See raw diff
|
|
{multi-item-experiments β old_experiments/multi-item-experiments}/classification_results2.csv
RENAMED
File without changes
|
{multi-item-experiments β old_experiments/multi-item-experiments}/multifood2.py
RENAMED
File without changes
|
{multi-item-experiments β old_experiments/multi-item-experiments}/multifood_viz.py
RENAMED
File without changes
|
preseed.py β old_experiments/preseed.py
RENAMED
File without changes
|
playground.py
CHANGED
@@ -1,108 +1 @@
|
|
1 |
-
|
2 |
-
import csv
|
3 |
-
import json
|
4 |
-
import time
|
5 |
-
import heapq
|
6 |
-
import pandas as pd
|
7 |
-
from openai import OpenAI
|
8 |
-
from dotenv import load_dotenv
|
9 |
-
from Levenshtein import distance
|
10 |
-
from tqdm import tqdm
|
11 |
-
from db.db_utils import get_connection, store_mapping_to_db, get_mapping_from_db
|
12 |
-
from ask_gpt import query_gpt
|
13 |
-
|
14 |
-
# For any unreviewed mappings, we ask chatgpt to consider:
|
15 |
-
# 1. The similar_words list
|
16 |
-
# 2. Similar words from the dictionary based on small levenstein distance
|
17 |
-
|
18 |
-
# ChatGPT should confirm that the current mapping is the best one. If not, they should provide the better mapping.
|
19 |
-
# If its a Non-Food Item, we should confirm that
|
20 |
-
# If it's a homogenous or hetergeneous mixture, we should confirm that
|
21 |
-
|
22 |
-
load_dotenv()
|
23 |
-
|
24 |
-
api_key = os.getenv("OPENAI_API_KEY")
|
25 |
-
client = OpenAI(api_key=api_key)
|
26 |
-
|
27 |
-
|
28 |
-
def save_to_csv(results):
|
29 |
-
output_file_path = f'./audits/{int(time.time())}.csv'
|
30 |
-
df_results = pd.DataFrame(results, columns=['input_word', 'original_dictionary_word', 'new_dictionary_word',])
|
31 |
-
df_results.to_csv(output_file_path, index=False)
|
32 |
-
|
33 |
-
def find_close_levenshtein_words(input_word, dictionary, threshold=3):
|
34 |
-
# Calculate Levenshtein distances for each word in the dictionary
|
35 |
-
close_words = [word for word in dictionary if distance(input_word, word) <= threshold]
|
36 |
-
return close_words
|
37 |
-
|
38 |
-
def query_gpt(food_item, dictionary_word, similar_words):
|
39 |
-
line_separated_words = '\n'.join(similar_words)
|
40 |
-
|
41 |
-
prompt = (
|
42 |
-
f"""I have a particular food item and a mapping to a USDA word. Can you confirm if the food item is most similar to the mapping?
|
43 |
-
|
44 |
-
Generally, you should prefer the mapped word, but if you believe there is a better fit, please provide it.
|
45 |
-
|
46 |
-
I will also provide a list of other similar words that you could be a better fit.
|
47 |
-
|
48 |
-
If it's not a food item, return 'Non-Food Item'.
|
49 |
-
|
50 |
-
You should respond in JSON format with an object that has the key `guess`, and the value is the most similar food item.
|
51 |
-
|
52 |
-
The food item is: "{food_item}"
|
53 |
-
It has been mapped to: "{dictionary_word}"
|
54 |
-
|
55 |
-
Similar words:
|
56 |
-
{line_separated_words}"""
|
57 |
-
)
|
58 |
-
|
59 |
-
completion = client.chat.completions.create(
|
60 |
-
messages=[
|
61 |
-
{"role": "system", "content": "You are a helpful assistant."},
|
62 |
-
{"role": "user", "content": prompt}
|
63 |
-
],
|
64 |
-
model="gpt-3.5-turbo-1106",
|
65 |
-
response_format={"type": "json_object"},
|
66 |
-
)
|
67 |
-
response = completion.choices[0].message.content
|
68 |
-
parsed = parse_response(response)
|
69 |
-
print(f"Q: '{food_item}'")
|
70 |
-
print(f"A: '{parsed}'")
|
71 |
-
print()
|
72 |
-
return parsed
|
73 |
-
|
74 |
-
# Define the function to parse the GPT response
|
75 |
-
def parse_response(response):
|
76 |
-
try:
|
77 |
-
result = json.loads(response)
|
78 |
-
return result['guess']
|
79 |
-
except (json.JSONDecodeError, KeyError) as e:
|
80 |
-
print(f"Error parsing response: {response} - {e}")
|
81 |
-
return None
|
82 |
-
|
83 |
-
|
84 |
-
csv_file_paths = ['./dictionary/dictionary.csv','./dictionary/additions.csv']
|
85 |
-
dictionary = []
|
86 |
-
for csv_file_path in csv_file_paths:
|
87 |
-
df_dictionary = pd.read_csv(csv_file_path)
|
88 |
-
_dictionary = df_dictionary['description'].astype(str).tolist()
|
89 |
-
dictionary.extend(_dictionary)
|
90 |
-
|
91 |
-
db_conn = get_connection()
|
92 |
-
db_cursor = db_conn.cursor()
|
93 |
-
|
94 |
-
# select all mappings that have not been reviewed
|
95 |
-
db_cursor.execute("SELECT input_word, dictionary_word, similar_words FROM mappings")
|
96 |
-
results = db_cursor.fetchall()
|
97 |
-
|
98 |
-
# iterate through each row, grab the input_word and ask chatgpt to compare it to the dictionary_word
|
99 |
-
csv_data = []
|
100 |
-
for row in results:
|
101 |
-
input_word = row[0]
|
102 |
-
print(f"input_word: {input_word}")
|
103 |
-
dictionary_word = row[1]
|
104 |
-
if dictionary_word not in dictionary:
|
105 |
-
db_cursor.execute("UPDATE mappings SET reviewed = 0 WHERE input_word = ?", (input_word,))
|
106 |
-
|
107 |
-
|
108 |
-
print(csv_data)
|
|
|
1 |
+
# Nothing here
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
similarity_fast.py
CHANGED
@@ -10,7 +10,6 @@ from utils import generate_embedding, clean_word, cosine_similarity, calculate_c
|
|
10 |
|
11 |
# model_name = 'sentence-transformers/all-MiniLM-L6-v2'
|
12 |
model_name = 'sentence-transformers/all-mpnet-base-v2'
|
13 |
-
csv_file_paths = ['./dictionary/dictionary.csv','./dictionary/additions.csv']
|
14 |
filename = model_name.replace('/', '-')
|
15 |
pickle_file_path = f'./embeddings/fast/{filename}.pkl'
|
16 |
|
@@ -20,11 +19,9 @@ class SimilarityFast:
|
|
20 |
self.db_cursor = db_cursor
|
21 |
self.model = SentenceTransformer(model_name)
|
22 |
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
_dictionary = df_dictionary['description'].astype(str).tolist()
|
27 |
-
dictionary.extend(_dictionary)
|
28 |
|
29 |
self.dictionary_embeddings = self.load_dictionary_embeddings(dictionary)
|
30 |
|
|
|
10 |
|
11 |
# model_name = 'sentence-transformers/all-MiniLM-L6-v2'
|
12 |
model_name = 'sentence-transformers/all-mpnet-base-v2'
|
|
|
13 |
filename = model_name.replace('/', '-')
|
14 |
pickle_file_path = f'./embeddings/fast/{filename}.pkl'
|
15 |
|
|
|
19 |
self.db_cursor = db_cursor
|
20 |
self.model = SentenceTransformer(model_name)
|
21 |
|
22 |
+
self.db_cursor.execute("SELECT description FROM dictionary")
|
23 |
+
dictionary = self.db_cursor.fetchall()
|
24 |
+
dictionary = [item[0] for item in dictionary]
|
|
|
|
|
25 |
|
26 |
self.dictionary_embeddings = self.load_dictionary_embeddings(dictionary)
|
27 |
|
similarity_slow.py
CHANGED
@@ -10,7 +10,6 @@ from utils import generate_embedding, cosine_similarity, clean_word, calculate_c
|
|
10 |
|
11 |
# model_name = 'sentence-transformers/all-MiniLM-L6-v2'
|
12 |
model_name = 'sentence-transformers/all-mpnet-base-v2'
|
13 |
-
csv_file_paths = ['./dictionary/dictionary.csv','./dictionary/additions.csv']
|
14 |
filename = model_name.replace('/', '-')
|
15 |
pickle_file_path = f'./embeddings/slow/{filename}.pkl'
|
16 |
|
@@ -21,12 +20,9 @@ class SimilaritySlow:
|
|
21 |
self.db_conn = db_conn
|
22 |
self.model = SentenceTransformer(model_name)
|
23 |
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
_dictionary = df_dictionary['description'].astype(str).tolist()
|
28 |
-
dictionary.extend(_dictionary)
|
29 |
-
|
30 |
self.dictionary_embeddings = self.load_dictionary_embeddings(dictionary)
|
31 |
|
32 |
def preprocess_dictionary_word(self, text):
|
|
|
10 |
|
11 |
# model_name = 'sentence-transformers/all-MiniLM-L6-v2'
|
12 |
model_name = 'sentence-transformers/all-mpnet-base-v2'
|
|
|
13 |
filename = model_name.replace('/', '-')
|
14 |
pickle_file_path = f'./embeddings/slow/{filename}.pkl'
|
15 |
|
|
|
20 |
self.db_conn = db_conn
|
21 |
self.model = SentenceTransformer(model_name)
|
22 |
|
23 |
+
self.db_cursor.execute("SELECT description FROM dictionary")
|
24 |
+
dictionary = self.db_cursor.fetchall()
|
25 |
+
dictionary = [item[0] for item in dictionary]
|
|
|
|
|
|
|
26 |
self.dictionary_embeddings = self.load_dictionary_embeddings(dictionary)
|
27 |
|
28 |
def preprocess_dictionary_word(self, text):
|