Spaces:
Sleeping
Sleeping
File size: 52,439 Bytes
00b00eb |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 |
"""
MMO Build Sprint 3
date :
changes : capability to tune MixedLM as well as simple LR in the same page
"""
import os
import streamlit as st
import pandas as pd
from data_analysis import format_numbers
import pickle
from utilities import set_header, load_local_css
import statsmodels.api as sm
import re
from sklearn.preprocessing import MaxAbsScaler
import matplotlib.pyplot as plt
from statsmodels.stats.outliers_influence import variance_inflation_factor
import statsmodels.formula.api as smf
from data_prep import *
import sqlite3
from utilities import (
set_header,
load_local_css,
update_db,
project_selection,
retrieve_pkl_object,
)
import numpy as np
from post_gres_cred import db_cred
import re
from constants import (
NUM_FLAG_COLS_TO_DISPLAY,
HALF_YEAR_THRESHOLD,
FULL_YEAR_THRESHOLD,
TREND_MIN,
ANNUAL_FREQUENCY,
QTR_FREQUENCY_FACTOR,
HALF_YEARLY_FREQUENCY_FACTOR,
)
from log_application import log_message
import sys, traceback
schema = db_cred["schema"]
st.set_option("deprecation.showPyplotGlobalUse", False)
st.set_page_config(
page_title="AI Model Tuning",
page_icon=":shark:",
layout="wide",
initial_sidebar_state="collapsed",
)
load_local_css("styles.css")
set_header()
# Define functions
# Get random effect from MixedLM Model
def get_random_effects(media_data, panel_col, _mdf):
# create an empty dataframe
random_eff_df = pd.DataFrame(columns=[panel_col, "random_effect"])
# Iterate over all panel values and add to dataframe
for i, market in enumerate(media_data[panel_col].unique()):
intercept = _mdf.random_effects[market].values[0]
random_eff_df.loc[i, "random_effect"] = intercept
random_eff_df.loc[i, panel_col] = market
return random_eff_df
# Predict on df using MixedLM model
def mdf_predict(X_df, mdf, random_eff_df):
# Create a copy of input df and predict using MixedLM model i.e fixed effect
X = X_df.copy()
X["fixed_effect"] = mdf.predict(X)
# Merge random effects
X = pd.merge(X, random_eff_df, on=panel_col, how="left")
# Get final predictions by adding random effect to fixed effect
X["pred"] = X["fixed_effect"] + X["random_effect"]
# Drop intermediate columns
X.drop(columns=["fixed_effect", "random_effect"], inplace=True)
return X["pred"]
def format_display(inp):
# Format display titles
return inp.title().replace("_", " ").strip()
if "username" not in st.session_state:
st.session_state["username"] = None
if "project_name" not in st.session_state:
st.session_state["project_name"] = None
if "project_dct" not in st.session_state:
project_selection()
st.stop()
if "Flags" not in st.session_state:
st.session_state["Flags"] = {}
try:
# Check Authentications
if "username" in st.session_state and st.session_state["username"] is not None:
if (
retrieve_pkl_object(
st.session_state["project_number"], "Model_Build", "best_models", schema
)
is None
): # db
st.error("Please save a model before tuning")
log_message(
"warning",
"No models saved",
"Model Tuning",
)
st.stop()
# Read previous progress (persistence)
if (
"session_state_saved"
in st.session_state["project_dct"]["model_build"].keys()
):
for key in [
"Model",
"date",
"saved_model_names",
"media_data",
"X_test_spends",
"spends_data",
]:
if key not in st.session_state:
st.session_state[key] = st.session_state["project_dct"][
"model_build"
]["session_state_saved"][key]
st.session_state["bin_dict"] = st.session_state["project_dct"][
"model_build"
]["session_state_saved"]["bin_dict"]
if (
"used_response_metrics" not in st.session_state
or st.session_state["used_response_metrics"] == []
):
st.session_state["used_response_metrics"] = st.session_state[
"project_dct"
]["model_build"]["session_state_saved"]["used_response_metrics"]
else:
st.error("Please load a session with a built model")
log_message(
"error",
"Session state saved not found in Project Dictionary",
"Model Tuning",
)
st.stop()
for key in ["select_all_flags_check", "selected_flags", "sel_model"]:
if key not in st.session_state["project_dct"]["model_tuning"].keys():
st.session_state["project_dct"]["model_tuning"][key] = {}
# is_panel = st.session_state['is_panel']
# panel_col = 'markets' # set the panel column
date_col = "date"
# set the panel column
panel_col = "panel"
is_panel = (
True if st.session_state["media_data"][panel_col].nunique() > 1 else False
)
if "Model_Tuned" not in st.session_state:
st.session_state["Model_Tuned"] = {}
cols1 = st.columns([2, 1])
with cols1[0]:
st.markdown(f"**Welcome {st.session_state['username']}**")
with cols1[1]:
st.markdown(f"**Current Project: {st.session_state['project_name']}**")
st.title("AI Model Tuning")
# flag indicating there is not tuned model till now
if "is_tuned_model" not in st.session_state:
st.session_state["is_tuned_model"] = {}
# # Read all saved models
model_dict = retrieve_pkl_object(
st.session_state["project_number"], "Model_Build", "best_models", schema
)
saved_models = model_dict.keys()
# Get list of response metrics
st.session_state["used_response_metrics"] = list(
set([model.split("__")[1] for model in saved_models])
)
# Select previously selected response_metric (persistence)
default_target_idx = (
st.session_state["project_dct"]["model_tuning"].get("sel_target_col", None)
if st.session_state["project_dct"]["model_tuning"].get(
"sel_target_col", None
)
is not None
else st.session_state["used_response_metrics"][0]
)
# Dropdown to select response metric
sel_target_col = st.selectbox(
"Select the response metric",
st.session_state["used_response_metrics"],
index=st.session_state["used_response_metrics"].index(default_target_idx),
format_func=format_display,
)
# Format selected response metrics (target col)
target_col = (
sel_target_col.lower()
.replace(" ", "_")
.replace("-", "")
.replace(":", "")
.replace("__", "_")
)
st.session_state["project_dct"]["model_tuning"][
"sel_target_col"
] = sel_target_col
# Look through all saved models, only show saved models of the selected resp metric (target_col)
# Get a list of models saved for selected response metric
required_saved_models = [
m.split("__")[0] for m in saved_models if m.split("__")[1] == target_col
]
# Get previously seelcted model if available (persistence)
default_model_idx = st.session_state["project_dct"]["model_tuning"][
"sel_model"
].get(sel_target_col, required_saved_models[0])
sel_model = st.selectbox(
"Select the model to tune",
required_saved_models,
index=required_saved_models.index(default_model_idx),
)
st.session_state["project_dct"]["model_tuning"]["sel_model"][
sel_target_col
] = default_model_idx
sel_model_dict = model_dict[
sel_model + "__" + target_col
] # get the model obj of the selected model
X_train = sel_model_dict["X_train"]
X_test = sel_model_dict["X_test"]
y_train = sel_model_dict["y_train"]
y_test = sel_model_dict["y_test"]
df = st.session_state["media_data"]
st.markdown("### Event Flags")
st.markdown("Helps in quantifying the impact of specific occurrences of events")
try:
# Dropdown to add event flags
with st.expander("Apply Event Flags"):
model = sel_model_dict["Model_object"]
date = st.session_state["date"]
date = pd.to_datetime(date)
X_train = sel_model_dict["X_train"]
features_set = sel_model_dict["feature_set"]
col = st.columns(3)
# Get date range
min_date = min(date).date()
max_date = max(date).date()
# Get previously selected start and end date of flag (persistence)
start_date_default = (
st.session_state["project_dct"]["model_tuning"].get(
"start_date_default"
)
if st.session_state["project_dct"]["model_tuning"].get(
"start_date_default"
)
is not None
else min_date
)
start_date_default = (
start_date_default if start_date_default > min_date else min_date
)
start_date_default = (
start_date_default if start_date_default < max_date else min_date
)
end_date_default = (
st.session_state["project_dct"]["model_tuning"].get(
"end_date_default"
)
if st.session_state["project_dct"]["model_tuning"].get(
"end_date_default"
)
is not None
else max_date
)
end_date_default = (
end_date_default if end_date_default > min_date else max_date
)
end_date_default = (
end_date_default if end_date_default < max_date else max_date
)
# Flag start and end date input boxes
with col[0]:
start_date = st.date_input(
"Select Start Date",
start_date_default,
min_value=min_date,
max_value=max_date,
)
if (start_date < min_date) or (start_date > max_date):
st.error(
"Please select dates in the range of the dates in the data"
)
st.stop()
with col[1]:
# Check if end date default > selected start date
end_date_default = (
end_date_default
if pd.Timestamp(end_date_default) >= pd.Timestamp(start_date)
else start_date
)
end_date = st.date_input(
"Select End Date",
end_date_default,
min_value=max(
pd.to_datetime(min_date), pd.to_datetime(start_date)
),
max_value=pd.to_datetime(max_date),
)
if (
(start_date < min_date)
or (end_date < min_date)
or (start_date > max_date)
or (end_date > max_date)
):
st.error(
"Please select dates in the range of the dates in the data"
)
st.stop()
if end_date < start_date:
st.error("Please select end date after start date")
st.stop()
with col[2]:
# Get default value of repeat check box (persistence)
repeat_default = (
st.session_state["project_dct"]["model_tuning"].get(
"repeat_default"
)
if st.session_state["project_dct"]["model_tuning"].get(
"repeat_default"
)
is not None
else "No"
)
repeat_default_idx = 0 if repeat_default.lower() == "yes" else 1
repeat = st.selectbox(
"Repeat Annually", ["Yes", "No"], index=repeat_default_idx
)
# Update selected values to session dictionary (persistence)
st.session_state["project_dct"]["model_tuning"][
"start_date_default"
] = start_date
st.session_state["project_dct"]["model_tuning"][
"end_date_default"
] = end_date
st.session_state["project_dct"]["model_tuning"][
"repeat_default"
] = repeat
if repeat == "Yes":
repeat = True
else:
repeat = False
if "flags" in st.session_state["project_dct"]["model_tuning"].keys():
st.session_state["Flags"] = st.session_state["project_dct"][
"model_tuning"
]["flags"]
if is_panel:
# Create flag on Train
met, line_values, fig_flag = plot_actual_vs_predicted(
X_train[date_col],
y_train,
model.fittedvalues,
model,
target_column=sel_target_col,
flag=(start_date, end_date),
repeat_all_years=repeat,
is_panel=True,
)
st.plotly_chart(fig_flag, use_container_width=True)
# create flag on test
met, test_line_values, fig_flag = plot_actual_vs_predicted(
X_test[date_col],
y_test,
sel_model_dict["pred_test"],
model,
target_column=sel_target_col,
flag=(start_date, end_date),
repeat_all_years=repeat,
is_panel=True,
)
else:
pred_train = model.predict(X_train[features_set])
# Create flag on Train
met, line_values, fig_flag = plot_actual_vs_predicted(
X_train[date_col],
y_train,
pred_train,
model,
flag=(start_date, end_date),
repeat_all_years=repeat,
is_panel=False,
)
st.plotly_chart(fig_flag, use_container_width=True)
# create flag on test
pred_test = model.predict(X_test[features_set])
met, test_line_values, fig_flag = plot_actual_vs_predicted(
X_test[date_col],
y_test,
pred_test,
model,
flag=(start_date, end_date),
repeat_all_years=repeat,
is_panel=False,
)
flag_name = "f1_flag"
flag_name = st.text_input("Enter Flag Name")
# add selected target col to flag name
# Save the flag name, flag train values, flag test values to session state
if st.button("Save flag"):
st.session_state["Flags"][flag_name + "_flag__" + target_col] = {}
st.session_state["Flags"][flag_name + "_flag__" + target_col][
"train"
] = line_values
st.session_state["Flags"][flag_name + "_flag__" + target_col][
"test"
] = test_line_values
st.success(f'{flag_name + "_flag__" + target_col} stored')
st.session_state["project_dct"]["model_tuning"]["flags"] = (
st.session_state["Flags"]
)
# Only show flags created for the particular target col
target_model_flags = [
f.split("__")[0]
for f in st.session_state["Flags"].keys()
if f.split("__")[1] == target_col
]
options = list(target_model_flags)
num_rows = -(-len(options) // NUM_FLAG_COLS_TO_DISPLAY)
tick = False
# Select all flags checkbox
if st.checkbox(
"Select all",
value=st.session_state["project_dct"]["model_tuning"][
"select_all_flags_check"
].get(sel_target_col, False),
):
tick = True
st.session_state["project_dct"]["model_tuning"][
"select_all_flags_check"
][sel_target_col] = True
else:
st.session_state["project_dct"]["model_tuning"][
"select_all_flags_check"
][sel_target_col] = False
# Get previous flag selection (persistence)
selection_defualts = st.session_state["project_dct"]["model_tuning"][
"selected_flags"
].get(sel_target_col, [])
selected_options = selection_defualts
# create a checkbox for each available flag for selected response metric
for row in range(num_rows):
cols = st.columns(NUM_FLAG_COLS_TO_DISPLAY)
for col in cols:
if options:
option = options.pop(0)
option_default = True if option in selection_defualts else False
selected = col.checkbox(option, value=(tick or option_default))
if selected:
selected_options.append(option)
else:
if option in selected_options:
selected_options.remove(option)
selected_options = list(set(selected_options))
# Check if flag values match Data
# This is necessary because different models can have different train/test dates
remove_flags = []
for opt in selected_options:
train_match = len(
st.session_state["Flags"][opt + "__" + target_col]["train"]
) == len(X_train[date_col])
test_match = len(
st.session_state["Flags"][opt + "__" + target_col]["test"]
) == len(X_test[date_col])
if not train_match:
st.warning(f"Flag {opt} can not be used due to train date mismatch")
# selected_options.remove(opt)
remove_flags.append(opt)
if not test_match:
st.warning(f"Flag {opt} can not be used due to test date mismatch")
# selected_options.remove(opt)
remove_flags.append(opt)
if (
len(remove_flags) > 0
and len(list(set(selected_options).intersection(set(remove_flags)))) > 0
):
selected_options = list(set(selected_options) - set(remove_flags))
st.session_state["project_dct"]["model_tuning"]["selected_flags"][
sel_target_col
] = selected_options
except:
# Capture the error details
exc_type, exc_value, exc_traceback = sys.exc_info()
error_message = "".join(
traceback.format_exception(exc_type, exc_value, exc_traceback)
)
log_message(
"error", f"Error while creating flags: {error_message}", "Model Tuning"
)
st.warning("An error occured, please try again", icon="⚠️")
try:
st.markdown("### Trend and Seasonality Calibration")
parameters = st.columns(3)
# Trend checkbox
with parameters[0]:
Trend = st.checkbox(
"**Trend**",
value=st.session_state["project_dct"]["model_tuning"].get(
"trend_check", False
),
)
st.markdown(
"Helps account for long-term trends or seasonality that could influence advertising effectiveness"
)
# Day of Week (week number) checkbox
with parameters[1]:
day_of_week = st.checkbox(
"**Day of Week**",
value=st.session_state["project_dct"]["model_tuning"].get(
"week_num_check", False
),
)
st.markdown(
"Assists in detecting and incorporating weekly patterns or seasonality"
)
# Sine and cosine Waves checkbox
with parameters[2]:
sine_cosine = st.checkbox(
"**Sine and Cosine Waves**",
value=st.session_state["project_dct"]["model_tuning"].get(
"sine_cosine_check", False
),
)
st.markdown(
"Helps in capturing long term cyclical patterns or seasonality in the data"
)
if sine_cosine:
# Drop down to select Frequency of waves
xtrain_time_period_months = (
X_train[date_col].max() - X_train[date_col].min()
).days / 30
# If we have 6 months of data, only quarter frequency is possible
if xtrain_time_period_months <= HALF_YEAR_THRESHOLD:
available_frequencies = ["Quarter"]
# If we have less than 12 months of data, we have quarter and semi-annual frequencies
elif xtrain_time_period_months < FULL_YEAR_THRESHOLD:
available_frequencies = ["Quarter", "Semi-Annual"]
# If we have 12 months of data or more, we have quarter, semi-annual and annual frequencies
elif xtrain_time_period_months >= FULL_YEAR_THRESHOLD:
available_frequencies = ["Quarter", "Semi-Annual", "Annual"]
wave_freq = st.selectbox("Select Frequency", available_frequencies)
except:
# Capture the error details
exc_type, exc_value, exc_traceback = sys.exc_info()
error_message = "".join(
traceback.format_exception(exc_type, exc_value, exc_traceback)
)
log_message(
"error",
f"Error while selecting tuning parameters: {error_message}",
"Model Tuning",
)
st.warning("An error occured, please try again", icon="⚠️")
try:
# Build tuned model
if st.button(
"Build model with Selected Parameters and Flags",
key="build_tuned_model",
use_container_width=True,
):
new_features = features_set
st.header("2.1 Results Summary")
ss = MaxAbsScaler()
if is_panel == True:
X_train_tuned = X_train[features_set]
X_train_tuned[target_col] = X_train[target_col]
X_train_tuned[date_col] = X_train[date_col]
X_train_tuned[panel_col] = X_train[panel_col]
X_test_tuned = X_test[features_set]
X_test_tuned[target_col] = X_test[target_col]
X_test_tuned[date_col] = X_test[date_col]
X_test_tuned[panel_col] = X_test[panel_col]
else:
X_train_tuned = X_train[features_set]
X_test_tuned = X_test[features_set]
for flag in selected_options:
# Get the flag values of train and test and add to the data
X_train_tuned[flag] = st.session_state["Flags"][
flag + "__" + target_col
]["train"]
X_test_tuned[flag] = st.session_state["Flags"][
flag + "__" + target_col
]["test"]
if Trend:
st.session_state["project_dct"]["model_tuning"][
"trend_check"
] = True
# group by panel, calculate trend of each panel spearately. Add trend to new feature set
if is_panel:
newdata = pd.DataFrame()
panel_wise_end_point_train = {}
for panel, groupdf in X_train_tuned.groupby(panel_col):
groupdf.sort_values(date_col, inplace=True)
groupdf["Trend"] = np.arange(
TREND_MIN, len(groupdf) + TREND_MIN, 1
) # Trend is a straight line with starting point as TREND_MIN
newdata = pd.concat([newdata, groupdf])
panel_wise_end_point_train[panel] = len(groupdf) + TREND_MIN
X_train_tuned = newdata.copy()
test_newdata = pd.DataFrame()
for panel, test_groupdf in X_test_tuned.groupby(panel_col):
test_groupdf.sort_values(date_col, inplace=True)
start = panel_wise_end_point_train[panel]
end = start + len(test_groupdf)
test_groupdf["Trend"] = np.arange(start, end, 1)
test_newdata = pd.concat([test_newdata, test_groupdf])
X_test_tuned = test_newdata.copy()
new_features = new_features + ["Trend"]
else:
X_train_tuned["Trend"] = np.arange(
TREND_MIN, len(X_train_tuned) + TREND_MIN, 1
) # Trend is a straight line with starting point as TREND_MIN
X_test_tuned["Trend"] = np.arange(
len(X_train_tuned) + TREND_MIN,
len(X_train_tuned) + len(X_test_tuned) + TREND_MIN,
1,
)
new_features = new_features + ["Trend"]
else:
st.session_state["project_dct"]["model_tuning"][
"trend_check"
] = False # persistence
# Add day of week (Week_num) to test & train
if day_of_week:
st.session_state["project_dct"]["model_tuning"][
"week_num_check"
] = True
if is_panel:
X_train_tuned[date_col] = pd.to_datetime(
X_train_tuned[date_col]
)
X_train_tuned["day_of_week"] = X_train_tuned[
date_col
].dt.day_of_week # Day of week
# if all the dates in the data have the same day of week number this feature cant be used
if X_train_tuned["day_of_week"].nunique() == 1:
st.error(
"All dates in the data are of the same week day. Hence Week number can't be used."
)
else:
X_test_tuned[date_col] = pd.to_datetime(
X_test_tuned[date_col]
)
X_test_tuned["day_of_week"] = X_test_tuned[
date_col
].dt.day_of_week # Day of week
new_features = new_features + ["day_of_week"]
else:
date = pd.to_datetime(date.values)
X_train_tuned["day_of_week"] = pd.to_datetime(
X_train[date_col]
).dt.day_of_week # Day of week
X_test_tuned["day_of_week"] = pd.to_datetime(
X_test[date_col]
).dt.day_of_week # Day of week
# if all the dates in the data have the same day of week number this feature cant be used
if X_train_tuned["day_of_week"].nunique() == 1:
st.error(
"All dates in the data are of the same week day. Hence Week number can't be used."
)
else:
new_features = new_features + ["day_of_week"]
else:
st.session_state["project_dct"]["model_tuning"][
"week_num_check"
] = False
# create sine and cosine wave and add to data
if sine_cosine:
st.session_state["project_dct"]["model_tuning"][
"sine_cosine_check"
] = True
frequency = ANNUAL_FREQUENCY # Annual Frequency
if wave_freq == "Quarter":
frequency = frequency * QTR_FREQUENCY_FACTOR
elif wave_freq == "Semi-Annual":
frequency = frequency * HALF_YEARLY_FREQUENCY_FACTOR
# create panel wise sine cosine waves in xtrain tuned. add to new feature set
if is_panel:
new_features = new_features + ["sine_wave", "cosine_wave"]
newdata = pd.DataFrame()
newdata_test = pd.DataFrame()
groups = X_train_tuned.groupby(panel_col)
train_panel_wise_end_point = {}
for panel, groupdf in groups:
num_samples = len(groupdf)
train_panel_wise_end_point[panel] = num_samples
days_since_start = np.arange(num_samples)
sine_wave = np.sin(frequency * days_since_start)
cosine_wave = np.cos(frequency * days_since_start)
sine_cosine_df = pd.DataFrame(
{"sine_wave": sine_wave, "cosine_wave": cosine_wave}
)
assert len(sine_cosine_df) == len(groupdf)
groupdf["sine_wave"] = sine_wave
groupdf["cosine_wave"] = cosine_wave
newdata = pd.concat([newdata, groupdf])
X_train_tuned = newdata.copy()
test_groups = X_test_tuned.groupby(panel_col)
for panel, test_groupdf in test_groups:
num_samples = len(test_groupdf)
start = train_panel_wise_end_point[panel]
days_since_start = np.arange(start, start + num_samples, 1)
# print("##", panel, num_samples, start, len(np.arange(start, start+num_samples, 1)))
sine_wave = np.sin(frequency * days_since_start)
cosine_wave = np.cos(frequency * days_since_start)
sine_cosine_df = pd.DataFrame(
{"sine_wave": sine_wave, "cosine_wave": cosine_wave}
)
assert len(sine_cosine_df) == len(test_groupdf)
# groupdf = pd.concat([groupdf, sine_cosine_df], axis=1)
test_groupdf["sine_wave"] = sine_wave
test_groupdf["cosine_wave"] = cosine_wave
newdata_test = pd.concat([newdata_test, test_groupdf])
X_test_tuned = newdata_test.copy()
else:
new_features = new_features + ["sine_wave", "cosine_wave"]
num_samples = len(X_train_tuned)
days_since_start = np.arange(num_samples)
sine_wave = np.sin(frequency * days_since_start)
cosine_wave = np.cos(frequency * days_since_start)
sine_cosine_df = pd.DataFrame(
{"sine_wave": sine_wave, "cosine_wave": cosine_wave}
)
# Concatenate the sine and cosine waves with the scaled X DataFrame
X_train_tuned = pd.concat(
[X_train_tuned, sine_cosine_df], axis=1
)
test_num_samples = len(X_test_tuned)
start = num_samples
days_since_start = np.arange(start, start + test_num_samples, 1)
sine_wave = np.sin(frequency * days_since_start)
cosine_wave = np.cos(frequency * days_since_start)
sine_cosine_df = pd.DataFrame(
{"sine_wave": sine_wave, "cosine_wave": cosine_wave}
)
# Concatenate the sine and cosine waves with the scaled X DataFrame
X_test_tuned = pd.concat([X_test_tuned, sine_cosine_df], axis=1)
else:
st.session_state["project_dct"]["model_tuning"][
"sine_cosine_check"
] = False
# Build model
# Get list of parameters added and scale
# previous features are scaled already during model build
added_params = list(set(new_features) - set(features_set))
if len(added_params) > 0:
concat_df = pd.concat([X_train_tuned, X_test_tuned]).reset_index(
drop=True
)
if is_panel:
train_max_date = X_train_tuned[date_col].max()
# concat_df = concat_df.reset_index(drop=True)
# concat_df=concat_df[added_params]
train_idx = X_train_tuned.index[-1]
concat_df[added_params] = ss.fit_transform(concat_df[added_params])
# added_params_df = pd.DataFrame(added_params_df)
# added_params_df.columns = added_params
if is_panel:
X_train_tuned[added_params] = concat_df[
concat_df[date_col] <= train_max_date
][added_params].reset_index(drop=True)
X_test_tuned[added_params] = concat_df[
concat_df[date_col] > train_max_date
][added_params].reset_index(drop=True)
else:
added_params_df = concat_df[added_params]
X_train_tuned[added_params] = added_params_df[: train_idx + 1]
X_test_tuned[added_params] = added_params_df.loc[
train_idx + 1 :
].reset_index(drop=True)
# Add flags (flags are 0, 1 only so need to scale)
if selected_options:
new_features = new_features + selected_options
# Build Mixed LM model for panel level data
if is_panel:
X_train_tuned.sort_values([date_col, panel_col]).reset_index(
drop=True, inplace=True
)
new_features = list(set(new_features))
inp_vars_str = " + ".join(new_features)
md_str = target_col + " ~ " + inp_vars_str
md_tuned = smf.mixedlm(
md_str,
data=X_train_tuned[[target_col] + new_features],
groups=X_train_tuned[panel_col],
)
model_tuned = md_tuned.fit()
# plot actual vs predicted for original model and tuned model
metrics_table, line, actual_vs_predicted_plot = (
plot_actual_vs_predicted(
X_train[date_col],
y_train,
model.fittedvalues,
model,
target_column=sel_target_col,
is_panel=True,
)
)
metrics_table_tuned, line, actual_vs_predicted_plot_tuned = (
plot_actual_vs_predicted(
X_train_tuned[date_col],
X_train_tuned[target_col],
model_tuned.fittedvalues,
model_tuned,
target_column=sel_target_col,
is_panel=True,
)
)
# Build OLS model for panel level data
else:
new_features = list(set(new_features))
model_tuned = sm.OLS(y_train, X_train_tuned[new_features]).fit()
metrics_table, line, actual_vs_predicted_plot = (
plot_actual_vs_predicted(
X_train[date_col],
y_train,
model.predict(X_train[features_set]),
model,
target_column=sel_target_col,
)
)
metrics_table_tuned, line, actual_vs_predicted_plot_tuned = (
plot_actual_vs_predicted(
X_train[date_col],
y_train,
model_tuned.predict(X_train_tuned[new_features]),
model_tuned,
target_column=sel_target_col,
)
)
# # ----------------------------------- TESTING -----------------------------------
#
# Plot Sine & cosine wave to test
# sine_cosine_plot = plot_actual_vs_predicted(
# X_train[date_col],
# y_train,
# X_train_tuned['sine_wave'],
# model_tuned,
# target_column=sel_target_col,
# is_panel=True,
# )
# st.plotly_chart(sine_cosine_plot, use_container_width=True)
# # Plot Trend line to test
# trend_plot = plot_tuned_params(
# X_train[date_col],
# y_train,
# X_train_tuned['Trend'],
# model_tuned,
# target_column=sel_target_col,
# is_panel=True,
# )
# st.plotly_chart(trend_plot, use_container_width=True)
#
# # Plot week number to test
# week_num_plot = plot_tuned_params(
# X_train[date_col],
# y_train,
# X_train_tuned['day_of_week'],
# model_tuned,
# target_column=sel_target_col,
# is_panel=True,
# )
# st.plotly_chart(week_num_plot, use_container_width=True)
# Get model metrics from metric table & display them
mape = np.round(metrics_table.iloc[0, 1], 2)
r2 = np.round(metrics_table.iloc[1, 1], 2)
adjr2 = np.round(metrics_table.iloc[2, 1], 2)
mape_tuned = np.round(metrics_table_tuned.iloc[0, 1], 2)
r2_tuned = np.round(metrics_table_tuned.iloc[1, 1], 2)
adjr2_tuned = np.round(metrics_table_tuned.iloc[2, 1], 2)
parameters_ = st.columns(3)
with parameters_[0]:
st.metric("R-squared", r2_tuned, np.round(r2_tuned - r2, 2))
with parameters_[1]:
st.metric(
"Adj. R-squared", adjr2_tuned, np.round(adjr2_tuned - adjr2, 2)
)
with parameters_[2]:
st.metric(
"MAPE", mape_tuned, np.round(mape_tuned - mape, 2), "inverse"
)
st.write(model_tuned.summary())
X_train_tuned[date_col] = X_train[date_col]
X_train_tuned[target_col] = y_train
X_test_tuned[date_col] = X_test[date_col]
X_test_tuned[target_col] = y_test
st.header("2.2 Actual vs. Predicted Plot (Train)")
if is_panel:
metrics_table, line, actual_vs_predicted_plot = (
plot_actual_vs_predicted(
X_train_tuned[date_col],
X_train_tuned[target_col],
model_tuned.fittedvalues,
model_tuned,
target_column=sel_target_col,
is_panel=True,
)
)
else:
metrics_table, line, actual_vs_predicted_plot = (
plot_actual_vs_predicted(
X_train_tuned[date_col],
X_train_tuned[target_col],
model_tuned.predict(X_train_tuned[new_features]),
model_tuned,
target_column=sel_target_col,
is_panel=False,
)
)
st.plotly_chart(actual_vs_predicted_plot, use_container_width=True)
st.markdown("## 2.3 Residual Analysis (Train)")
if is_panel:
columns = st.columns(2)
with columns[0]:
fig = plot_residual_predicted(
y_train, model_tuned.fittedvalues, X_train_tuned
)
st.plotly_chart(fig)
with columns[1]:
st.empty()
fig = qqplot(y_train, model_tuned.fittedvalues)
st.plotly_chart(fig)
with columns[0]:
fig = residual_distribution(y_train, model_tuned.fittedvalues)
st.pyplot(fig)
else:
columns = st.columns(2)
with columns[0]:
fig = plot_residual_predicted(
y_train,
model_tuned.predict(X_train_tuned[new_features]),
X_train,
)
st.plotly_chart(fig)
with columns[1]:
st.empty()
fig = qqplot(
y_train, model_tuned.predict(X_train_tuned[new_features])
)
st.plotly_chart(fig)
with columns[0]:
fig = residual_distribution(
y_train, model_tuned.predict(X_train_tuned[new_features])
)
st.pyplot(fig)
# st.session_state['is_tuned_model'][target_col] = True
# Save tuned model in a dict
st.session_state["Model_Tuned"][sel_model + "__" + target_col] = {
"Model_object": model_tuned,
"feature_set": new_features,
"X_train_tuned": X_train_tuned,
"X_test_tuned": X_test_tuned,
}
with st.expander("Results Summary Test data"):
if is_panel:
random_eff_df = get_random_effects(
st.session_state.media_data.copy(), panel_col, model_tuned
)
test_pred = mdf_predict(
X_test_tuned, model_tuned, random_eff_df
)
else:
test_pred = model_tuned.predict(X_test_tuned[new_features])
st.header("2.2 Actual vs. Predicted Plot (Test)")
metrics_table, line, actual_vs_predicted_plot = (
plot_actual_vs_predicted(
X_test_tuned[date_col],
y_test,
test_pred,
model,
target_column=sel_target_col,
is_panel=is_panel,
)
)
st.plotly_chart(actual_vs_predicted_plot, use_container_width=True)
st.markdown("## 2.3 Residual Analysis (Test)")
columns = st.columns(2)
with columns[0]:
fig = plot_residual_predicted(y_test, test_pred, X_test_tuned)
st.plotly_chart(fig)
with columns[1]:
st.empty()
fig = qqplot(y_test, test_pred)
st.plotly_chart(fig)
with columns[0]:
fig = residual_distribution(y_test, test_pred)
st.pyplot(fig)
except:
# Capture the error details
exc_type, exc_value, exc_traceback = sys.exc_info()
error_message = "".join(
traceback.format_exception(exc_type, exc_value, exc_traceback)
)
log_message(
"error",
f"Error while building tuned model: {error_message}",
"Model Tuning",
)
st.warning("An error occured, please try again", icon="⚠️")
if (
st.session_state["Model_Tuned"] is not None
and len(list(st.session_state["Model_Tuned"].keys())) > 0
):
if st.button("Use This model for Media Planning", use_container_width=True):
# remove previous tuned models saved for this target col
_remove = [
m
for m in st.session_state["Model_Tuned"].keys()
if m.split("__")[1] == target_col and m.split("__")[0] != sel_model
]
if len(_remove) > 0:
for m in _remove:
del st.session_state["Model_Tuned"][m]
# Flag depicting tuned model for selected response metric
st.session_state["is_tuned_model"][target_col] = True
tuned_model_pkl = pickle.dumps(st.session_state["Model_Tuned"])
update_db(
st.session_state["project_number"],
"Model_Tuning",
"tuned_model",
tuned_model_pkl,
schema,
# resp_mtrc=None,
) # db
log_message(
"info",
f"Tuned model {' '.join(_remove)} removed due to overwrite",
"Model Tuning",
)
# Save session state variables (persistence)
st.session_state["project_dct"]["model_tuning"][
"session_state_saved"
] = {}
for key in [
"bin_dict",
"used_response_metrics",
"is_tuned_model",
"media_data",
"X_test_spends",
"spends_data",
]:
st.session_state["project_dct"]["model_tuning"][
"session_state_saved"
][key] = st.session_state[key]
project_dct_pkl = pickle.dumps(st.session_state["project_dct"])
update_db(
st.session_state["project_number"],
"Model_Tuning",
"project_dct",
project_dct_pkl,
schema,
# resp_mtrc=None,
) # db
log_message(
"info",
f'Tuned Model {sel_model + "__" + target_col} Saved',
"Model Tuning",
)
# Clear page metadata
st.session_state["project_dct"]["scenario_planner"][
"modified_metadata_file"
] = None
st.session_state["project_dct"]["response_curves"][
"modified_metadata_file"
] = None
st.success(sel_model + " for " + target_col + " Tuned saved!")
except:
# Capture the error details
exc_type, exc_value, exc_traceback = sys.exc_info()
error_message = "".join(
traceback.format_exception(exc_type, exc_value, exc_traceback)
)
log_message("error", f"An error has occured : {error_message}", "Model Tuning")
st.warning("An error occured, please try again", icon="⚠️")
# st.write(error_message)
|