update

main.py

@@ -17,8 +17,9 @@ from subpages import (
     RawDataPage,
 )
 from subpages.attention import AttentionPage
-from subpages.embeddings import EmbeddingsPage
+from subpages.hidden_states import HiddenStatesPage
 from subpages.inspect import InspectPage
+from utils import classmap
 
 sts = st.sidebar
 st.set_page_config(
@@ -54,25 +55,11 @@ def _write_color_legend(context):
     def style(x):
         return [f"background-color: {rgb}; opacity: 1;" for rgb in colors]
 
-    labelmap = {
-        "O": "O",
-        "person": "🙎",
-        "PER": "🙎",
-        "location": "🌎",
-        "LOC": "🌎",
-        "corporation": "🏤",
-        "ORG": "🏤",
-        "product": "📱",
-        "creative": "🎷",
-        "group": "🎷",
-        "MISC": "🎷",
-    }
-
     labels = list(set([lbl.split("-")[1] if "-" in lbl else lbl for lbl in context.labels]))
     colors = [st.session_state.get(f"color_{lbl}", "#000000") for lbl in labels]
 
     color_legend_df = pd.DataFrame(
-        [labelmap[l] for l in labels], columns=["label"], index=labels
+        [classmap[l] for l in labels], columns=["label"], index=labels
     ).T
     st.sidebar.write(
         color_legend_df.T.style.apply(style, axis=0).set_properties(
@@ -85,7 +72,7 @@ def main():
     pages: list[Page] = [
         HomePage(),
         AttentionPage(),
-        EmbeddingsPage(),
+        HiddenStatesPage(),
         ProbingPage(),
         MetricsPage(),
         MisclassifiedPage(),

subpages/__init__.py

@@ -1,7 +1,7 @@
 from subpages.attention import AttentionPage
 from subpages.debug import DebugPage
-from subpages.embeddings import EmbeddingsPage
 from subpages.find_duplicates import FindDuplicatesPage
+from subpages.hidden_states import HiddenStatesPage
 from subpages.home import HomePage
 from subpages.inspect import InspectPage
 from subpages.losses import LossesPage

subpages/{embeddings.py → hidden_states.py}

@@ -28,8 +28,8 @@ def reduce_dim_umap(X, n_neighbors=5, min_dist=0.1, metric="euclidean"):
     return UMAP(n_neighbors=n_neighbors, min_dist=min_dist, metric=metric).fit_transform(X)
 
 
-class EmbeddingsPage(Page):
-    name = "Embeddings"
+class HiddenStatesPage(Page):
+    name = "Hidden States"
     icon = "grid-3x3"
 
     def get_widget_defaults(self):

subpages/home.py

@@ -3,11 +3,10 @@ import random
 from typing import Optional
 
 import streamlit as st
-from pandas import wide_to_long
 
 from data import get_data
 from subpages.page import Context, Page
-from utils import color_map_color
+from utils import classmap, color_map_color
 
 _SENTENCE_ENCODER_MODEL = (
     "sentence-transformers/all-MiniLM-L6-v2",
@@ -53,7 +52,7 @@ class HomePage(Page):
 
         with st.expander("💡", expanded=True):
             st.write(
-                "**Error Analysis is an important but often overlooked part of the data science project lifecycle**, for which there is still very little tooling available. Practitioners tend to write throwaway code or, worse, skip this crucial step of understanding their models' errors altogether. This project tries to provide an **extensive toolkit to probe any NER model/dataset combination**, find labeling errors and understand the models' and datasets' limitations, leading the user on her way to further improvements."
+                "**Error Analysis is an important but often overlooked part of the data science project lifecycle**, for which there is still very little tooling available. Practitioners tend to write throwaway code or, worse, skip this crucial step of understanding their models' errors altogether. This project tries to provide an **extensive toolkit to probe any NER model/dataset combination**, find labeling errors and understand the models' and datasets' limitations, leading the user on her way to further **improving both model AND dataset**."
             )
 
         col1, _, col2a, col2b = st.columns([1, 0.05, 0.15, 0.15])
@@ -91,7 +90,7 @@ class HomePage(Page):
                 st.text_input(
                     label="Encoder Model:",
                     key="encoder_model_name",
-                    help="Path or name of the encoder to use",
+                    help="Path or name of the encoder to use for duplicate detection",
                 )
                 ds_name = st.text_input(
                     label="Dataset:",
@@ -136,8 +135,9 @@ class HomePage(Page):
             emojis = list(json.load(open("subpages/emoji-en-US.json")).keys())
             for label in labels:
                 if f"icon_{label}" not in st.session_state:
-                    st.session_state[f"icon_{label}"] = "🤗"  #  labels[label]
+                    st.session_state[f"icon_{label}"] = classmap[label]
                 st.selectbox(label, key=f"icon_{label}", options=emojis)
+                classmap[label] = st.session_state[f"icon_{label}"]
 
         # if st.button("Reset to defaults"):
         #     st.session_state.update(**get_home_page_defaults())

utils.py

@@ -14,6 +14,19 @@ tokenizer_hash_funcs = {
 # device = torch.device("cuda" if torch.cuda.is_available() else "cpu" if torch.has_mps else "cpu")
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 
+classmap = {
+    "O": "O",
+    "PER": "🙎",
+    "person": "🙎",
+    "LOC": "🌎",
+    "location": "🌎",
+    "ORG": "🏤",
+    "corporation": "🏤",
+    "product": "📱",
+    "creative": "🎷",
+    "MISC": "🎷",
+}
+
 
 def aggrid_interactive_table(df: pd.DataFrame) -> dict:
     """Creates an st-aggrid interactive table based on a dataframe.
@@ -159,18 +172,6 @@ def colorize_classes(df: pd.DataFrame) -> pd.DataFrame:
 
 def htmlify_labeled_example(example: pd.DataFrame) -> str:
     html = []
-    classmap = {
-        "O": "O",
-        "PER": "🙎",
-        "person": "🙎",
-        "LOC": "🌎",
-        "location": "🌎",
-        "ORG": "🏤",
-        "corporation": "🏤",
-        "product": "📱",
-        "creative": "🎷",
-        "MISC": "🎷",
-    }
 
     for _, row in example.iterrows():
         pred = row.preds.split("-")[1] if "-" in row.preds else "O"