forecose

PyPI Python versions Tests

A time-series forecasting extension for pydexcom using Google's TimesFM. Readings from the previous 24 hours are captured from the Dexcom Share API service and fed into the model to create a forecast of blood glucose values for the next hour.

All modelling and forecasting is performed locally on your device. The only external connections made are with:

  • Dexcom Share API: fetching CGM readings following the pydexcom approach.
  • HuggingFace: one-time download of the forecasting model weights on the first run.

Quick Start

  1. Ensure that you have also installed the pydexcom package and enabled the Share service within your Dexcom G7 / G6 / G5 / G4 mobile app.

pip install pydexcom forecose

  1. Initialise pydexcom with your Dexcom credentials (below shows the simplist route, refer to pydexcom for further instruction).
>>> from pydexcom import Dexcom
>>> dexcom = Dexcom(username="username", password="password")
  1. Generate a prediction. By default, the DexcomForecast class predicts the upcoming hour (the next 12 readings) to prevent overextending the forecast. Set custom prediction lengths by adjusting the horizon argument.
>>> from forecose import DexcomForecast
>>> predictions = DexcomForecast(horizon=12).get_forecast(dexcom)
>>> print(predictions)
                          timestamp  predicted_glucose    q10    q25    q50    q75    q90
0  2026-06-30 11:56:43.332500+01:00              156.0  156.0  147.0  155.0  159.0  162.0
1  2026-06-30 12:01:43.332000+01:00              159.0  159.0  142.0  156.0  165.0  169.0
2  2026-06-30 12:06:43.331500+01:00              160.0  159.0  137.0  156.0  169.0  176.0
3  2026-06-30 12:11:43.331000+01:00              160.0  161.0  132.0  155.0  172.0  180.0
4  2026-06-30 12:16:43.330500+01:00              162.0  161.0  128.0  156.0  176.0  185.0
5  2026-06-30 12:21:43.330000+01:00              162.0  162.0  125.0  155.0  177.0  188.0
6  2026-06-30 12:26:43.329500+01:00              163.0  164.0  120.0  155.0  180.0  192.0
7  2026-06-30 12:31:43.329000+01:00              162.0  162.0  116.0  153.0  181.0  194.0
8  2026-06-30 12:36:43.328500+01:00              162.0  162.0  113.0  152.0  183.0  197.0
9  2026-06-30 12:41:43.328000+01:00              161.0  161.0  109.0  151.0  183.0  197.0
10 2026-06-30 12:46:43.327500+01:00              162.0  162.0  107.0  151.0  186.0  201.0
11 2026-06-30 12:51:43.327000+01:00              161.0  160.0  104.0  149.0  186.0  201.0

>>> print(predictions.mmol_l)
                          timestamp  predicted_glucose  q10  q25  q50   q75   q90
0  2026-06-30 11:56:43.332500+01:00                8.7  8.7  8.2  8.6   8.8   9.0
1  2026-06-30 12:01:43.332000+01:00                8.8  8.8  7.9  8.7   9.2   9.4
2  2026-06-30 12:06:43.331500+01:00                8.9  8.8  7.6  8.7   9.4   9.8
3  2026-06-30 12:11:43.331000+01:00                8.9  8.9  7.3  8.6   9.5  10.0
4  2026-06-30 12:16:43.330500+01:00                9.0  8.9  7.1  8.7   9.8  10.3
5  2026-06-30 12:21:43.330000+01:00                9.0  9.0  6.9  8.6   9.8  10.4
6  2026-06-30 12:26:43.329500+01:00                9.0  9.1  6.7  8.6  10.0  10.7
7  2026-06-30 12:31:43.329000+01:00                9.0  9.0  6.4  8.5  10.0  10.8
8  2026-06-30 12:36:43.328500+01:00                9.0  9.0  6.3  8.4  10.2  10.9
9  2026-06-30 12:41:43.328000+01:00                8.9  8.9  6.0  8.4  10.2  10.9
10 2026-06-30 12:46:43.327500+01:00                9.0  9.0  5.9  8.4  10.3  11.2
11 2026-06-30 12:51:43.327000+01:00                8.9  8.9  5.8  8.3  10.3  11.2

What do these predictions mean?

  • predicted-glucose: The average trajectory of your blood sugar forecast (smoothed, centred baseline of the confidence bands).
  • q10 to q90: The range of confidence bands provide a realistic upper and lower estimate boundaries, showing the full probability distribution of predicted glucose values.

Event Modelling

To account for key events (e.g., insulin administration or carbohydrate (carbs) intake) that act on blood glucose values without distorting the underlying TimesFM probability distribution, you can apply a deterministic overlay to your baseline forecast.

Drawing on mathematical frameworks utilised in closed-loop Artifical Pancreas systems and the Hovorka/Bergman meal submodels, event impacts are computed as a second-order linear delay process. Here, event unit rates (e.g., the absorption of insulin or carbs) are translated into a physiological curve that begins slowly, reaches a peak, and then gradually decays over time.

By default, forecose updates the forecast predictions using standard clinical baselines (a 55-minute peak for insulin, and a 40-minute peak for carbs):

>>> carb_predictions = predictions.add_event(type="carbs", units=30, minutes_ago=0)
>>> print(carb_predictions.mmol_l)
                          timestamp  predicted_glucose   q10  q25   q50   q75   q90
0  2026-06-30 11:56:43.332500+01:00                8.7   8.7  8.2   8.7   8.9   9.0
1  2026-06-30 12:01:43.332000+01:00                9.0   9.0  8.0   8.8   9.3   9.5
2  2026-06-30 12:06:43.331500+01:00                9.3   9.2  8.0   9.0   9.8  10.2
3  2026-06-30 12:11:43.331000+01:00                9.5   9.5  7.9   9.2  10.2  10.6
4  2026-06-30 12:16:43.330500+01:00                9.9   9.8  8.0   9.5  10.7  11.2
5  2026-06-30 12:21:43.330000+01:00               10.2  10.2  8.1   9.8  11.0  11.6
6  2026-06-30 12:26:43.329500+01:00               10.5  10.5  8.1  10.0  11.4  12.1
7  2026-06-30 12:31:43.329000+01:00               10.8  10.8  8.2  10.3  11.8  12.5
8  2026-06-30 12:36:43.328500+01:00               11.0  11.0  8.3  10.5  12.2  13.0
9  2026-06-30 12:41:43.328000+01:00               11.3  11.3  8.4  10.8  12.5  13.3
10 2026-06-30 12:46:43.327500+01:00               11.7  11.7  8.6  11.0  13.0  13.8
11 2026-06-30 12:51:43.327000+01:00               11.9  11.8  8.7  11.2  13.3  14.1

>>> insulin_predictions = predictions.add_event(type="insulin", units=5, minutes_ago=0)
>>> print(insulin_predictions.mmol_l)
                          timestamp  predicted_glucose  q10  q25  q50  q75  q90
0  2026-06-30 11:56:43.332500+01:00                8.6  8.6  8.1  8.5  8.8  8.9
1  2026-06-30 12:01:43.332000+01:00                8.7  8.7  7.7  8.5  9.0  9.2
2  2026-06-30 12:06:43.331500+01:00                8.5  8.5  7.3  8.3  9.0  9.4
3  2026-06-30 12:11:43.331000+01:00                8.3  8.4  6.8  8.0  9.0  9.4
4  2026-06-30 12:16:43.330500+01:00                8.2  8.1  6.3  7.8  8.9  9.4
5  2026-06-30 12:21:43.330000+01:00                7.8  7.8  5.8  7.4  8.7  9.3
6  2026-06-30 12:26:43.329500+01:00                7.5  7.6  5.2  7.1  8.5  9.2
7  2026-06-30 12:31:43.329000+01:00                7.2  7.2  4.6  6.7  8.2  8.9
8  2026-06-30 12:36:43.328500+01:00                6.8  6.8  4.1  6.2  7.9  8.7
9  2026-06-30 12:41:43.328000+01:00                6.4  6.4  3.5  5.8  7.6  8.4
10 2026-06-30 12:46:43.327500+01:00                6.0  6.0  3.0  5.4  7.4  8.2
11 2026-06-30 12:51:43.327000+01:00                5.6  5.6  2.4  4.9  7.0  7.8

Sensitivity to insulin (ISF) and carbohydrates (CSF) is set at 40 mg/dL per 1U and 4 mg/dL per gram, respectively. These values are placeholders meant to represent reasonable values and should be adjusted through the add_event method using the tau and sensitivity parameters. In future, I hope to develop a method for calculating estimate values from historic data.

>>> insensitive_insulin_predictions = predictions.add_event(type="insulin", units=5, minutes_ago=0, sensitivity=20.0)
>>> print(insensitive_insulin_predictions.mmol_l)
                          timestamp  predicted_glucose  q10  q25  q50  q75  q90
0  2026-06-30 11:56:43.332500+01:00                8.7  8.7  8.2  8.6  8.8  9.0
1  2026-06-30 12:01:43.332000+01:00                8.8  8.8  7.8  8.6  9.1  9.3
2  2026-06-30 12:06:43.331500+01:00                8.7  8.7  7.4  8.5  9.2  9.6
3  2026-06-30 12:11:43.331000+01:00                8.6  8.7  7.0  8.3  9.3  9.7
4  2026-06-30 12:16:43.330500+01:00                8.5  8.5  6.7  8.2  9.3  9.8
5  2026-06-30 12:21:43.330000+01:00                8.4  8.4  6.4  8.0  9.3  9.9
6  2026-06-30 12:26:43.329500+01:00                8.3  8.4  5.9  7.9  9.3  9.9
7  2026-06-30 12:31:43.329000+01:00                8.0  8.0  5.5  7.5  9.1  9.8
8  2026-06-30 12:36:43.328500+01:00                7.9  7.9  5.2  7.3  9.0  9.8
9  2026-06-30 12:41:43.328000+01:00                7.7  7.7  4.8  7.1  8.9  9.7
10 2026-06-30 12:46:43.327500+01:00                7.5  7.5  4.5  6.9  8.9  9.7
11 2026-06-30 12:51:43.327000+01:00                7.3  7.2  4.1  6.6  8.7  9.5

Key dates

PyPI data
Data sourced directly from PyPI's database.
  • Released:
Newer release available (0.5.0)

1 maintainer

PyPI data
Data sourced directly from PyPI's database.
Avatar for aes21 from gravatar.com aes21

Credits

Author: Alexander Sadler

Requires

Python >=3.10

Report project as malware

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

forecose-0.4.3.tar.gz (11.7 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

forecose-0.4.3-py3-none-any.whl (11.8 kB view details)

Uploaded Python 3

File details

Details for the file forecose-0.4.3.tar.gz.

File metadata

  • Download URL: forecose-0.4.3.tar.gz
  • Upload date:
  • Size: 11.7 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/7.0.0 CPython/3.13.14

File hashes

Hashes for forecose-0.4.3.tar.gz
Algorithm Hash digest
SHA256 ef1b1b6c7726a71e906848c8935e50ba1fb7f8817fcbfe0c9253803a1816716a
MD5 aa25ab40dd5117761513d84dd20554df
BLAKE2b-256 5f8c06a94067750f5496f9271003811cb1443d9b1c71fcdc221b6f4c5f3b1127

See more details on using hashes here.

Provenance

The following attestation bundles were made for forecose-0.4.3.tar.gz:

Publisher: publish.yml on aes21/forecose

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file forecose-0.4.3-py3-none-any.whl.

File metadata

  • Download URL: forecose-0.4.3-py3-none-any.whl
  • Upload date:
  • Size: 11.8 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/7.0.0 CPython/3.13.14

File hashes

Hashes for forecose-0.4.3-py3-none-any.whl
Algorithm Hash digest
SHA256 f970a69098f8ba978fdfdd0d5ff082c248d78b248d68de88388ba7f8762d1cf8
MD5 971f643c7d3d24952beffdf5fa464050
BLAKE2b-256 e3d79f776fb7d09283d2c5a49ca9837e64711e2eb318caffcfccbdc186fea4d2

See more details on using hashes here.

Provenance

The following attestation bundles were made for forecose-0.4.3-py3-none-any.whl:

Publisher: publish.yml on aes21/forecose

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Release history Release notifications | RSS feed

0.5.0

2 files

This release

0.4.3 This release

2 files

0.4.2

2 files

0.4.1

2 files

0.4.0

2 files

0.3.0

2 files

0.2.2

2 files

0.2.1

2 files

0.2.0

2 files

0.1.1

2 files

0.1.0

2 files

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page