File size: 21,194 Bytes
dc94239
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
##############################################################################
# Understanding the impact of Growth and Margin profile on B2B SaaS Valuations
# Dataset: 106 B2B SaaS companies
#   Author: Ramu Arunachalam (ramu@acapital.com)
#   Created: 06/20/21
#   Datset last updated: 06/09/21
###############################################################################

import joblib as jl
import pandas as pd
import plotly.express as px
import plotly.graph_objects as go
import statsmodels.api as sm
import streamlit as st
from statsmodels.stats.outliers_influence import variance_inflation_factor
from transformers import pipeline
from transformers import TapasTokenizer, TapasForQuestionAnswering
import json
import requests

file_date = '2021-06-11'
saas_filename_all = f'{file_date}-comps_B2B_ALL.csv'
saas_filename_high_growth = f'{file_date}-comps_B2B_High_Growth.csv'


def get_scatter_fig(df, x, y):
    fig = px.scatter(df,
                     x=x,
                     y=y,
                     hover_data=['Name'],
                     title=f'{y} vs {x}')
    df_r = df[[y] + [x]].dropna()
    model = sm.OLS(df_r[y], sm.add_constant(df_r[x])).fit()
    regline = sm.OLS(df_r[y], sm.add_constant(df_r[x])).fit().fittedvalues
    fig.add_trace(go.Scatter(x=df_r[x],
                             y=model.predict(),
                             mode='lines',
                             marker_color='black',
                             name='Best-fit',
                             line=dict(width=4, dash='dot')))
    return fig


latex_dict = {'EV / NTM Revenue': r'''\frac{EV}{Rev_{NTM}}''',
              'EV / 2021 Revenue': r'''\frac{EV}{Rev_{2021}}''',
              'EV / NTM Gross Profit': r'''\frac{EV}{GP_{NTM}}''',
              'EV / 2021 Gross Profit': r'''\frac{EV}{GP_{2021}}''',
              'NTM Revenue Growth': r'''Rev\,Growth_{NTM}''',
              '2021 Revenue Growth': r'''Rev\,Growth_{2021}''',
              'Growth adjusted EV / LTM Revenue': r'''\frac{EV}{Rev_{LTM}}\cdot\frac{1}{Growth_{NTM}}''',
              'Growth adjusted EV / 2020 Revenue': r'''\frac{EV}{Rev_{2020}}\cdot\frac{1}{Growth_{2021}}'''
              }


class RegressionInput:
    def __init__(self, df, x_vars, y_var):
        self.df = df
        self.x_vars = x_vars
        self.y_var = y_var
        self._hash = tuple([jl.hash(df), tuple(self.x_vars), tuple([self.y_var])])
        return

    def hash(self):
        return self._hash


class RegressionOutput:
    def __init__(self, reg_input, df_r, model, df_pvalues, vif_data):
        self.df_r = df_r
        self.model = model
        self.df_pvalues = df_pvalues
        self.vif_data = vif_data
        self.plot_figs = dict()

        # Regression equation
        self.eq_str = latex_dict.get(reg_input.y_var, reg_input.y_var) + r'''= \beta_0'''

        for x, i in zip(reg_input.x_vars, range(len(reg_input.x_vars))):
            self.eq_str += rf'''+\beta_{{{i + 1}}}\cdot {{{latex_dict.get(x, x)}}}'''
        self.eq_str += r'''+\epsilon'''
        self.eq_str = self.eq_str.replace("%", "\%").replace("&", "\&").replace("$", "\$")

        # Compute regression plots and save them
        # Plot residuals
        for x in reg_input.x_vars:
            self.plot_figs[x] = sm.graphics.plot_regress_exog(self.model, x)

        self._hash = tuple([reg_input.hash(), jl.hash(df_r), id(model), jl.hash(df_pvalues), jl.hash(vif_data)])
        return

    def hash(self):
        return self._hash


def reg_input_hash(reg_input):
    h = reg_input.hash()
    # st.info(f"reg_input_hash: h = {h}")
    return h


def reg_output_hash(reg_output):
    h = reg_output.hash()
    # st.info(f'reg_output hash = {h}')
    return h


class Experiment:
    id_num = 0

    def __init__(self):
        df_main = pd.read_csv(saas_filename_all)

        # Clean: 2x --> 2, 80% --> 80, $3,000 --> 3000
        df_obj = df_main[set(df_main.columns) - {'Name'}].select_dtypes(['object'])
        df_main[df_obj.columns] = df_obj \
            .apply(lambda x: x.str.strip('x')) \
            .apply(lambda x: x.str.strip('%')) \
            .replace(',', '', regex=True) \
            .replace('\$', '', regex=True)

        cols = df_main.columns
        for c in cols:
            try:
                df_main[c] = pd.to_numeric(df_main[c])
            except:
                pass

        df_main['2021 Revenue Growth'] = (df_main['2021 Analyst Revenue Estimates'].astype(float) / df_main[
            '2020 Revenue'].astype(
            float) - 1) * 100

        df_main = df_main[df_main['Name'].notna()]
        self.tickers_all = list(df_main[df_main['Name'].isin(['Median', 'Mean']) == False]['Name'])
        df_main_hg = pd.read_csv(saas_filename_high_growth)
        df_main_hg = df_main_hg[df_main_hg['Name'].notna()]
        self.tickers_hg = list(df_main_hg[df_main_hg['Name'].isin(['Median', 'Mean']) == False]['Name'])
        self.tickers_excl_hg = list(set(self.tickers_all) - set(self.tickers_hg))


        self.df_main = df_main
        self.df = df_main
        self.reg_input = None
        self.reg_output = None

        return

    def get_tickers(self, growth='High'):
        if growth == 'High':
            return self.tickers_hg
        elif growth == 'Low':
            return self.tickers_excl_hg
        else:
            return self.tickers_all

    def filter(self, by):
        if by == 'High growth only':
            tickers = self.tickers_hg
        elif by == 'All (excl. high growth)':
            tickers = self.tickers_excl_hg
        else:
            tickers = self.tickers_all

        self.df = self.df_main[self.df_main['Name'].isin(tickers)]  # type of dataset
        return self

    def set_fwd_timeline(self, type):
        self.rev_g = f'{type} Revenue Growth'
        self.rev_mult = f'EV / {type} Revenue'
        self.gp_mult = f'EV / {type} Gross Profit'
        self.gm = f'Gross Margin'

        # To avoid double counting growth, for growth-adjusted multiples
        # we take the forward growth rate with the current revenue multiple
        rev_mult = 'LTM' if type == 'NTM' else '2020'
        self.growth_adj_mult = f'Growth adjusted EV / {rev_mult} Revenue'
        self.df[self.growth_adj_mult] = self.df[f'EV / {rev_mult} Revenue'] / self.df[self.rev_g]
        return self

    def get_y_metric_list(self):
        return [self.rev_mult, self.gp_mult, self.growth_adj_mult, self.rev_g]

    def get_x_metric_list(self):
        return self.df.select_dtypes(['float', 'int']).columns

    def to_frame(self):
        return self.df

    @st.cache(suppress_st_warning=True,
              hash_funcs={RegressionInput: reg_input_hash, RegressionOutput: reg_output_hash})
    def _regression(self, reg_input):
        df = reg_input.df
        reg_x_vars = reg_input.x_vars
        reg_y = reg_input.y_var

        if not reg_x_vars:
            return None

        df_r = df[[reg_y] + reg_x_vars].dropna()

        # Run the regression
        X = df_r[reg_x_vars]
        X = sm.add_constant(X)

        model = sm.OLS(df_r[reg_y], X).fit()

        # Compute Variance Inflation Factors
        df_v = df_r[reg_x_vars]
        vif_data = None
        if len(df_v.columns) >= 2:
            # VIF dataframe
            vif_data = pd.DataFrame()

            vif_data["feature"] = df_v.columns
            # calculating VIF for each feature
            vif_data["VIF"] = [variance_inflation_factor(df_v.values, i)
                               for i in range(len(df_v.columns))]

        # pvalue dataframe
        df_pvalues = model.params.to_frame().reset_index().rename(columns={'index': 'vars', 0: 'Beta'})
        df_pvalues['p-value'] = model.pvalues.to_frame().reset_index().rename(columns={0: 'p-value'})['p-value']
        df_pvalues['Statistical Significance'] = 'Low'
        df_pvalues.loc[df_pvalues['p-value'] <= 0.05, 'Statistical Significance'] = 'High'
        df_pvalues = df_pvalues[df_pvalues['vars'] != 'const']

        return RegressionOutput(reg_input, df_r, model, df_pvalues, vif_data)

    def regression(self, reg_x_vars, reg_y_var):
        self.reg_input = RegressionInput(self.df, reg_x_vars, reg_y_var)
        self.reg_output = self._regression(self.reg_input)
        return self

    def print(self, show_detail=False):
        # Print regression equation
        st.latex(self.reg_output.eq_str)

        def highlight_significant_rows(val):
            color = 'green' if val['p-value'] <= 0.05 else 'red'
            return [f"color: {color}"] * len(val)

        st.subheader("Summary", anchor='summary')
        st.write(f"1. N = {len(self.reg_output.df_r)} companies")

        # Assess model fit
        if self.reg_output.model.rsquared * 100 > 30:
            st.write(f"2. Model fit is **good** R ^ 2 = {self.reg_output.model.rsquared * 100: .2f}%")
            if self.reg_output.model.f_pvalue < 0.05:
                st.write(f"3. Model is **statistically significant** (F-test = {self.reg_output.model.f_pvalue:.2f})")
            else:
                st.write(
                    f"3. The regression is **NOT statistically significant** (F-test = {self.reg_output.model.f_pvalue:.2f})")

        else:
            st.write(f"2. Model fit is **poor** (R ^ 2 = {self.reg_output.model.rsquared * 100: .2f}%)")

        # Check for Multicolinearity
        if (
                self.reg_output.vif_data is not None
                and len(self.reg_output.vif_data[self.reg_output.vif_data['VIF'] > 10]) > 0
        ):
            st.write("4. **Potential multicolinearity**")
        else:
            st.write("4. **NO multicolinearity**")

        # print p-values
        st.write('***')
        for _, row in self.reg_output.df_pvalues.iterrows():
            str = 'strong' if row['Statistical Significance'] == 'High' else 'weak'
            st.write(f"* There is a **{str} relationship** between *'{self.reg_input.y_var}'* and *'{row['vars']}'*")

        st.table(self.reg_output.df_pvalues.set_index('vars').style.apply(highlight_significant_rows, axis=1))

        if show_detail:
            # Show details
            st.subheader("Details:", anchor='details')
            # Plot residuals
            for k, f in self.reg_output.plot_figs.items():
                st.write(f"Plotting residuals for **{k}**")
                st.pyplot(f)
                # st.pyplot(f)
            st.markdown('***')
            st.write(self.reg_output.model.summary())
            st.markdown('***')
            if self.reg_output.vif_data is not None:
                st.write("Variance Inflation Factors")
                st.table(self.reg_output.vif_data.set_index('feature'))

        return self


def workbench(show_detail):
    fwd_time = st.sidebar.selectbox('Timeline', ('2021', 'NTM'))
    slice_by_growth = st.sidebar.radio("B2B SaaS Dataset", ['High growth only', 'All', 'All (excl. high growth)'])

    e = Experiment().set_fwd_timeline(fwd_time).filter(slice_by_growth)

    st.sidebar.write("**Regression:**")
    y_sel = st.sidebar.radio("Target metric", e.get_y_metric_list())
    st.sidebar.text("Select independent variable(s)")

    st.header("Regression")
    # Check if user selected revenue growth and/or gross margin
    reg_x_cols = [i for i in [e.rev_g, e.gm] if st.sidebar.checkbox(i, value={e.rev_g: True, e.gm: True}, key=i)]
    remaining_cols = list(set(e.get_x_metric_list()) - {e.rev_g, e.gm})
    reg_x_cols += st.sidebar.multiselect("Additional independent variables:", remaining_cols)
    e.regression(reg_x_vars=reg_x_cols, reg_y_var=y_sel).print(show_detail)


    ## Plots
    #st.header("Plots")
    #for _, x in zip(range(4), e.reg_input.x_vars):
    #    st.plotly_chart(get_scatter_fig(e.to_frame(), x=x, y=e.reg_input.y_var))
    #st.plotly_chart(get_scatter_fig(e.to_frame(), x=e.gm, y=e.reg_input.y_var))

    st.subheader("Dataset")
    st.expander('Table Output') \
        .table(e.to_frame()[['Name'] + [y_sel] + reg_x_cols]
               .set_index('Name')
               .sort_values(y_sel, ascending=False))
    st.expander('Full Raw Table Output').table(e.df_main)
    st.sidebar.info(f"""*{len(e.df)} companies selected*    
        *Prices as of {file_date}*""")

    return

def get_dataset(filter='All'):

   e = Experiment()
   df = e.set_fwd_timeline('2021').to_frame()[['Name'] + ['EV / 2021 Revenue', '2021 Revenue Growth','Gross Margin']].set_index('Name')
   
   high_growth_tickers = e.get_tickers(growth='High')
   low_growth_tickers = e.get_tickers(growth='Low')
   high_growth_tickers = set(high_growth_tickers).intersection(set(df.index.values.tolist()))
   df.loc[high_growth_tickers,'Category'] = 'High Growth'
   low_growth_tickers = set(low_growth_tickers).intersection(set(df.index.values.tolist()))
   df.loc[low_growth_tickers,'Category'] = 'Low Growth'
   df = df[df['Category'].notna()]
   if filter == 'High Growth':
       return df.loc[high_growth_tickers]
   elif filter == 'Low Growth':
       return df.loc[low_growth_tickers]

   return df


def summary(e1, e2, e3, e4):
    st.header("High Growth B2B SaaS")
    st.markdown("""
    For high growth B2B SaaS, ***revenue growth*** (*not profitability*) ***drives valuation***
    * *Valuation multiples* are well explained by *revenue growth* 
        * Model fit is good (High R^2)
        * Revenue growth is a statistically significant factor (low p-value)
    * *Gross Margin* does not influence *valuation multiples*
        * Poor relationship between Revenue multiples and Gross margin (high p-value)
    """)
    with st.expander("More info"):
        e1.print(True)

    st.markdown("""
        * Looking at Free Cash Flow % instead of Gross Margin yield similar results
            * Model fit is good (High R^2)
            * *Revenue growth* is a statistically significant factor (low p-value)
        * *FCF Margin* does not influence *valuation multiples*
            * Poor relationship between Revenue multiples and FCF margin (high p-value)
        """)
    with st.expander("More info"):
        e2.print(True)

    st.markdown('***')
    st.header("B2B SaaS (excluding high growth)")
    st.markdown("""
        
        For the rest of B2B SaaS (i.e non high growth SaaS), the picture is less clear
        * *Revenue growth* by itself doesn't adequately explain *valuation multiples* 
            * Model fit is poor (low R^2)
        * But *Revenue growth* is still a statistically significant factor (low p-value)
        * *Gross Margin* does not influence *valuation multiples*
            * Poor relationship between Revenue multiples and Gross margin (high p-value)
        """)
    with st.expander("More info"):
        e3.print(True)
    st.markdown("""
            * Looking at Free Cash Flow % instead of Gross Margin improves model fit
            * FCF Margin* has a **small positive effect** on *valuation multiples*
                * Low p-value but small Beta.
            * But overall *revenue growth* still has a much **larger effect** on valuation multiples than profitability
                * Low p-value and higher Beta relative to FCF %
            """)
    with st.expander("More info"):
        e4.print(True)
    return


def make_api_call(queries, df:pd.DataFrame):
    API_TOKEN = "api_DjJYjFpAQfQkhpfzncoRuuKuuLWrSzHdav"
    headers = {"Authorization": f"Bearer {API_TOKEN}"}
    API_URL = "https://api-inference.huggingface.co/models/google/tapas-base-finetuned-wtq"

    st.sidebar.info("Using ** google/tapas-large-finetuned-wtq**")
    def query(payload):
        response = requests.post(API_URL, headers=headers, json=payload)
        return response.json()

    table_dict = df.to_dict(orient='list')
    output = query({
    "inputs": {
		"query": queries,
        "table": table_dict
    }})

    return output

def nlu_query_use_api():
 
    df = get_dataset(filter='High Growth').dropna().reset_index().rename(columns={'Name':'Company','EV / 2021 Revenue':'Revenue Multiple','2021 Revenue Growth':'Growth Rate'})

    df['Revenue Multiple'] = df['Revenue Multiple'].round(2).apply(str)
    df['Growth Rate'] = df['Growth Rate'].round(2).apply(str)
    df['Gross Margin'] = df['Gross Margin'].round(2).apply(str)
    df = df[['Company','Revenue Multiple','Growth Rate','Gross Margin']]

    with st.expander("Dataset"):
        st.table(df)
    
    questions = ['How many companies are in this dataset?', 
    'Which company has the highest growth rate?',
    'Which company has the highest gross margin?',
    'Which company trades at the highest revenue multiple?',
    'List all companies with growth rates greater than 40?',
    'What is the average gross margin for companies in this dataset?',
    'What is the average trading multiple for companies in this dataset?'
    
    ]

    st.sidebar.write("Sample questions:")
    st.sidebar.caption("[Copy and Paste any of these questions into the textbox below]")
    for i in questions: st.sidebar.markdown(f"* {i}") 
        
    queries = st.text_area("Enter Question:", 'How many companies are in this dataset?')
    #queries = ['Which company has the highest gross margin?']
    output = make_api_call(queries=queries, df=df)
    if 'error' in output:
        st.write(output['error'])
    else:
        df_output = pd.DataFrame.from_dict(output['cells'])
        if output['aggregator'] == 'COUNT':
            st.info(f"[COUNT] Answer: {df_output[0].count()}")
        elif output['aggregator'] == 'SUM':
            st.info(f"[SUM] Answer: {df_output[0].astype(float).sum().round(2)}")
        elif output['aggregator'] == 'AVERAGE':
            st.info(f"[AVERAGE] Answer: {df_output[0].astype(float).mean().round(2)}")
        else:
            st.info(f"Answer is {output['answer']}")

        with st.expander("Raw Output"):
            st.write(output)
    return

def nlu_query():
    from torch_scatter import scatter

    st.header("NLU Query")
    model_name = 'google/tapas-large-finetuned-wtq'
    model = TapasForQuestionAnswering.from_pretrained(model_name)
    tokenizer = TapasTokenizer.from_pretrained(model_name)
    df = get_dataset().dropna().reset_index().rename(columns={'Name':'Company','EV / 2021 Revenue':'Revenue Multiple','2021 Revenue Growth':'Growth Rate'})

    df['Revenue Multiple'] = df['Revenue Multiple'].round(2).apply(str)
    df['Growth Rate'] = df['Growth Rate'].round(2).apply(str)
    df['Gross Margin'] = df['Gross Margin'].round(2).apply(str)
    df = df[['Company','Revenue Multiple','Growth Rate','Gross Margin']]

    st.table(df)
    queries = ['How many companies are in the dataset', 'Which company has the highest growth rate?','Which company has the highest gross margin?']
    st.write(queries)
    #queries = st.text_area('Ask a question')
    
    inputs = tokenizer(table=df, queries=queries, padding='max_length', return_tensors="pt")
    outputs = model(**inputs)
    predicted_answer_coordinates, predicted_aggregation_indices = tokenizer.convert_logits_to_predictions(
        inputs, 
        outputs.logits.detach(), 
        outputs.logits_aggregation.detach())

    inputs = tokenizer(table=df, queries=queries, padding='max_length', return_tensors="pt")
    id2aggregation = {0: "NONE", 1: "SUM", 2: "AVERAGE", 3:"COUNT"}
    aggregation_predictions_string = [id2aggregation[x] for x in predicted_aggregation_indices]

    answers = []
    for coordinates in predicted_answer_coordinates:
        if len(coordinates) == 1:
            # only a single cell:
            answers.append(df.iat[coordinates[0]])
        else:
            # multiple cells
            cell_values = []
            for coordinate in coordinates:
                cell_values.append(df.iat[coordinate])
            answers.append(", ".join(cell_values))

    st.write(answers)
    st.write(aggregation_predictions_string)
    return

def main():
    #st.set_page_config(initial_sidebar_state="collapsed")
    sel = st.sidebar.radio("Menu", ['NLU Question Answer','Summary', 'Workbench'])
    
    show_detail = True

    # pre compute three experiments
    # Experiment 1
    e1 = Experiment() \
        .set_fwd_timeline('2021') \
        .filter('High growth only')
    e1.regression(reg_x_vars=[e1.rev_g, e1.gm], reg_y_var=e1.rev_mult)

    # Experiment 2
    e2 = Experiment() \
        .set_fwd_timeline('2021') \
        .filter('High growth only')
    e2.regression(reg_x_vars=[e2.rev_g, 'LTM FCF %'], reg_y_var=e2.rev_mult)

    # Experiment 3
    e3 = Experiment() \
        .set_fwd_timeline('2021') \
        .filter('All (excl. high growth)')
    e3.regression(reg_x_vars=[e3.rev_g, e3.gm], reg_y_var=e3.rev_mult)

    # Experiment 4
    e4 = Experiment() \
        .set_fwd_timeline('2021') \
        .filter('All (excl. high growth)')
    e4.regression(reg_x_vars=[e4.rev_g, 'LTM FCF %'], reg_y_var=e4.rev_mult)

    if sel == 'NLU Question Answer':
        st.title("Query Dataset")
        return nlu_query_use_api()
    elif sel == 'Workbench':
        st.title('Impact of Growth and Margins on Valuation')
        return workbench(True)
    else:
        st.title('Impact of Growth and Margins on Valuation')
        return summary(e1, e2, e3, e4)


if __name__ == "__main__":
    main()