Calvin
commited on
Commit
•
f94a42e
1
Parent(s):
8a1aceb
final touches
Browse files- Exploration.ipynb +0 -0
- offer_pipeline.py +119 -24
- requirements.txt +0 -1
Exploration.ipynb
ADDED
The diff for this file is too large to render.
See raw diff
|
|
offer_pipeline.py
CHANGED
@@ -3,7 +3,6 @@ from transformers import pipeline
|
|
3 |
import pickle
|
4 |
import os
|
5 |
import pandas as pd
|
6 |
-
# import seaborn as sns
|
7 |
import ast
|
8 |
import string
|
9 |
import re
|
@@ -14,79 +13,165 @@ st.set_page_config(
|
|
14 |
layout="wide"
|
15 |
)
|
16 |
|
17 |
-
|
|
|
18 |
model = SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2')
|
19 |
|
|
|
20 |
dire = "DS_NLP_search_data"
|
21 |
|
|
|
22 |
@st.cache_data
|
23 |
def get_processed_offers():
|
|
|
|
|
|
|
|
|
|
|
|
|
24 |
processed_offers = pd.read_csv(os.path.join(dire, "processed_offers.csv"))
|
25 |
processed_offers["CATEGORY"] = processed_offers["CATEGORY"].map(ast.literal_eval)
|
|
|
26 |
return processed_offers
|
27 |
|
|
|
28 |
@st.cache_data
|
29 |
def get_categories_data():
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
30 |
cats = pd.read_csv(os.path.join(dire, "categories.csv"))
|
|
|
31 |
return cats
|
32 |
|
|
|
33 |
@st.cache_data
|
34 |
def get_offers_data():
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
35 |
offers = pd.read_csv(os.path.join(dire, "offer_retailer.csv"))
|
|
|
36 |
return offers
|
37 |
|
|
|
38 |
@st.cache_data
|
39 |
def get_categories(cats_):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
40 |
categories = list(cats_["IS_CHILD_CATEGORY_TO"].unique())
|
41 |
for x in ["Mature"]:
|
42 |
if x in categories:
|
43 |
categories.remove(x)
|
|
|
44 |
return categories
|
45 |
|
|
|
46 |
def check_in_offer(search_str, offer_rets):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
47 |
offers = []
|
48 |
-
# print(offer_rets)
|
49 |
for i in range(len(offer_rets)):
|
50 |
offer_str = offer_rets.iloc[i]["OFFER"]
|
51 |
-
# print(offer_str)
|
52 |
parsed_str = offer_str.lower().translate(str.maketrans('', '', string.punctuation))
|
53 |
parsed_str = re.sub('[^a-zA-Z0-9 \n\.]', '', parsed_str)
|
54 |
-
|
55 |
if search_str.lower() in parsed_str.split(" "):
|
56 |
offers.append(offer_str)
|
57 |
df = pd.DataFrame({"OFFER":offers})
|
58 |
-
|
59 |
return df
|
60 |
|
|
|
61 |
def is_retailer(search_str, threshold=0.5):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
62 |
processed_search_str = search_str.lower().capitalize()
|
63 |
labels = pipe(processed_search_str,
|
64 |
candidate_labels=["brand", "retailer", "item"],
|
65 |
)
|
66 |
|
67 |
-
|
|
|
|
|
|
|
68 |
|
69 |
def perform_cat_inference(search_str, categories, cats, processed_offers):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
70 |
labels = pipe(search_str,
|
71 |
candidate_labels=categories,
|
72 |
)
|
73 |
-
print(labels)
|
74 |
# labels = [l for i, l in enumerate(labels["labels"]) if labels["scores"][i] > 0.20]
|
75 |
filtered_cats = list(cats[cats["IS_CHILD_CATEGORY_TO"].isin(labels["labels"][:3])]["PRODUCT_CATEGORY"].unique())
|
76 |
labels_2 = pipe(search_str,
|
77 |
candidate_labels=filtered_cats,
|
78 |
)
|
79 |
-
print(labels_2)
|
80 |
top_labels = labels_2["labels"][:3]
|
81 |
-
|
82 |
-
|
83 |
-
|
84 |
-
print(top_labels)
|
85 |
offers = processed_offers[processed_offers["CATEGORY"].apply(lambda x: bool(set(x) & set(top_labels)))]["OFFER"].reset_index()
|
86 |
|
87 |
return offers, labels, labels_2
|
88 |
|
|
|
89 |
def sort_by_similarity(search_str, related_offers):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
90 |
temp_dict = {}
|
91 |
embedding_1 = model.encode(search_str, convert_to_tensor=True)
|
92 |
|
@@ -96,42 +181,52 @@ def sort_by_similarity(search_str, related_offers):
|
|
96 |
temp_dict[offer] = float(util.pytorch_cos_sim(embedding_1, embedding_2))
|
97 |
|
98 |
sorted_dict = dict(sorted(temp_dict.items(), key=lambda x : x[1], reverse=True))
|
99 |
-
# casted_scores = list(map(lambda x : int(x), ))
|
100 |
df = pd.DataFrame({"OFFER":list(sorted_dict.keys())[:20], "scores":list(sorted_dict.values())[:20]})
|
|
|
101 |
return df
|
102 |
|
|
|
103 |
def main():
|
|
|
104 |
col_1, col_2, col_3 = st.columns(3)
|
105 |
-
search_str =
|
106 |
processed_offers = get_processed_offers()
|
107 |
cats = get_categories_data()
|
108 |
offer_rets = get_offers_data()
|
109 |
categories = get_categories(cats)
|
110 |
-
# retail_mapping = get_prod_categories()
|
111 |
|
112 |
-
if
|
|
|
113 |
retail = is_retailer(search_str)
|
114 |
direct_offers = check_in_offer(search_str, offer_rets)
|
115 |
col_2.write("Directly related offers")
|
116 |
-
|
|
|
|
|
|
|
|
|
117 |
|
118 |
if retail:
|
|
|
119 |
related_offers = offer_rets[~offer_rets["OFFER"].isin(list(direct_offers["OFFER"]))]
|
120 |
else:
|
|
|
121 |
related_offers, labels_1, labels_2 = perform_cat_inference(search_str, categories, cats, processed_offers)
|
122 |
related_offers = related_offers[~related_offers["OFFER"].isin(list(direct_offers["OFFER"]))]
|
123 |
|
|
|
124 |
col_2.table(pd.DataFrame({"labels": labels_1["labels"][:5], "scores": labels_1["scores"][:5]}))
|
|
|
125 |
col_2.table(pd.DataFrame({"labels": labels_2["labels"][:5], "scores": labels_2["scores"][:5]}))
|
126 |
-
|
127 |
-
|
128 |
-
# df = get_confidence_charts(labels_2)
|
129 |
-
# st.table(df)
|
130 |
|
131 |
col_2.write("Other related offers")
|
132 |
sorted_offers = sort_by_similarity(search_str, related_offers)
|
133 |
-
col_2.table(sorted_offers)
|
134 |
-
if __name__ == "__main__":
|
135 |
|
|
|
|
|
|
|
|
|
|
|
|
|
136 |
main()
|
137 |
|
|
|
3 |
import pickle
|
4 |
import os
|
5 |
import pandas as pd
|
|
|
6 |
import ast
|
7 |
import string
|
8 |
import re
|
|
|
13 |
layout="wide"
|
14 |
)
|
15 |
|
16 |
+
# Download and cache models
|
17 |
+
pipe = pipeline(task="zero-shot-classification", model="valhalla/distilbart-mnli-12-3")
|
18 |
model = SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2')
|
19 |
|
20 |
+
# Directory of csv files
|
21 |
dire = "DS_NLP_search_data"
|
22 |
|
23 |
+
# Use Streamlit caching to load data once
|
24 |
@st.cache_data
|
25 |
def get_processed_offers():
|
26 |
+
'''
|
27 |
+
Load processed offers from exploration notebook and cache
|
28 |
+
|
29 |
+
Returns:
|
30 |
+
processed_offers (pd.DataFrame) : zero-shot categorized offers
|
31 |
+
'''
|
32 |
processed_offers = pd.read_csv(os.path.join(dire, "processed_offers.csv"))
|
33 |
processed_offers["CATEGORY"] = processed_offers["CATEGORY"].map(ast.literal_eval)
|
34 |
+
|
35 |
return processed_offers
|
36 |
|
37 |
+
|
38 |
@st.cache_data
|
39 |
def get_categories_data():
|
40 |
+
'''
|
41 |
+
Load raw category data and cache
|
42 |
+
|
43 |
+
Returns:
|
44 |
+
cats (pd.DataFrame) : raw category data
|
45 |
+
'''
|
46 |
+
|
47 |
cats = pd.read_csv(os.path.join(dire, "categories.csv"))
|
48 |
+
|
49 |
return cats
|
50 |
|
51 |
+
|
52 |
@st.cache_data
|
53 |
def get_offers_data():
|
54 |
+
'''
|
55 |
+
Load raw offfers data and cache
|
56 |
+
|
57 |
+
Returns:
|
58 |
+
cats (pd.DataFrame) : raw offers data
|
59 |
+
'''
|
60 |
+
|
61 |
offers = pd.read_csv(os.path.join(dire, "offer_retailer.csv"))
|
62 |
+
|
63 |
return offers
|
64 |
|
65 |
+
|
66 |
@st.cache_data
|
67 |
def get_categories(cats_):
|
68 |
+
'''
|
69 |
+
Extract, load categories and cache
|
70 |
+
|
71 |
+
Parameters:
|
72 |
+
cats_ (pd.DataFrame) : raw categories data
|
73 |
+
|
74 |
+
Returns:
|
75 |
+
categories (List) : child categories
|
76 |
+
'''
|
77 |
+
|
78 |
categories = list(cats_["IS_CHILD_CATEGORY_TO"].unique())
|
79 |
for x in ["Mature"]:
|
80 |
if x in categories:
|
81 |
categories.remove(x)
|
82 |
+
|
83 |
return categories
|
84 |
|
85 |
+
|
86 |
def check_in_offer(search_str, offer_rets):
|
87 |
+
'''
|
88 |
+
Determine if the input text is directly in the offer with basic string matching
|
89 |
+
|
90 |
+
Parameters:
|
91 |
+
search_str (string) : user text input
|
92 |
+
offer_rets (pd.DataFrame) : raw offer data
|
93 |
+
|
94 |
+
Returns:
|
95 |
+
df (pd.DataFrame) : offers with text input
|
96 |
+
'''
|
97 |
+
|
98 |
offers = []
|
|
|
99 |
for i in range(len(offer_rets)):
|
100 |
offer_str = offer_rets.iloc[i]["OFFER"]
|
|
|
101 |
parsed_str = offer_str.lower().translate(str.maketrans('', '', string.punctuation))
|
102 |
parsed_str = re.sub('[^a-zA-Z0-9 \n\.]', '', parsed_str)
|
103 |
+
|
104 |
if search_str.lower() in parsed_str.split(" "):
|
105 |
offers.append(offer_str)
|
106 |
df = pd.DataFrame({"OFFER":offers})
|
107 |
+
|
108 |
return df
|
109 |
|
110 |
+
|
111 |
def is_retailer(search_str, threshold=0.5):
|
112 |
+
'''
|
113 |
+
Determine if the text input is highly likely to be a retailer
|
114 |
+
|
115 |
+
Parameters:
|
116 |
+
search_str (string) : user text input
|
117 |
+
threshold (int) : probability threshold
|
118 |
+
|
119 |
+
Returns:
|
120 |
+
is_ret (boolean) : true if retailer, false otherwise
|
121 |
+
'''
|
122 |
+
|
123 |
processed_search_str = search_str.lower().capitalize()
|
124 |
labels = pipe(processed_search_str,
|
125 |
candidate_labels=["brand", "retailer", "item"],
|
126 |
)
|
127 |
|
128 |
+
is_ret = labels["labels"][0] == "retailer" and labels["scores"][0] > threshold
|
129 |
+
|
130 |
+
return is_ret
|
131 |
+
|
132 |
|
133 |
def perform_cat_inference(search_str, categories, cats, processed_offers):
|
134 |
+
'''
|
135 |
+
Perform zero shot learning twice and return the offers relevant to the child categories
|
136 |
+
|
137 |
+
Parameters:
|
138 |
+
search_str (string) : user text input
|
139 |
+
categories (pd.DataFrame) : list of categories
|
140 |
+
cats (pd.DataFrame) : raw category data
|
141 |
+
processed_offers (pd.DataFrame) : processed_offer_data
|
142 |
+
|
143 |
+
Returns:
|
144 |
+
offers (pd.DataFrame) : relevant offers
|
145 |
+
labels (dict) : parent categories and their probability scores
|
146 |
+
labels_2 (dict) : child categories and their probability scores
|
147 |
+
'''
|
148 |
+
|
149 |
labels = pipe(search_str,
|
150 |
candidate_labels=categories,
|
151 |
)
|
|
|
152 |
# labels = [l for i, l in enumerate(labels["labels"]) if labels["scores"][i] > 0.20]
|
153 |
filtered_cats = list(cats[cats["IS_CHILD_CATEGORY_TO"].isin(labels["labels"][:3])]["PRODUCT_CATEGORY"].unique())
|
154 |
labels_2 = pipe(search_str,
|
155 |
candidate_labels=filtered_cats,
|
156 |
)
|
|
|
157 |
top_labels = labels_2["labels"][:3]
|
|
|
|
|
|
|
|
|
158 |
offers = processed_offers[processed_offers["CATEGORY"].apply(lambda x: bool(set(x) & set(top_labels)))]["OFFER"].reset_index()
|
159 |
|
160 |
return offers, labels, labels_2
|
161 |
|
162 |
+
|
163 |
def sort_by_similarity(search_str, related_offers):
|
164 |
+
'''
|
165 |
+
Use sentence embeddings to evaluate the similarity of relevant offers to the text input
|
166 |
+
|
167 |
+
Parameters:
|
168 |
+
search_str (string) : user text input
|
169 |
+
related_offers (pd.DataFrame) : relevant offers discovered by zero shot learning
|
170 |
+
|
171 |
+
Returns:
|
172 |
+
df (pd.DataFrame) : relevant offers and their similiarity scores
|
173 |
+
'''
|
174 |
+
|
175 |
temp_dict = {}
|
176 |
embedding_1 = model.encode(search_str, convert_to_tensor=True)
|
177 |
|
|
|
181 |
temp_dict[offer] = float(util.pytorch_cos_sim(embedding_1, embedding_2))
|
182 |
|
183 |
sorted_dict = dict(sorted(temp_dict.items(), key=lambda x : x[1], reverse=True))
|
|
|
184 |
df = pd.DataFrame({"OFFER":list(sorted_dict.keys())[:20], "scores":list(sorted_dict.values())[:20]})
|
185 |
+
|
186 |
return df
|
187 |
|
188 |
+
|
189 |
def main():
|
190 |
+
# Load and cache data
|
191 |
col_1, col_2, col_3 = st.columns(3)
|
192 |
+
search_str = col_1.text_input("Enter a retailer, brand, or category").capitalize()
|
193 |
processed_offers = get_processed_offers()
|
194 |
cats = get_categories_data()
|
195 |
offer_rets = get_offers_data()
|
196 |
categories = get_categories(cats)
|
|
|
197 |
|
198 |
+
if col_1.button("Search", type="primary"):
|
199 |
+
# Check offers where the text is directly in it
|
200 |
retail = is_retailer(search_str)
|
201 |
direct_offers = check_in_offer(search_str, offer_rets)
|
202 |
col_2.write("Directly related offers")
|
203 |
+
|
204 |
+
if len(direct_offers) == 0:
|
205 |
+
col_2.write("None found")
|
206 |
+
else:
|
207 |
+
col_2.table(direct_offers)
|
208 |
|
209 |
if retail:
|
210 |
+
# If retail, we directly compare every offer using sentence embeddings
|
211 |
related_offers = offer_rets[~offer_rets["OFFER"].isin(list(direct_offers["OFFER"]))]
|
212 |
else:
|
213 |
+
# Otherwise, we use zero shot learning with processed offers to narrow down our search
|
214 |
related_offers, labels_1, labels_2 = perform_cat_inference(search_str, categories, cats, processed_offers)
|
215 |
related_offers = related_offers[~related_offers["OFFER"].isin(list(direct_offers["OFFER"]))]
|
216 |
|
217 |
+
col_2.write("Parent categories probabilities")
|
218 |
col_2.table(pd.DataFrame({"labels": labels_1["labels"][:5], "scores": labels_1["scores"][:5]}))
|
219 |
+
col_2.write("Child categories probabilities")
|
220 |
col_2.table(pd.DataFrame({"labels": labels_2["labels"][:5], "scores": labels_2["scores"][:5]}))
|
|
|
|
|
|
|
|
|
221 |
|
222 |
col_2.write("Other related offers")
|
223 |
sorted_offers = sort_by_similarity(search_str, related_offers)
|
|
|
|
|
224 |
|
225 |
+
if len(sorted_offers) == 0:
|
226 |
+
col_2.write("None found")
|
227 |
+
else:
|
228 |
+
col_2.table(sorted_offers)
|
229 |
+
|
230 |
+
if __name__ == "__main__":
|
231 |
main()
|
232 |
|
requirements.txt
CHANGED
@@ -1,6 +1,5 @@
|
|
1 |
streamlit
|
2 |
transformers
|
3 |
pandas
|
4 |
-
seaborn
|
5 |
torch
|
6 |
sentence-transformers
|
|
|
1 |
streamlit
|
2 |
transformers
|
3 |
pandas
|
|
|
4 |
torch
|
5 |
sentence-transformers
|