import streamlit as st
import pandas as pd

st.set_page_config(layout="wide")

st.sidebar.markdown("**Data Filter Options**")
split = st.sidebar.selectbox("Dataset Split", ["train", "test"])
display = st.sidebar.selectbox("Source", ["All", "Only MRPC", "Only MRPC-R1"])

ptype = st.sidebar.radio("Paraphrase Pair Types", ["All",
                                                   "Only Paraphrases (MRPC-R1)",
                                                   "Only Paraphrases (MRPC)",
                                                   "Rejected Paraphrases from MRPC",
                                                   "Corrected Paraphrases from MRPC"])

st.sidebar.markdown("**Score Filter Options**")
filter_by = st.sidebar.selectbox("Filter By Scores From", ["MRPC", "MRPC-R1"])
display_range_wpd = st.sidebar.slider(
    "Filter by WPD Scores", min_value=0.0, max_value=1.0, value=(0.1, 0.7))
display_range_ld = st.sidebar.slider(
    "Filter by LD Scores", min_value=0.0, max_value=1.0, value=(0.1, 0.4))
display_scores = st.sidebar.checkbox("Display scores", value=False)

st.sidebar.markdown("""**Explanation**
This demo allows you to explore the data inside [MRPC](https://www.microsoft.com/en-us/download/details.aspx?id=52398), showing how we can use Word Position Deviation (WPD) and Lexical Deviation (LD) to find different types of paraphrases. By using what we observe from the data, we can also correct numerous labelling errors inside MRPC, presenting the a revision of MRPC termed as MRPC-R1. This demo accompanies the paper ["Towards Better Characterization of Paraphrases" (ACL 2022)](https://github.com/tlkh/paraphrase-metrics).""")

st.markdown("**MRPC Paraphrase Data Explorer**")


def load_df(split):
    if split == "train":
        df = pd.read_csv("./mrpc_train_scores.csv")
    else:
        df = pd.read_csv("./mrpc_test_scores.csv")
    df.reset_index(drop=True, inplace=True)
    return df


def filter_df(df, display, ptype, filter_by, display_scores):
    # filter data
    if display == "MRPC":
        df = df.drop(["new_s1", "new_s2"], axis=1)
    elif display == "MRPC-R1":
        df = df.drop(["og_s1", "og_s2"], axis=1)
    # filter paraphrase type
    if ptype == "Only Paraphrases (MRPC)":
        condition = df.og_label == 1
        df_sel = df[condition]
    elif ptype == "Only Paraphrases (MRPC-R1)":
        condition = df.new_label == 1
        df_sel = df[condition]
    elif ptype == "Rejected Paraphrases from MRPC":
        condition = (df.new_label == 0) & (df.og_label == 1)
        df_sel = df[condition]
    elif ptype == "Corrected Paraphrases from MRPC":
        condition = df.remarks == "corrected"
        df_sel = df[condition]
    else:
        # all
        df_sel = df
    # sort by scores
    if filter_by == "MRPC":
        # wpd
        condition = (df_sel.og_wpd >= display_range_wpd[0]) & (
            df_sel.og_wpd < display_range_wpd[1])
        df_sel = df_sel[condition]
        # ld
        condition = (df_sel.og_ld >= display_range_ld[0]) & (
            df_sel.og_ld < display_range_ld[1])
        df_sel = df_sel[condition]
    else:
        # wpd
        condition = (df_sel.new_wpd >= display_range_wpd[0]) & (
            df_sel.new_wpd < display_range_wpd[1])
        df_sel = df_sel[condition]
        # ld
        condition = (df_sel.new_ld >= display_range_ld[0]) & (
            df_sel.new_ld < display_range_ld[1])
        df_sel = df_sel[condition]
    # filter scores
    if filter_by == "MRPC":
        df_sel.sort_values("og_ld", inplace=True)
        df_sel.sort_values("og_wpd", inplace=True)
    else:
        df_sel.sort_values("new_ld", inplace=True)
        df_sel.sort_values("new_wpd", inplace=True)
    if not display_scores:
        df_sel.drop(["og_ld", "og_wpd", "new_ld", "new_wpd"], axis=1, inplace=True)
    label_col = df_sel["og_label"].astype(str)+"->"+df_sel["new_label"].astype(str)
    df_sel["og/new label"] = label_col
    df_sel.drop(["remarks", "og_label", "new_label"], axis=1, inplace=True)
    return df_sel


df = load_df(split)

df_sel = filter_df(df, display, ptype, filter_by, display_scores)

st.markdown("Displaying "+str(len(df_sel))+" items")

st.table(data=df_sel)