Time-series forecasting (Exponential Smoothing) with Darts Integration #1851

martinb-ai · 2022-05-21T03:06:44Z

Description

Newly updated exponential smoothing with probabilistic forecasting utilizing Darts library
(Found inside stocks/pred/expo)

automated historical backtesting and retraining
modifiable sliding window and forecasting look-ahead
mean average precision error (MAPE) over backtest
supports N number of forecasting days; best results are 3-7 days in the future.

Relevant motivation and context.
Standardization, flexibility and readability of a common library.
Improved performance and features.
List any dependencies that are required for this change.
torch, darts

How has this been tested?

Please describe the tests that you ran to verify your changes.

stocks
load aapl
pred
expo -w 0.9 -n 6 -t M -p 5 -d T -f 4 -s M

Please also list any relevant details for your test configuration.

Checklist:

Update our Hugo documentation following these guidelines.
Update our tests following these guidelines.
Make sure you are following our CONTRIBUTING guidelines.
If a feature was added make sure to add it to the corresponding scripts file.

Others

I have performed a self-review of my own code.
I have commented my code, particularly in hard-to-understand areas.
My code passes all the checks pylint, flake8, black, ... To speed up development you should run pre-commit install.
New and existing unit tests pass locally with my changes. You can test this locally using pytest tests/....
--> No, breaks on covid scripts.

jmaslek · 2022-05-21T03:20:44Z

Gonna wanna also update requirements and requirements-full, which can be done as

poetry export -f requirements.txt --without-hashes  --output requirements.txt

poetry export \
       -f requirements.txt \
       --dev \
       -E prediction \
       --without-hashes \
       --output requirements-full.txt

jmaslek · 2022-05-21T03:37:18Z

Only other thing missing is the Hugo website stuff. If you look in the website folder, you can find common/prediction_techniques then follow the same pattern for editing content/ as well as data/main.yml

jmaslek · 2022-05-21T03:41:47Z

So on my not M1 Mac, I seem to be able to install everything except pystan. Ill spare the errors, but they are the usual clang stuff:

    error: Command "clang -Wno-unused-result -Wsign-compare -Wunreachable-code -DNDEBUG -g -fwrapv -O2 -Wall -fPIC -O2 -isystem /opt/anaconda3/envs/openbb/include -fPIC -O2 -isystem /opt/anaconda3/envs/openbb/include -I/opt/anaconda3/envs/openbb/include/python3.10 -c pystan/_misc.cpp -o build/temp.macosx-10.9-x86_64-3.10/pystan/_misc.o -Os -ftemplate-depth-256 -Wno-unused-function -Wno-uninitialized -std=c++1y" failed with exit status 1
    ----------------------------------------
ERROR: Command errored out with exit status 1: /opt/anaconda3/envs/openbb/bin/python -u -c 'import io, os, sys, setuptools, tokenize; sys.argv[0] = '"'"'/private/var/folders/44/whtr9jk97n9638w2kt7ys1pr0000gn/T/pip-install-3ilzvueq/pystan_e1d671dcfc12414b8ea473f679cc8273/setup.py'"'"'; __file__='"'"'/private/var/folders/44/whtr9jk97n9638w2kt7ys1pr0000gn/T/pip-install-3ilzvueq/pystan_e1d671dcfc12414b8ea473f679cc8273/setup.py'"'"';f = getattr(tokenize, '"'"'open'"'"', open)(__file__) if os.path.exists(__file__) else io.StringIO('"'"'from setuptools import setup; setup()'"'"');code = f.read().replace('"'"'\r\n'"'"', '"'"'\n'"'"');f.close();exec(compile(code, __file__, '"'"'exec'"'"'))' install --record /private/var/folders/44/whtr9jk97n9638w2kt7ys1pr0000gn/T/pip-record-vzyjam8o/install-record.txt --single-version-externally-managed --compile --install-headers /opt/anaconda3/envs/openbb/include/python3.10/pystan Check the logs for full command output.

Can any other non-m1 folks confirm this (cough @DidierRLopes )

DidierRLopes · 2022-05-21T11:34:16Z

This is what I'm getting 😞

martinb-ai · 2022-05-21T12:30:38Z

This is what I'm getting 😞

@DidierRLopes

MacOS Issues with LightGBM

At the time of writing, there is an issue with libomp 12.0.1 that results in
segmentation fault on Mac OS Big Sur.
Here's the procedure to downgrade the libomp library (from the
original Github issue):

Install brew if you don't already have it.
Install wget if you don't already have it : brew install wget.
Run the commands below:

wget https://raw.githubusercontent.com/Homebrew/homebrew-core/fb8323f2b170bd4ae97e1bac9bf3e2983af3fdb0/Formula/libomp.rb
brew unlink libomp
brew install libomp.rb

If this works for you, I will have to update the install README for our prediction menu with regards to apple products.
I had to do this also, but figured it might only be an M1 issue...

martinb-ai · 2022-05-21T12:33:54Z

So on my not M1 Mac, I seem to be able to install everything except pystan. Ill spare the errors, but they are the usual clang stuff:

    error: Command "clang -Wno-unused-result -Wsign-compare -Wunreachable-code -DNDEBUG -g -fwrapv -O2 -Wall -fPIC -O2 -isystem /opt/anaconda3/envs/openbb/include -fPIC -O2 -isystem /opt/anaconda3/envs/openbb/include -I/opt/anaconda3/envs/openbb/include/python3.10 -c pystan/_misc.cpp -o build/temp.macosx-10.9-x86_64-3.10/pystan/_misc.o -Os -ftemplate-depth-256 -Wno-unused-function -Wno-uninitialized -std=c++1y" failed with exit status 1
    ----------------------------------------
ERROR: Command errored out with exit status 1: /opt/anaconda3/envs/openbb/bin/python -u -c 'import io, os, sys, setuptools, tokenize; sys.argv[0] = '"'"'/private/var/folders/44/whtr9jk97n9638w2kt7ys1pr0000gn/T/pip-install-3ilzvueq/pystan_e1d671dcfc12414b8ea473f679cc8273/setup.py'"'"'; __file__='"'"'/private/var/folders/44/whtr9jk97n9638w2kt7ys1pr0000gn/T/pip-install-3ilzvueq/pystan_e1d671dcfc12414b8ea473f679cc8273/setup.py'"'"';f = getattr(tokenize, '"'"'open'"'"', open)(__file__) if os.path.exists(__file__) else io.StringIO('"'"'from setuptools import setup; setup()'"'"');code = f.read().replace('"'"'\r\n'"'"', '"'"'\n'"'"');f.close();exec(compile(code, __file__, '"'"'exec'"'"'))' install --record /private/var/folders/44/whtr9jk97n9638w2kt7ys1pr0000gn/T/pip-record-vzyjam8o/install-record.txt --single-version-externally-managed --compile --install-headers /opt/anaconda3/envs/openbb/include/python3.10/pystan Check the logs for full command output.

Can any other non-m1 folks confirm this (cough @DidierRLopes )

I think this might be an issue with using anaconda3 as opposed to miniconda. (Still confirming though)

hrzn · 2022-05-21T14:39:08Z

The clang issue is due to pystan, which is a dependency of prophet, which provides the Prophet model in Darts by default. Prophet and a couple of other dependencies of Darts can be notoriously painful to install. For these reasons, we also provide versions of Darts that do not include all models. In your case, it could be easier to depend on "u8darts[torch]" (PyPi) or u8darts-torch (conda) than the full darts (which contains all models). "u8darts[torch]" contains all models of Darts except Prophet and pmdarima-based models, which you are unlikely to need. You can see a list of available flavours here

hrzn · 2022-05-21T14:48:19Z

openbb_terminal/common/prediction_techniques/expo_model.py

+        time_col="date",
+        value_cols=["AdjClose"],
+        freq="B",
+        fill_missing_dates=True,


It is fine to leave this here, but it might incur a small performance penalty. If you happen to already have the guarantee beforehand that your DataFrame does not have missing dates (even if it has missing values), you can consider setting this to False.

hrzn · 2022-05-21T14:50:55Z

openbb_terminal/common/prediction_techniques/expo_model.py

+    )
+
+    # Show forecast over validation # and then +n_predict afterwards sampled 10 times per point
+    probabilistic_forecast = model_es.predict(n_predict, num_samples=10)


10 samples is somewhat on the low side - something like 500 would seem more reasonable to estimate the distribution. Generating these samples is fast in Numpy so it shouldn't incur any noticeable penalty, and it'll improve precision (especially as later you need the 10th and 90th percentiles).

Great, will update. Thanks for the insight.

Thanks for the feedback @hrzn ❤️

We are going to revamp our entire prediction menu with Darts and create a release around it 🚀 🚀

jmaslek · 2022-05-21T19:23:01Z

So after installing with conda install -c conda-forge -c pytorch u8darts-torch, I broke it :(

2022 May 21, 15:21 (🦋) /stocks/ $ pred
[2022-05-21 15:21:29,852] WARNING | darts.models | Support for AutoARIMA, BATS and TBATS is not available.To enable it, install u8darts[pmdarima] or u8darts[all].
[2022-05-21 15:21:29,853] WARNING | darts.models | Support for Facebook Prophet is not available. To enable it, install "darts", "u8darts[prophet]" or "u8darts[all]" (with pip);or "u8darts-all" (with conda).
libc++abi: terminating with uncaught exception of type std::length_error: vector
zsh: abort      openbb

jmaslek · 2022-05-21T20:06:48Z

It works!

Had to clean up my env from goofing around yesterday with the full install, but conda install -c conda-forge -c pytorch u8darts-torch installed on an intel Mac.

I still get these messages on import, but I assume nothing we can do

[2022-05-21 15:21:29,852] WARNING | darts.models | Support for AutoARIMA, BATS and TBATS is not available.To enable it, install u8darts[pmdarima] or u8darts[all].
[2022-05-21 15:21:29,853] WARNING | darts.models | Support for Facebook Prophet is not available. To enable it, install "darts", "u8darts[prophet]" or "u8darts[all]" (with pip);or "u8darts-all" (with conda).

martinb-ai · 2022-05-21T20:40:02Z

I still get these messages on import, but I assume nothing we can do

[2022-05-21 15:21:29,852] WARNING | darts.models | Support for AutoARIMA, BATS and TBATS is not available.To enable it, install u8darts[pmdarima] or u8darts[all].
[2022-05-21 15:21:29,853] WARNING | darts.models | Support for Facebook Prophet is not available. To enable it, install "darts", "u8darts[prophet]" or "u8darts[all]" (with pip);or "u8darts-all" (with conda).

This is being worked on. unit8co/darts#927

DidierRLopes · 2022-05-21T23:39:50Z

I still get this:

My poetry seems to be f*cked somewhere, but unrelated. But with pip install -r requirements-full.txt all is good!

jmaslek · 2022-05-21T23:51:25Z

My poetry seems to be f*cked somewhere, but unrelated. But with pip install -r requirements-full.txt all is good!

Yeah in order to get it I had to nuke the env and recreate. I don't think we should add this to poetry.

martinb-ai · 2022-05-22T00:43:11Z

Yeah in order to get it I had to nuke the env and recreate. I don't think we should add this to poetry.

Agreed. I am trying to find the proper steps to have it installed on all system architectures to then create a simple setup script. Poetry does not play nicely with all library versions.

For now I am updating the install README on setup that works for M1 and Intel Macs. (5f5c618)

hrzn · 2022-05-22T06:31:50Z

I still get these messages on import, but I assume nothing we can do

[2022-05-21 15:21:29,852] WARNING | darts.models | Support for AutoARIMA, BATS and TBATS is not available.To enable it, install u8darts[pmdarima] or u8darts[all].
[2022-05-21 15:21:29,853] WARNING | darts.models | Support for Facebook Prophet is not available. To enable it, install "darts", "u8darts[prophet]" or "u8darts[all]" (with pip);or "u8darts-all" (with conda).

This is being worked on. unit8co/darts#927

I'll put this one higher on our backlog so we address it rather sooner than later.

martinb-ai · 2022-05-22T14:10:20Z

Can confirm this branch runs smoothly on my linux seup.

conda env create -n gst --file build/conda/conda-3-8-env.yaml
conda activate gst
poetry install -E prediction
conda install -c conda-forge -c pytorch u8darts-torch

jmaslek · 2022-05-22T21:23:10Z

Looks like a side effect of loading in these libraries. When loading in stocks

Loading Daily AAPL stock with starting period 2019-05-20 for analysis.
/Users/james/PycharmProjects/OpenBBTerminal/openbb_terminal/stocks/stocks_helper.py:1128: FutureWarning: DatetimeIndex.union_many is deprecated and will be removed in a future version. Use obj.union instead.
  sch = calendar.schedule(
/Users/james/PycharmProjects/OpenBBTerminal/openbb_terminal/stocks/stocks_helper.py:1128: FutureWarning: DatetimeIndex.union_many is deprecated and will be removed in a future version. Use obj.union instead.
  sch = calendar.schedule(
/Users/james/PycharmProjects/OpenBBTerminal/openbb_terminal/stocks/stocks_helper.py:1128: FutureWarning: DatetimeIndex.union_many is deprecated and will be removed in a future version. Use obj.union instead.
  sch = calendar.schedule(

Can you confirm or is this just me. Its just a warning so we can suppress.

martinb-ai · 2022-05-22T22:14:40Z

Looks like a side effect of loading in these libraries. When loading in stocks

Loading Daily AAPL stock with starting period 2019-05-20 for analysis.
/Users/james/PycharmProjects/OpenBBTerminal/openbb_terminal/stocks/stocks_helper.py:1128: FutureWarning: DatetimeIndex.union_many is deprecated and will be removed in a future version. Use obj.union instead.
  sch = calendar.schedule(
/Users/james/PycharmProjects/OpenBBTerminal/openbb_terminal/stocks/stocks_helper.py:1128: FutureWarning: DatetimeIndex.union_many is deprecated and will be removed in a future version. Use obj.union instead.
  sch = calendar.schedule(
/Users/james/PycharmProjects/OpenBBTerminal/openbb_terminal/stocks/stocks_helper.py:1128: FutureWarning: DatetimeIndex.union_many is deprecated and will be removed in a future version. Use obj.union instead.
  sch = calendar.schedule(

Can you confirm or is this just me. Its just a warning so we can suppress.

Can confirm, see it on my side too - definitely a lib version issue. This warning is not present on the main branch when using a normal env just from poetry install .

martinb-ai and others added 17 commits May 17, 2022 15:21

Working exponential with bugs

1e6546a

utilizing business datatime + fillers

42f840d

working figures

5edcad7

expo with more functionality

0f0aab4

historical forcasting feat.

09f4ebb

numeric values for forcasted days

93c3c4f

print pretty predictions

b6d7a87

formatting

580cde9

code cleanup

22b0b2a

doc strings fix

ddc88cf

more docs

0ef0ba1

quick fix

3ebbb2a

Merge branch 'OpenBB-finance:main' into timeseries-forcasting

223bcd7

refactoring

3493af6

poetry updates

0fc8e5e

typo

d93c717

test script update

a4f1834

update requirements.txt

bf25431

jmaslek added the feat S Small T-Shirt size Feature label May 21, 2022

martinb-ai and others added 2 commits May 21, 2022 00:11

refactoring and Hugo website

d05625a

Update _index.md

0c2cbdb

DidierRLopes mentioned this pull request May 21, 2022

[FR] New and improved prediction techniques (DARTS) #1631

Closed

hrzn reviewed May 21, 2022

View reviewed changes

martinb-ai added 2 commits May 21, 2022 16:16

minor mod and removal of space

6beb2b6

Update readme for forecasting

5f5c618

removal of old TF version

0b0f4b9

Merge branch 'main' into timeseries-forcasting

50d5f3e

martinb-ai and others added 14 commits May 23, 2022 15:47

reqs and poetry

adfaedc

linting

5e415b3

merge main.yml from main

4de697c

supress warnings due to lib version change

3f491c3

reverting main.yml

8bcf6b2

retrying expo on main.yml

d354f3f

Merge branch 'main' into timeseries-forcasting

caa867f

fixing fundamentalanalysis version

5ebdcd1

update poetry fundamentalanalysis

d9b5bd7

type casting arg parse

2372379

seasonal period default update

5340c45

Merge branch 'main' into timeseries-forcasting

184edd7

Merge branch 'main' into timeseries-forcasting

aaf46f6

update en.yml

a712011

DidierRLopes approved these changes May 24, 2022

View reviewed changes

DidierRLopes merged commit 5f8eb43 into OpenBB-finance:main May 24, 2022

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Time-series forecasting (Exponential Smoothing) with Darts Integration #1851

Time-series forecasting (Exponential Smoothing) with Darts Integration #1851

martinb-ai commented May 21, 2022 •

edited

Loading

jmaslek commented May 21, 2022

jmaslek commented May 21, 2022

jmaslek commented May 21, 2022

DidierRLopes commented May 21, 2022

martinb-ai commented May 21, 2022 •

edited

Loading

martinb-ai commented May 21, 2022

hrzn commented May 21, 2022 •

edited

Loading

hrzn May 21, 2022

hrzn May 21, 2022

martinb-ai May 21, 2022

DidierRLopes May 21, 2022

hrzn May 21, 2022

jmaslek commented May 21, 2022

jmaslek commented May 21, 2022

martinb-ai commented May 21, 2022

DidierRLopes commented May 21, 2022

jmaslek commented May 21, 2022

martinb-ai commented May 22, 2022

hrzn commented May 22, 2022

martinb-ai commented May 22, 2022

jmaslek commented May 22, 2022

martinb-ai commented May 22, 2022

Time-series forecasting (Exponential Smoothing) with Darts Integration #1851

Time-series forecasting (Exponential Smoothing) with Darts Integration #1851

Conversation

martinb-ai commented May 21, 2022 • edited Loading

Description

How has this been tested?

Checklist:

Others

jmaslek commented May 21, 2022

jmaslek commented May 21, 2022

jmaslek commented May 21, 2022

DidierRLopes commented May 21, 2022

martinb-ai commented May 21, 2022 • edited Loading

MacOS Issues with LightGBM

martinb-ai commented May 21, 2022

hrzn commented May 21, 2022 • edited Loading

hrzn May 21, 2022

Choose a reason for hiding this comment

hrzn May 21, 2022

Choose a reason for hiding this comment

martinb-ai May 21, 2022

Choose a reason for hiding this comment

DidierRLopes May 21, 2022

Choose a reason for hiding this comment

hrzn May 21, 2022

Choose a reason for hiding this comment

jmaslek commented May 21, 2022

jmaslek commented May 21, 2022

martinb-ai commented May 21, 2022

DidierRLopes commented May 21, 2022

jmaslek commented May 21, 2022

martinb-ai commented May 22, 2022

hrzn commented May 22, 2022

martinb-ai commented May 22, 2022

jmaslek commented May 22, 2022

martinb-ai commented May 22, 2022

martinb-ai commented May 21, 2022 •

edited

Loading

martinb-ai commented May 21, 2022 •

edited

Loading

hrzn commented May 21, 2022 •

edited

Loading