v2.3.0

Prod Features

# %%
from IPython import get_ipython
from jinja2 import Template

if get_ipython() is not None:
    get_ipython().run_line_magic("load_ext", "autoreload")
    get_ipython().run_line_magic("autoreload", "2")
from dotenv import load_dotenv
import pandas as pd
from tqds.sql import db, sql

import pandas as pd
import numpy as np

import seaborn as sns
import matplotlib.pyplot as plt
sns.set_theme(style='whitegrid', rc={'figure.figsize': (14, 12)})

pd.options.display.max_columns = None
pd.options.display.max_rows = 100
pd.options.display.max_colwidth = None
pd.set_option('display.float_format', lambda x: '%.3f' % x)
trino_conn = db.TrinoClient().conn
redshift_conn = db.RedshiftClient().conn
query = sql.Query()



# %%
# NEW FEATURES:
# [CLD-v2.3.0] anesthesia accuracy bounds 
# [CLD-v2.3.0] drug accuracy scores 
# [CLD-v2.3.0] exchange networks 
# [CLD-v2.3.0] OPG base rate update params 
# [CLD-v2.3.0] Update to Labs Spines
# [CLD-v2.3.0] add 21 place of service code 
# [CLD-v2.3.0] more bh plan bridge exceptions 
# [CLD-v2.3.0] CAR-T Plausibility 
# [CLD-v2.3.0] OP outlier lower bound 
# [CLD-v2.3.0] PG accuracy logic 
# [CLD-v2.3.0] update analyze simulation configurations 
# [CLD-v2.3.0] external prod clear rates 
# [CLD-v2.3.0] add infusion centers 

# other misc:
# plan bridge was using network_name Preferred, but network spine using network_name PPO


# %%
# [CLD-v2.3.0] anesthesia accuracy bounds 

# %%
df = pd.read_sql(f"""
SELECT 
    provider_type, 
    canonical_rate_score, 
    COUNT(*) as n_rates,
    MIN(canonical_rate_percent_of_medicare) as min_pct_medicare,
    MAX(canonical_rate_percent_of_medicare) as max_pct_medicare,
    MAX(canonical_rate) as max_canonical_rate
FROM tq_dev.internal_dev_csong_cld_v2_3_0.prod_combined_abridged
WHERE billing_code IN (
    SELECT hcpcs
    FROM tq_production.reference_internal.anesthesia_reference_pricing
)
GROUP BY 1, 2
ORDER BY 1, 2
""", con=trino_conn)
print(df.to_markdown(index=False))

"""
| provider_type   |   canonical_rate_score |   n_rates |   min_pct_medicare |   max_pct_medicare |   max_canonical_rate |
|:----------------|-----------------------:|----------:|-------------------:|-------------------:|---------------------:|
| Imaging Center  |                      0 |    235337 |             nan    |             nan    |        nan           |
| Imaging Center  |                      1 |       120 |               0.5  |              90.09 |       1878.33        |
| Imaging Center  |                      4 |      4201 |               0.83 |              29.71 |        619.4         |
| Physician Group |                      0 | 123219406 |             nan    |             nan    |        nan           |
| Physician Group |                      1 |   1132766 |               0    |          746405    |          1.55402e+07 |
| Physician Group |                      2 |   6623106 |               3.5  |              10.46 |        200           |
| Physician Group |                      3 |   9644892 |               1    |               3.5  |         96.75        |
| Physician Group |                      4 |    110304 |               3.51 |               5.37 |        111.85        |
| Physician Group |                      5 |     15174 |               1    |               3.48 |         97           |
"""

# %%
# [CLD-v2.3.0] drug accuracy scores 

# %%
df = pd.read_sql(f"""
SELECT 
    canonical_rate_source,
    canonical_rate_score,
    COUNT(*) as n_rates,
    MIN(canonical_rate_percent_of_medicare) as min_canonical_rate,
    MAX(canonical_rate_percent_of_medicare) as max_canonical_rate
FROM tq_dev.internal_dev_csong_cld_v2_3_0.prod_combined_abridged
WHERE is_drug_code = True
GROUP BY 1,2
ORDER BY 1,2
LIMIT 100
""", con=trino_conn)
print(df.to_markdown(index=False))

"""
| canonical_rate_source   |   canonical_rate_score |   n_rates |   min_canonical_rate |   max_canonical_rate |
|:------------------------|-----------------------:|----------:|---------------------:|---------------------:|
| hospital                |                      1 |   4451328 |                 0    |                inf   |
| hospital                |                      2 |    180508 |                 0.8  |                  0.9 |
| hospital                |                      3 |   1790825 |                 0.9  |                  4   |
| hospital                |                      4 |   5255449 |                 1.25 |                  3.5 |
| hospital                |                      5 |    787512 |                 1.2  |                  4   |
| imputation              |                      1 |   4018773 |                 0    |             362871   |
| imputation              |                      2 |   4353234 |                 0.8  |                  4   |
| payer                   |                      1 |   7512310 |                 0    |                inf   |
| payer                   |                      2 |   2389249 |                 0.8  |                 10   |
| payer                   |                      3 |  17208127 |                 0.9  |                  4   |
| payer                   |                      4 |   8379973 |                 1.25 |                  3.5 |
| payer                   |                      5 |    289458 |                 1.2  |                  4   |
"""

# %%
# [CLD-v2.3.0] exchange networks 

# %%
pd.read_sql(f"""
SELECT 
    provider_type,
    COUNT(DISTINCT provider_id) as n_providers
FROM tq_dev.internal_dev_csong_cld_v2_3_0.prod_combined_abridged
WHERE network_class = 'Exchange'
AND canonical_rate_score > 1
GROUP BY 1
LIMIT 10
""", con=trino_conn)

# %%
df = pd.read_sql(f"""
SELECT 
    payer_id,
    payer_name,
    COUNT(DISTINCT provider_id) as n_providers,
    COUNT(*) as n_rates
FROM tq_dev.internal_dev_csong_cld_v2_3_0.prod_combined_abridged
WHERE network_class = 'Exchange'
AND canonical_rate_score > 1
GROUP BY 1, 2
ORDER BY n_providers DESC
""", con=trino_conn)
print(df.to_markdown(index=False))

# %%
df = pd.read_sql(f"""
SELECT 
    payer_id,
    payer_name,
    COUNT(DISTINCT provider_id) as n_providers,
    COUNT(*) as n_rates
FROM tq_dev.internal_dev_csong_cld_v2_3_0.prod_combined_abridged
WHERE network_class = 'Exchange'
AND canonical_rate_score > 1
AND state IN ('LA', 'NC', 'PA')
GROUP BY 1, 2
ORDER BY n_providers DESC
""", con=trino_conn)
print(df.to_markdown(index=False))


# %%
# [CLD-v2.3.0] OPG base rate update params 

# %%
pd.read_sql(f"""
SELECT 
    MIN(1.00 * opg_n_freq / opg_n_total) as min_opg_freq_per_total,
    MIN(opg_n_total) as min_opg_n_total
FROM tq_dev.internal_dev_csong_cld_v2_3_0.prod_combined_all
WHERE opg_base_rate IS NOT NULL
AND (opg_n_freq / opg_n_total_possible) < 0.5
LIMIT 10
""", con=trino_conn)

# %%
pd.read_sql(f"""
SELECT 
    MIN(1.00 * opg_n_freq / opg_n_total_possible) as min_opg_freq_per_total_possible 
FROM tq_dev.internal_dev_csong_cld_v2_3_0.prod_combined_all
WHERE opg_base_rate IS NOT NULL
AND (
    (1.00 * opg_n_freq / opg_n_total < 0.8) 
    OR opg_n_total < 100
)
LIMIT 10
""", con=trino_conn)

# %%
df = pd.read_sql(f"""
WITH
new AS (
    SELECT COUNT(*) as n_rates
    FROM tq_dev.internal_dev_csong_cld_v2_3_0.prod_combined_all
    WHERE opg_base_rate IS NOT NULL
),
old AS (
    SELECT COUNT(*) as n_rates
    FROM tq_dev.internal_dev_csong_cld_v2_2_2.prod_combined_all
    WHERE opg_base_rate IS NOT NULL
)
SELECT 
    new.n_rates as new_n_rates,
    old.n_rates as old_n_rates
FROM new
JOIN old ON TRUE
""", con=trino_conn)
print(df.to_markdown(index=False, floatfmt=".0f"))

"""
|   new_n_rates |   old_n_rates |
|--------------:|--------------:|
|      66680326 |      90324394 |
"""
66680326 / 90324394

# %%
df = pd.read_sql(f"""
WITH
new AS (
    SELECT canonical_rate_type, COUNT(*) as n_rates
    FROM tq_dev.internal_dev_csong_cld_v2_3_0.prod_combined_abridged
    WHERE canonical_rate_type LIKE '%opg_%'
    AND canonical_rate_score > 1
    GROUP BY 1
),
old AS (
    SELECT canonical_rate_type, COUNT(*) as n_rates
    FROM tq_dev.internal_dev_csong_cld_v2_2_2.prod_combined_abridged
    WHERE canonical_rate_type LIKE '%opg_%'
    AND canonical_rate_score > 1
    GROUP BY 1
)
SELECT 
    new.canonical_rate_type,
    new.n_rates as new_n_rates,
    old.n_rates as old_n_rates
FROM new
JOIN old ON new.canonical_rate_type = old.canonical_rate_type
""", con=trino_conn)
print(df.to_markdown(index=False, floatfmt=".0f"))

# | canonical_rate_type   |   new_n_rates |   old_n_rates |
# |:----------------------|--------------:|--------------:|
# | impute: opg_base_rate |      29033871 |      50744655 |

# %%
# [CLD-v2.3.0] Update to Labs Spines 
# check for national coverage

# %%
df = pd.read_sql(f"""
SELECT 
    provider_id,
    provider_name,
    CASE
        WHEN provider_id IN (
            SELECT provider_id
            FROM tq_dev.internal_dev_csong_cld_v2_3_0.tmp_ref_provider_spine_2025_09
            WHERE national_payer_coverage = True
        )
        THEN 'national_coverage'
        ELSE 'non_national_coverage'
    END as coverage_type,
    COUNT(DISTINCT payer_id) as n_payers
FROM tq_dev.internal_dev_csong_cld_v2_3_0.prod_combined_abridged
WHERE canonical_rate_score > 1
GROUP BY 1, 2, 3
""", con=trino_conn)
df

# %%
df = pd.read_sql(f"""
SELECT 
    provider_id,
    provider_name,
    COUNT(DISTINCT payer_id) as n_payers,
    COUNT(DISTINCT billing_code) as n_billing_codes,
    COUNT(
        DISTINCT
        CASE
            WHEN canonical_rate_score > 1 THEN payer_id
        END
    ) as n_payers_w_rates,
    COUNT(
        DISTINCT
        CASE
            WHEN canonical_rate_score > 1 THEN billing_code
        END
    ) as n_billing_codes_w_rates
FROM tq_dev.internal_dev_csong_cld_v2_3_0.prod_combined_all
WHERE national_payer_coverage = True
AND is_lab_code = True
AND canonical_rate_score > 1
GROUP BY 1, 2
""", con=trino_conn)
df

# %%
# [CLD-v2.3.0] add 21 place of service code 

# %%
pd.read_sql(f"""
SELECT canonical_rate_score, count(*)
FROM tq_dev.internal_dev_csong_cld_v2_3_0.prod_combined_abridged p 
WHERE payer_id = '42'
AND billing_code like '00%'
and NOT regexp_like(billing_code, '[A-Za-z]')
AND canonical_rate IS NOT NULL
AND provider_name = 'Emory Specialty Associates, Llc'
AND provider_type = 'Physician Group'
GROUP BY 1
""", con=trino_conn)

# %%
# [CLD-v2.3.0] more bh plan bridge exceptions 

# %%
pd.read_sql(f"""
select *
FROM tq_dev.internal_dev_csong_cld_v2_3_0.prod_combined_abridged
where billing_code = '927' 
and billing_code_type = 'MS-DRG'
and provider_name = 'Jackson South Medical Center'
and payer_id = '76'
""", con=trino_conn)

# %%
test = pd.read_sql(f"""
SELECT *
FROM tq_production.hospital_data.hospital_rates
WHERE provider_id = '7040'
AND billing_code = '927'
AND billing_code_type = 'MS-DRG'
AND negotiated_dollar = 274466.07
""", con=trino_conn)
test.T
# CIGNA HMO/PPO - ALL OTHER PLANS


# %%
# [CLD-v2.3.0] CAR-T Plausibility 

# %%
df = pd.read_sql(f"""
SELECT 
    COUNT(*) as n_rates,
    COUNT(DISTINCT provider_id) as n_providers
FROM tq_dev.internal_dev_csong_cld_v2_3_0.prod_combined_abridged
WHERE billing_code = '018'
AND billing_code_type = 'MS-DRG'
AND canonical_rate_score > 1
LIMIT 10
""", con=trino_conn)
df

# %%
df = pd.read_sql(f"""
SELECT 
    billing_code,
    min(canonical_rate_percent_of_cbsa_avg_medicare) as min_pct,
    max(canonical_rate_percent_of_cbsa_avg_medicare) as max_pct,
    avg(canonical_rate_percent_of_cbsa_avg_medicare) as avg_pct,
    min(canonical_rate) as min_rate,
    max(canonical_rate) as max_rate,
    avg(canonical_rate) as avg_rate
FROM tq_dev.internal_dev_csong_cld_v2_3_0.prod_combined_abridged
WHERE billing_code = '018'
AND billing_code_type = 'MS-DRG'
AND canonical_rate_score > 1
GROUP BY 1
""", con=trino_conn)
print(df.T.to_markdown(floatfmt=".3f"))


# %%
# [CLD-v2.3.0] OP outlier lower bound 

# %%
pd.read_sql(f"""
SELECT 
    min(canonical_rate_percent_of_medicare),
    min(canonical_rate_percent_of_cbsa_avg_medicare),
    min(canonical_rate_percent_of_state_avg_medicare)
FROM tq_dev.internal_dev_csong_cld_v2_3_0.prod_combined_abridged
WHERE bill_type = 'Outpatient'
AND canonical_rate_score > 1
AND is_drug_code = False
AND is_lab_code = False
AND billing_code NOT IN (
    SELECT hcpcs
    FROM tq_production.reference_internal.anesthesia_reference_pricing
)
AND provider_type = 'Hospital'
""", con=trino_conn)

# %%
# [CLD-v2.3.0] PG accuracy logic 

# %%
pd.read_sql(f"""
SELECT 
    canonical_rate_score,
    MIN(canonical_rate_percent_of_medicare) as min_pct,
    MAX(canonical_rate_percent_of_medicare) as max_pct,
    MIN(canonical_rate_percent_of_cbsa_avg_medicare) as min_cbsa_pct,
    MAX(canonical_rate_percent_of_cbsa_avg_medicare) as max_cbsa_pct
FROM tq_dev.internal_dev_csong_cld_v2_3_0.prod_combined_abridged
WHERE provider_type = 'Physician Group'
AND is_lab_code = False
AND is_drug_code = False
AND billing_code NOT IN (
    SELECT hcpcs
    FROM tq_production.reference_internal.anesthesia_reference_pricing
)
GROUP BY 1
ORDER BY 1 DESC
""", con=trino_conn)

# %%
# [CLD-v2.3.0] add infusion centers 

# %%
pd.read_sql(f"""
SELECT 
    canonical_rate_score,
    COUNT(*) as n_possible_rates,
    COUNT(DISTINCT provider_id) as n_possible_providers,
    SUM(
        CASE
            WHEN canonical_rate_score > 1 THEN 1
            ELSE 0
        END
    ) as n_rates,
    COUNT(
        DISTINCT 
        CASE
            WHEN canonical_rate_score > 1 THEN provider_id
        END
    ) as n_providers
FROM tq_dev.internal_dev_csong_cld_v2_3_0.prod_combined_abridged
WHERE provider_subtype = 'Infusion Center'
GROUP BY 1
ORDER BY 1 DESC
LIMIT 10
""", con=trino_conn)

# %%
pd.read_sql(f"""
SELECT 
    COUNT(*) as n_possible_rates,
    COUNT(DISTINCT provider_id) as n_possible_providers,
    SUM(
        CASE
            WHEN canonical_rate_score > 1 THEN 1
            ELSE 0
        END
    ) as n_rates,
    COUNT(
        DISTINCT 
        CASE
            WHEN canonical_rate_score > 1 THEN provider_id
        END
    ) as n_providers
FROM tq_dev.internal_dev_csong_cld_v2_3_0.prod_combined_abridged
WHERE provider_subtype = 'Infusion Center'
""", con=trino_conn)

# %%
# MISC

# %%
# Compare rate counts by network class between v2.3.0 and v2.2.2

# %%
df = pd.read_sql(f"""
WITH
new AS (
    SELECT network_class, COUNT(*) as new_count
    FROM tq_dev.internal_dev_csong_cld_v2_3_0.prod_combined_abridged
    WHERE canonical_rate_score > 1
    GROUP BY 1
),
old AS (
    SELECT network_class, COUNT(*) as old_count
    FROM tq_dev.internal_dev_csong_cld_v2_2_2.prod_combined_abridged
    WHERE canonical_rate_score > 1
    GROUP BY 1
)
SELECT 
    network_class,
    new_count,
    old_count
FROM new
LEFT JOIN old USING (network_class)
""", con=trino_conn)
print(df.to_markdown(index=False, floatfmt=".0f"))


# %%
# check that we are pulling hospital rates for BCBS TN
# (fixed plan bridge naming)

# %%
pd.read_sql(f"""
select COUNT(*), 'new'
FROM tq_dev.internal_dev_csong_cld_v2_3_0.prod_combined_abridged
WHERE provider_id in ('3442', '3447', '3448', '3445')
AND payer_id = '398'
AND canonical_rate_source = 'hospital'
union
select COUNT(*), 'old'
FROM tq_dev.internal_dev_csong_cld_v2_2_2.prod_combined_abridged
WHERE provider_id in ('3442', '3447', '3448', '3445')
AND payer_id = '398'
AND canonical_rate_source = 'hospital'
""", con=trino_conn)

# %%
# provider_type_new
pd.read_sql(f"""
SELECT DISTINCT provider_type, provider_type_new
FROM tq_dev.internal_dev_csong_cld_v2_3_0.prod_clickhouse_abridged
WHERE canonical_rate_score > 1
""", con=trino_conn)

# %%
pd.read_sql(f"""
SELECT network_class, COUNT(*), 'new'
FROM tq_dev.internal_dev_csong_cld_v2_3_0.prod_clickhouse_abridged
GROUP BY 1
UNION ALL
SELECT network_class, COUNT(*), 'old'
FROM tq_dev.internal_dev_csong_cld_v2_2_2.prod_clickhouse_abridged
GROUP BY 1
""", con=trino_conn)

Distributions

# %%
from IPython import get_ipython
from jinja2 import Template

if get_ipython() is not None:
    get_ipython().run_line_magic("load_ext", "autoreload")
    get_ipython().run_line_magic("autoreload", "2")
from dotenv import load_dotenv
import pandas as pd
from tqds.sql import db, sql

import pandas as pd
import numpy as np

import seaborn as sns
import matplotlib.pyplot as plt
sns.set_theme(style='whitegrid', rc={'figure.figsize': (14, 12)})

pd.options.display.max_columns = None
pd.options.display.max_rows = 100
pd.options.display.max_colwidth = None
pd.set_option('display.float_format', lambda x: '%.3f' % x)
trino_conn = db.TrinoClient().conn
redshift_conn = db.RedshiftClient().conn
query = sql.Query()


# %% [markdown]
"""
# Compare rate counts by bill_type, provider_type, canonical_rate_score between v2.3.0 and v2.2.2
"""

# %%
df = pd.read_sql(f"""
WITH
new AS (
    SELECT 
        bill_type,
        provider_type,
        COUNT(*) as n
    FROM tq_dev.internal_dev_csong_cld_v2_3_0.prod_combined_abridged
    WHERE network_type IN ('PPO', 'HMO')
    GROUP BY 1, 2
),
old AS (
    SELECT 
        bill_type,
        provider_type,
        COUNT(*) as n
    FROM tq_dev.internal_dev_csong_cld_v2_2_2.prod_combined_abridged
    WHERE network_type IN ('PPO', 'HMO')
    GROUP BY 1, 2
)
SELECT
    new.bill_type,
    new.provider_type,
    new.n as new_n,
    old.n as old_n
FROM new
JOIN old ON 
new.provider_type = old.provider_type
AND new.bill_type = old.bill_type
ORDER BY 
    new.bill_type,
    new.provider_type
""", con=trino_conn)
df['diff'] = df['new_n'] - df['old_n']
df['diff_pct'] = df['diff'] / df['old_n'] * 100
print(df.to_markdown(index=False, floatfmt=".0f"))


# %% [markdown]
"""
# Compare rate counts by bill_type, provider_type, canonical_rate_score between v2.3.0 and v2.2.2
"""

# %%
df = pd.read_sql(f"""
WITH
new AS (
    SELECT 
        bill_type,
        provider_type,
        canonical_rate_score,
        COUNT(*) as n
    FROM tq_dev.internal_dev_csong_cld_v2_3_0.prod_combined_abridged
    WHERE network_type IN ('PPO', 'HMO')
    GROUP BY 1, 2, 3
),
old AS (
    SELECT 
        bill_type,
        provider_type,
        canonical_rate_score,
        COUNT(*) as n
    FROM tq_dev.internal_dev_csong_cld_v2_2_2.prod_combined_abridged
    WHERE network_type IN ('PPO', 'HMO')
    GROUP BY 1, 2, 3
)
SELECT
    new.bill_type,
    new.provider_type,
    new.canonical_rate_score,
    new.n as new_n,
    old.n as old_n
FROM new
JOIN old ON 
new.provider_type = old.provider_type
AND new.canonical_rate_score = old.canonical_rate_score
AND new.bill_type = old.bill_type
ORDER BY 
    new.bill_type,
    new.provider_type, 
    new.canonical_rate_score
""", con=trino_conn)
df['diff'] = df['new_n'] - df['old_n']
df['diff_pct'] = df['diff'] / df['old_n'] * 100
print(df.to_markdown(index=False, floatfmt=".0f"))

# %%
df

# %% [markdown]
"""
# ASC drop analysis
"""

# %%
df_asc = pd.read_sql(f"""
SELECT 
    canonical_rate_score,
    SUM(
        CASE
            WHEN canonical_rate_percent_of_medicare < 0.5 THEN 1
            ELSE 0
        END
    ) as n_below_0_5,
    COUNT(*) as n
FROM tq_dev.internal_dev_csong_cld_v2_2_2.prod_combined_abridged
WHERE network_type IN ('PPO', 'HMO')
AND provider_type = 'ASC'
GROUP BY 1
""", con=trino_conn)
df_asc

# %%
dfasc2 = pd.read_sql(f"""
WITH
new AS (
    SELECT 
        canonical_rate_type,
        canonical_rate_score,
        COUNT(*) as n
    FROM tq_dev.internal_dev_csong_cld_v2_3_0.prod_combined_abridged
    WHERE network_type IN ('PPO', 'HMO')
    GROUP BY 1, 2
),
old AS (
    SELECT 
        canonical_rate_type,
        canonical_rate_score,
        COUNT(*) as n
    FROM tq_dev.internal_dev_csong_cld_v2_2_2.prod_combined_abridged
    WHERE network_type IN ('PPO', 'HMO')
    GROUP BY 1, 2
)
SELECT
    new.canonical_rate_type,
    new.canonical_rate_score,
    new.n as new_n,
    old.n as old_n
FROM new
JOIN old ON 
new.canonical_rate_score = old.canonical_rate_score
AND new.canonical_rate_type = old.canonical_rate_type
ORDER BY 
    new.canonical_rate_type,
    new.canonical_rate_score
""", con=trino_conn)
dfasc2['diff'] = dfasc2['new_n'] - dfasc2['old_n']
dfasc2['diff_pct'] = dfasc2['diff'] / dfasc2['old_n'] * 100
print(dfasc2.to_markdown(index=False, floatfmt=".0f"))

# %%
dfasc2.sort_values(by='diff_pct', ascending=True)


# %% [markdown]
"""
# NULLs
"""

# %%
columns = pd.read_sql(f"""
DESCRIBE tq_dev.internal_dev_csong_cld_v2_2_2.prod_combined_abridged
""", con=trino_conn)['Column'].tolist()

# %%
q = """
SELECT
    {% for col in columns %}
    1.000 * (COUNT_IF({{col}} IS NULL) / COUNT(*)) AS {{col}}{% if not loop.last %},{% endif %}
    {% endfor %}
FROM tq_dev.internal_dev_csong_cld_v2_3_0.prod_combined_abridged
WHERE canonical_rate_score > 1
"""
q_template = Template(q)
q_rendered = q_template.render(columns=columns)
nulls_v2_3_0 = pd.read_sql(q_rendered, con=trino_conn)
print(nulls_v2_3_0.T.sort_values(0).to_markdown(floatfmt=".3f"))

# %%