import streamlit as st
import pathlib
import json
import pandas as pd

st.header("Time Series Preprocessing Pipeline")
st.markdown("Users can load their time-series data and select a set of transformations to prepare a training set for univariate or multivariate time-series classification.\
 Go ahead and use the sidebar on the left to upload your data files in *.json* format and start exploring and transforming it!")
col1, col2 = st.columns(2)

file_names, file_bytes = [], []
with st.sidebar:
    files = st.file_uploader("Load files", accept_multiple_files = True)
    if files:
        file_names = [file.name for file in files]
        file_bytes = [file.getvalue() for file in files]
        st.text("\n".join(file_names))

data_dict = dict({'trial_id':[], 'pupil_dilation':[], 'baseline':[], 'rating':[]})
with st.spinner("building base dictionary..."):
    for file_data in file_bytes:
        data = json.loads(file_data)
        for k in data:
          for i in data[k]:
            for k, v in i.items():
              data_dict[k].append(v)

df_base = pd.DataFrame()  # {'<fields>' : []})
with col1:
    if file_bytes:
        with st.spinner("building base dataframe..."):
            df_base = pd.DataFrame.from_dict(data_dict)
            df_base["trial_id"] = df_base.trial_id.map(lambda s: "".join([c for c in s if c.isdigit()]))
            df_base["len_pupil_dilation"] = df_base.pupil_dilation.map(lambda l: len(l))
            df_base["len_baseline"] = df_base.baseline.map(lambda l: len(l))
            st.info(f"number of files: {len(file_names)}")
            st.markdown("Your original data")
            st.dataframe(df_base)
    else:
        st.caption("Upload your data from the sidebar to start :sunglasses:")

with col2:
    if not df_base.empty:
        st.markdown("**Cleaning actions**")
        detect_blinking = st.button("Detect blinking ('0.0' values)")
        number_of_blinks = 0
        if detect_blinking:
            # Initialization of session_state
            if 'df' not in st.session_state:
                st.session_state['df'] = df_base
            for ser in df_base['pupil_dilation']:
                for f in ser:
                    if f == 0.0:
                        number_of_blinks += 1
            
            for ser in df_base['baseline']:
                for f in ser:
                    if f == 0.0:
                        number_of_blinks += 1
            # Initialization of session_state
            if 'blinks' not in st.session_state:
                st.session_state['blinks'] = number_of_blinks
                        
        if "blinks" in st.session_state.keys():
            st.info(f"blinking values (0.0) were found in {number_of_blinks} time-steps in all your data")
            remove_blinking  = st.button("Remove blinking")
            # df in column 2
            if remove_blinking:
                df_right = st.session_state.df.copy(deep=True)
                df_right.pupil_dilation = df_right.pupil_dilation.map(lambda ser: [f for f in ser if f != 0.0])
                df_right.baseline = df_right.baseline.map(lambda ser: [f for f in ser if f != 0.0])
                st.success("blinking values have been removed!")
                st.info("after transformation")
                st.dataframe(df_right)
        elif detect_blinking and not number_of_blinks:
            st.caption("no blinking values were found in your data!")

if not df_base.empty:
    st.warning("consider running outlier detection to clean your data!", icon="⚠️")

# for key, value in st.session_state.items():
#    st.success(f"{key}: {value}")

# reloading new samples would damage the st-session_state loading, vars are only written once