isabel/testing-streamlit

1### ----------------------------- ###2###           libraries           ###3### ----------------------------- ###4 5import streamlit as st6import pickle as pkl7import pandas as pd8import numpy as np9from sklearn.model_selection import train_test_split10from sklearn.linear_model import LogisticRegression11from sklearn import metrics12 13 14### ----------------------------- ###15###        interface setup        ###16### ----------------------------- ###17 18with open('styles.css') as f:19  st.markdown(f'<style>{f.read()}</style>', unsafe_allow_html=True)20 21 22### ------------------------------ ###23###       data transformation      ###24### ------------------------------ ###25 26# load dataset27uncleaned_data = pd.read_csv('data.csv')28  29# remove timestamp from dataset (always first column)30uncleaned_data = uncleaned_data.iloc[: , 1:]31data = pd.DataFrame()32  33# keep track of which columns are categorical and what 34# those columns' value mappings are35# structure: {colname1: {...}, colname2: {...} }36cat_value_dicts = {}37final_colname = uncleaned_data.columns[len(uncleaned_data.columns) - 1]38  39# for each column...40for (colname, colval) in uncleaned_data.iteritems():41  42  # check if col is already a number; if so, add col directly43  # to new dataframe and skip to next column44  if isinstance(colval.values[0], (np.integer, float)):45    data[colname] = uncleaned_data[colname].copy()46    continue47  48  # structure: {0: "lilac", 1: "blue", ...}49  new_dict = {}50  val = 0 # first index per column51  transformed_col_vals = [] # new numeric datapoints52  53  # if not, for each item in that column...54  for (row, item) in enumerate(colval.values):55      56    # if item is not in this col's dict...57    if item not in new_dict:58      new_dict[item] = val59      val += 160      61    # then add numerical value to transformed dataframe62    transformed_col_vals.append(new_dict[item])63    64  # reverse dictionary only for final col (0, 1) => (vals)65  if colname == final_colname:66    new_dict = {value : key for (key, value) in new_dict.items()}67  68  cat_value_dicts[colname] = new_dict69  data[colname] = transformed_col_vals70 71 72### -------------------------------- ###73###           model training         ###74### -------------------------------- ###75 76def train_model():77  # select features and prediction; automatically selects last column as prediction78  cols = len(data.columns)79  num_features = cols - 180  x = data.iloc[: , :num_features]81  y = data.iloc[: , num_features:]82  83  # split data into training and testing sets84  x_train, x_test, y_train, y_test = train_test_split(x, y, test_size=0.25)85  86  # instantiate the model (using default parameters)87  model = LogisticRegression()88  model.fit(x_train, y_train.values.ravel())89  y_pred = model.predict(x_test)90  91  # save the model to file using the pickle package92  with open('model.pkl', 'wb') as f:93    pkl.dump(model, f)94  95  # save model accuracy to file using the pickle package96  with open('acc.txt', 'w+') as f:97    acc = metrics.accuracy_score(y_test, y_pred)98    f.write(str(round(acc * 100, 1)) + '%')99    100  return model101 102### -------------------------------- ###103###            rerun logic           ###104### -------------------------------- ###105 106# check to see if this is the first time running the script,107# if the model has already been trained and saved, load it108try:109  with open('model.pkl', 'rb') as f:110    model = pkl.load(f)111    112# if this is the first time running the script, train the model113# and save it to the file model.pkl114except FileNotFoundError as e:115  model = train_model()116 117# read the model accuracy from file118with open('acc.txt', 'r') as f:119  acc = f.read()120 121 122### ------------------------------- ###123###        interface creation       ###124### ------------------------------- ###125 126# uses the logistic regression to predict for a generic number127# of features128def general_predictor(input_list):129  features = []130 131  # transform categorical input132  for colname, input in zip(data.columns, input_list):133    if (colname in cat_value_dicts):134      features.append(cat_value_dicts[colname][input])135    else:136      features.append(input)137 138  # predict single datapoint139  new_input = [features]140  result = model.predict(new_input)141  return cat_value_dicts[final_colname][result[0]]142 143def get_feat():144  feats = [abs(x) for x in model.coef_[0]]145  max_val = max(feats)146  idx = feats.index(max_val)147  return data.columns[idx]148 149with open('info.md') as f:150  st.title(f.readline())151  st.subheader('Take the quiz to get a personalized recommendation using AI.')152 153form = st.form('ml-inputs')154 155# add data labels to replace those lost via star-args156inputls = []157for colname in data.columns:158  # skip last column159  if colname == final_colname:160    continue161 162  # access categories dict if data is categorical163  # otherwise, just use a number input164  if colname in cat_value_dicts:165    radio_options = list(cat_value_dicts[colname].keys())166    inputls.append(form.selectbox(colname, radio_options))167  else:168    # add numerical input169    inputls.append(form.number_imput(colname))170 171# generate gradio interface172if form.form_submit_button("Submit to get your recommendation!"):173  prediction = general_predictor(inputls)174  175  form.subheader(prediction)176 177col1, col2 = st.columns(2)178col1.metric("Number of Different Possible Results", len(cat_value_dicts[final_colname]))179col2.metric("Model Accuracy",  acc)180st.metric("Most Important Question", "")181st.subheader(get_feat())182st.markdown("***")183 184with open('info.md') as f:185  f.readline()186  st.markdown(f.read())