Open SolosoftNL opened 1 year ago
nateemma
commented
1 year ago Hmm, that doesn't show on my end. That's a Freqtrade variable, so you may not have your packages installed properly. Can you let me know which version of freqtrade that you are using, and also which version of python please (that line would be invalid for python 2)?
Also, I assume you are looking at the binance directory?
Thanks,
Phil
SolosoftNL
commented
1 year ago Hi. Thanks for your fast response. I've created a docker freqtrade container with freqtrade v2022.12 and running python 3.9+ And yes I'm looking at Binance folder. I think your right that I've to check my packages are all installed correctly. The one used in the strategy file. If I have troubles i always check the code at: https://extendsclass.com/python-tester.html But good change this site uses an older version of python. Will check it tonight but the issue is on my side i think 🤔
SolosoftNL
commented
1 year ago update: still get errors in log, i have installed scipy correctly, see below 2023-01-25 20:38:11,162 - freqtrade.configuration.configuration - INFO - Using pairlist from configuration. 2023-01-25 20:38:11,391 - freqtrade.resolvers.iresolver - WARNING - Could not import /freqtrade/user_data/strategies/DWT_LongShort.py due to 'No module named 'scipy''
Requirement already satisfied: scipy in /home/pi/.local/lib/python3.9/site-packages (1.8.1) Requirement already satisfied: numpy<1.25.0,>=1.17.3 in /usr/lib/python3/dist-packages (from scipy) (1.19.5)
what could possibly the problem?
SolosoftNL
commented
1 year ago No one?
tomjrtsmith
commented
1 year ago I am getting a similar error, but on another strategy and not using docker. I've also got various tensorflow issues that aren't to do with the strategies but more to do with numpy and freqtrade there are some unaddressed dependency issues floating around. Suggest opening an issue in the freqtrade repos and seeing if Mathais has any ideas or failing that the scipy repo.
On Sun, Jan 29, 2023 at 2:41 AM SolosoftNL @.***> wrote:
No one?
— Reply to this email directly, view it on GitHub https://github.com/nateemma/strategies/issues/6#issuecomment-1407401852, or unsubscribe https://github.com/notifications/unsubscribe-auth/AXXUNNIUBKM7HSSMPDEKVRLWUUO2DANCNFSM6AAAAAAUGI4SGI . You are receiving this because you are subscribed to this thread.Message ID: @.***>
nateemma
commented
1 year ago I don’t see these issues, but I do have issues with conflicts between freqtrade, tensorflow and PyTorch. Are you using pip or conda to install the packages? I have noticed that the freqtrade environment doesn’t always find conda packages (and I sometimes have to use conda to get Mac-specific packages)
Cheers
Phil
On Wed, Jan 25, 2023 at 12:43 PM SolosoftNL @.***> wrote:
update: still get errors in log, i have installed scipy correctly, see below 2023-01-25 20:38:11,162 - freqtrade.configuration.configuration - INFO - Using pairlist from configuration. 2023-01-25 20:38:11,391 - freqtrade.resolvers.iresolver - WARNING - Could not import /freqtrade/user_data/strategies/DWT_LongShort.py due to 'No module named 'scipy''
Requirement already satisfied: scipy in /home/pi/.local/lib/python3.9/site-packages (1.8.1) Requirement already satisfied: numpy<1.25.0,>=1.17.3 in /usr/lib/python3/dist-packages (from scipy) (1.19.5)
what could possibly the problem?
— Reply to this email directly, view it on GitHub https://github.com/nateemma/strategies/issues/6#issuecomment-1404200399, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABD4X54ETUVUK5DNIBFEOEDWUGF6XANCNFSM6AAAAAAUGI4SGI . You are receiving this because you commented.Message ID: @.***>
tomjrtsmith
commented
1 year ago I think this may be one issue, not sure how to best resolve as I have 5 PCA's running but it looks like all the sklearn references are being affected by a brownout schedule, it's not clear from reading this whether all sklearn file references need to be replaced with scikit-learn but I suspect so
https://pypi.org/project/sklearn/
On Sun, Jan 29, 2023 at 10:17 AM Nateemma @.***> wrote:
I don’t see these issues, but I do have issues with conflicts between freqtrade, tensorflow and PyTorch. Are you using pip or conda to install the packages? I have noticed that the freqtrade environment doesn’t always find conda packages (and I sometimes have to use conda to get Mac-specific packages)
Cheers
Phil
On Wed, Jan 25, 2023 at 12:43 PM SolosoftNL @.***> wrote:
update: still get errors in log, i have installed scipy correctly, see below 2023-01-25 20:38:11,162 - freqtrade.configuration.configuration - INFO - Using pairlist from configuration. 2023-01-25 20:38:11,391 - freqtrade.resolvers.iresolver - WARNING - Could not import /freqtrade/user_data/strategies/DWT_LongShort.py due to 'No module named 'scipy''
Requirement already satisfied: scipy in /home/pi/.local/lib/python3.9/site-packages (1.8.1) Requirement already satisfied: numpy<1.25.0,>=1.17.3 in /usr/lib/python3/dist-packages (from scipy) (1.19.5)
what could possibly the problem?
— Reply to this email directly, view it on GitHub <https://github.com/nateemma/strategies/issues/6#issuecomment-1404200399 , or unsubscribe < https://github.com/notifications/unsubscribe-auth/ABD4X54ETUVUK5DNIBFEOEDWUGF6XANCNFSM6AAAAAAUGI4SGI
. You are receiving this because you commented.Message ID: @.***>
— Reply to this email directly, view it on GitHub https://github.com/nateemma/strategies/issues/6#issuecomment-1407490200, or unsubscribe https://github.com/notifications/unsubscribe-auth/AXXUNNMYGKBPPD5SLZWXUG3WUWEHJANCNFSM6AAAAAAUGI4SGI . You are receiving this because you commented.Message ID: @.***>
nateemma
commented
1 year ago Ah, I'm already using scikit-learn so maybe that's why I don't see these issues...
I will update the code to explicitly import scikit-learn and upload (when I finish my current edit)
Cheers,
Phil
On Sat, Jan 28, 2023 at 3:20 PM tom s @.***> wrote:
I think this may be one issue, not sure how to best resolve as I have 5 PCA's running but it looks like all the sklearn references are being affected by a brownout schedule, it's not clear from reading this whether all sklearn file references need to be replaced with scikit-learn but I suspect so
https://pypi.org/project/sklearn/
On Sun, Jan 29, 2023 at 10:17 AM Nateemma @.***> wrote:
I don’t see these issues, but I do have issues with conflicts between freqtrade, tensorflow and PyTorch. Are you using pip or conda to install the packages? I have noticed that the freqtrade environment doesn’t always find conda packages (and I sometimes have to use conda to get Mac-specific packages)
Cheers
Phil
On Wed, Jan 25, 2023 at 12:43 PM SolosoftNL @.***> wrote:
update: still get errors in log, i have installed scipy correctly, see below 2023-01-25 20:38:11,162 - freqtrade.configuration.configuration - INFO
Using pairlist from configuration. 2023-01-25 20:38:11,391 - freqtrade.resolvers.iresolver - WARNING - Could not import /freqtrade/user_data/strategies/DWT_LongShort.py due to 'No module named 'scipy''
Requirement already satisfied: scipy in /home/pi/.local/lib/python3.9/site-packages (1.8.1) Requirement already satisfied: numpy<1.25.0,>=1.17.3 in /usr/lib/python3/dist-packages (from scipy) (1.19.5)
what could possibly the problem?
— Reply to this email directly, view it on GitHub < https://github.com/nateemma/strategies/issues/6#issuecomment-1404200399 , or unsubscribe <
https://github.com/notifications/unsubscribe-auth/ABD4X54ETUVUK5DNIBFEOEDWUGF6XANCNFSM6AAAAAAUGI4SGI
. You are receiving this because you commented.Message ID: @.***>
— Reply to this email directly, view it on GitHub <https://github.com/nateemma/strategies/issues/6#issuecomment-1407490200 , or unsubscribe < https://github.com/notifications/unsubscribe-auth/AXXUNNMYGKBPPD5SLZWXUG3WUWEHJANCNFSM6AAAAAAUGI4SGI
. You are receiving this because you commented.Message ID: @.***>
— Reply to this email directly, view it on GitHub https://github.com/nateemma/strategies/issues/6#issuecomment-1407509456, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABD4X57JZL4PYUJP6TQIEVLWUWSTVANCNFSM6AAAAAAUGI4SGI . You are receiving this because you commented.Message ID: @.***>
nateemma
commented
1 year ago never mind, the import remains unchanged.
FYI, there is a script user_data/strategies/scripts/install_packages.sh that will install all packages for you. It is perhaps Mac-specific, but it might work for other platforms
Cheers,
Phil
On Sat, Jan 28, 2023 at 3:26 PM Phil Price @.***> wrote:
Ah, I'm already using scikit-learn so maybe that's why I don't see these issues...
I will update the code to explicitly import scikit-learn and upload (when I finish my current edit)
Cheers,
Phil
On Sat, Jan 28, 2023 at 3:20 PM tom s @.***> wrote:
I think this may be one issue, not sure how to best resolve as I have 5 PCA's running but it looks like all the sklearn references are being affected by a brownout schedule, it's not clear from reading this whether all sklearn file references need to be replaced with scikit-learn but I suspect so
https://pypi.org/project/sklearn/
On Sun, Jan 29, 2023 at 10:17 AM Nateemma @.***> wrote:
I don’t see these issues, but I do have issues with conflicts between freqtrade, tensorflow and PyTorch. Are you using pip or conda to install the packages? I have noticed that the freqtrade environment doesn’t always find conda packages (and I sometimes have to use conda to get Mac-specific packages)
Cheers
Phil
On Wed, Jan 25, 2023 at 12:43 PM SolosoftNL @.***> wrote:
update: still get errors in log, i have installed scipy correctly, see below 2023-01-25 20:38:11,162 - freqtrade.configuration.configuration - INFO - Using pairlist from configuration. 2023-01-25 20:38:11,391 - freqtrade.resolvers.iresolver - WARNING - Could not import /freqtrade/user_data/strategies/DWT_LongShort.py due to 'No module named 'scipy''
Requirement already satisfied: scipy in /home/pi/.local/lib/python3.9/site-packages (1.8.1) Requirement already satisfied: numpy<1.25.0,>=1.17.3 in /usr/lib/python3/dist-packages (from scipy) (1.19.5)
what could possibly the problem?
— Reply to this email directly, view it on GitHub < https://github.com/nateemma/strategies/issues/6#issuecomment-1404200399 , or unsubscribe <
https://github.com/notifications/unsubscribe-auth/ABD4X54ETUVUK5DNIBFEOEDWUGF6XANCNFSM6AAAAAAUGI4SGI
. You are receiving this because you commented.Message ID: @.***>
— Reply to this email directly, view it on GitHub < https://github.com/nateemma/strategies/issues/6#issuecomment-1407490200>, or unsubscribe < https://github.com/notifications/unsubscribe-auth/AXXUNNMYGKBPPD5SLZWXUG3WUWEHJANCNFSM6AAAAAAUGI4SGI
. You are receiving this because you commented.Message ID: @.***>
— Reply to this email directly, view it on GitHub https://github.com/nateemma/strategies/issues/6#issuecomment-1407509456, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABD4X57JZL4PYUJP6TQIEVLWUWSTVANCNFSM6AAAAAAUGI4SGI . You are receiving this because you commented.Message ID: @.***>
tomjrtsmith
commented
1 year ago Turned out the Torch installer for Darts wouldn't work for linux so I had to install using the pip3 command from their website and that cleared up all the sklearn errors.
I still get a bunch of tensorflow file lock errors but I'm convinced that is a tensorflow issue so I've asked them to test it with freqtrade.
PCA_dwt is throwing out this sklearn error, but only occasionally and I don't know if it's affecting performance, I've only just gotten it running.
/home/ubuntu/freqtrade/.env/lib/python3.10/site-packages/sklearn/model_selection/_split.py:700: UserWarning: The least populated class in y has only 4 members, which is less than n_splits=5.
On Sun, Jan 29, 2023 at 12:47 PM Nateemma @.***> wrote:
never mind, the import remains unchanged.
FYI, there is a script user_data/strategies/scripts/install_packages.sh that will install all packages for you. It is perhaps Mac-specific, but it might work for other platforms
Cheers,
Phil
On Sat, Jan 28, 2023 at 3:26 PM Phil Price @.***> wrote:
Ah, I'm already using scikit-learn so maybe that's why I don't see these issues...
I will update the code to explicitly import scikit-learn and upload (when I finish my current edit)
Cheers,
Phil
On Sat, Jan 28, 2023 at 3:20 PM tom s @.***> wrote:
I think this may be one issue, not sure how to best resolve as I have 5 PCA's running but it looks like all the sklearn references are being affected by a brownout schedule, it's not clear from reading this whether all sklearn file references need to be replaced with scikit-learn but I suspect so
https://pypi.org/project/sklearn/
On Sun, Jan 29, 2023 at 10:17 AM Nateemma @.***> wrote:
I don’t see these issues, but I do have issues with conflicts between freqtrade, tensorflow and PyTorch. Are you using pip or conda to install the packages? I have noticed that the freqtrade environment doesn’t always find conda packages (and I sometimes have to use conda to get Mac-specific packages)
Cheers
Phil
On Wed, Jan 25, 2023 at 12:43 PM SolosoftNL @.***> wrote:
update: still get errors in log, i have installed scipy correctly, see below 2023-01-25 20:38:11,162 - freqtrade.configuration.configuration - INFO - Using pairlist from configuration. 2023-01-25 20:38:11,391 - freqtrade.resolvers.iresolver - WARNING - Could not import /freqtrade/user_data/strategies/DWT_LongShort.py due to 'No module named 'scipy''
Requirement already satisfied: scipy in /home/pi/.local/lib/python3.9/site-packages (1.8.1) Requirement already satisfied: numpy<1.25.0,>=1.17.3 in /usr/lib/python3/dist-packages (from scipy) (1.19.5)
what could possibly the problem?
— Reply to this email directly, view it on GitHub < https://github.com/nateemma/strategies/issues/6#issuecomment-1404200399 , or unsubscribe <
https://github.com/notifications/unsubscribe-auth/ABD4X54ETUVUK5DNIBFEOEDWUGF6XANCNFSM6AAAAAAUGI4SGI
. You are receiving this because you commented.Message ID: @.***>
— Reply to this email directly, view it on GitHub < https://github.com/nateemma/strategies/issues/6#issuecomment-1407490200 , or unsubscribe <
https://github.com/notifications/unsubscribe-auth/AXXUNNMYGKBPPD5SLZWXUG3WUWEHJANCNFSM6AAAAAAUGI4SGI
. You are receiving this because you commented.Message ID: @.***>
— Reply to this email directly, view it on GitHub < https://github.com/nateemma/strategies/issues/6#issuecomment-1407509456>, or unsubscribe < https://github.com/notifications/unsubscribe-auth/ABD4X57JZL4PYUJP6TQIEVLWUWSTVANCNFSM6AAAAAAUGI4SGI
. You are receiving this because you commented.Message ID: @.***>
— Reply to this email directly, view it on GitHub https://github.com/nateemma/strategies/issues/6#issuecomment-1407513559, or unsubscribe https://github.com/notifications/unsubscribe-auth/AXXUNNKCE6MGASONW26RMUTWUWVZLANCNFSM6AAAAAAUGI4SGI . You are receiving this because you commented.Message ID: @.***>
nateemma
commented
1 year ago On Sat, Jan 28, 2023 at 7:39 PM tom s @.***> wrote:
Turned out the Torch installer for Darts wouldn't work for linux so I had to install using the pip3 command from their website and that cleared up all the sklearn errors.
I still get a bunch of tensorflow file lock errors but I'm convinced that is a tensorflow issue so I've asked them to test it with freqtrade.
I’ve noticed issues with tensor flow and cpu/gpu interactions with multiple threads, especially with hyper opt. Backtesting is ok for me though
PCA_dwt is throwing out this sklearn error, but only occasionally and I don't know if it's affecting performance, I've only just gotten it running.
/home/ubuntu/freqtrade/.env/lib/python3.10/site-packages/sklearn/model_selection/_split.py:700: UserWarning: The least populated class in y has only 4 members, which is less than n_splits=5.
OK, that’s because the tree-based classifiers require at least 5 positive labels (buys/sells). You can likely ‘fix’ this by decreasing the entry and exit thresholds in the json file
Cheers
Phil
On Sun, Jan 29, 2023 at 12:47 PM Nateemma @.***> wrote:
never mind, the import remains unchanged.
FYI, there is a script user_data/strategies/scripts/install_packages.sh that will install all packages for you. It is perhaps Mac-specific, but it might work for other platforms
Cheers,
Phil
On Sat, Jan 28, 2023 at 3:26 PM Phil Price @.***> wrote:
Ah, I'm already using scikit-learn so maybe that's why I don't see these issues...
I will update the code to explicitly import scikit-learn and upload (when I finish my current edit)
Cheers,
Phil
On Sat, Jan 28, 2023 at 3:20 PM tom s @.***> wrote:
I think this may be one issue, not sure how to best resolve as I have 5 PCA's running but it looks like all the sklearn references are being affected by a brownout schedule, it's not clear from reading this whether all sklearn file references need to be replaced with scikit-learn but I suspect so
https://pypi.org/project/sklearn/
On Sun, Jan 29, 2023 at 10:17 AM Nateemma @.***> wrote:
I don’t see these issues, but I do have issues with conflicts between freqtrade, tensorflow and PyTorch. Are you using pip or conda to install the packages? I have noticed that the freqtrade environment doesn’t always find conda packages (and I sometimes have to use conda to get Mac-specific packages)
Cheers
Phil
On Wed, Jan 25, 2023 at 12:43 PM SolosoftNL @.***> wrote:
update: still get errors in log, i have installed scipy correctly, see below 2023-01-25 20:38:11,162 - freqtrade.configuration.configuration - INFO - Using pairlist from configuration. 2023-01-25 20:38:11,391 - freqtrade.resolvers.iresolver - WARNING
Could not import /freqtrade/user_data/strategies/DWT_LongShort.py due to 'No module named 'scipy''
Requirement already satisfied: scipy in /home/pi/.local/lib/python3.9/site-packages (1.8.1) Requirement already satisfied: numpy<1.25.0,>=1.17.3 in /usr/lib/python3/dist-packages (from scipy) (1.19.5)
what could possibly the problem?
— Reply to this email directly, view it on GitHub <
https://github.com/nateemma/strategies/issues/6#issuecomment-1404200399
, or unsubscribe <
https://github.com/notifications/unsubscribe-auth/ABD4X54ETUVUK5DNIBFEOEDWUGF6XANCNFSM6AAAAAAUGI4SGI
. You are receiving this because you commented.Message ID: @.***>
— Reply to this email directly, view it on GitHub <
https://github.com/nateemma/strategies/issues/6#issuecomment-1407490200 ,
or unsubscribe <
https://github.com/notifications/unsubscribe-auth/AXXUNNMYGKBPPD5SLZWXUG3WUWEHJANCNFSM6AAAAAAUGI4SGI
. You are receiving this because you commented.Message ID: @.***>
— Reply to this email directly, view it on GitHub < https://github.com/nateemma/strategies/issues/6#issuecomment-1407509456 , or unsubscribe <
https://github.com/notifications/unsubscribe-auth/ABD4X57JZL4PYUJP6TQIEVLWUWSTVANCNFSM6AAAAAAUGI4SGI
. You are receiving this because you commented.Message ID: @.***>
— Reply to this email directly, view it on GitHub <https://github.com/nateemma/strategies/issues/6#issuecomment-1407513559 , or unsubscribe < https://github.com/notifications/unsubscribe-auth/AXXUNNKCE6MGASONW26RMUTWUWVZLANCNFSM6AAAAAAUGI4SGI
. You are receiving this because you commented.Message ID: @.***>
— Reply to this email directly, view it on GitHub https://github.com/nateemma/strategies/issues/6#issuecomment-1407551784, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABD4X5YIGT57XKNDHEDUMSLWUXRAPANCNFSM6AAAAAAUGI4SGI . You are receiving this because you commented.Message ID: @.***>
nateemma
commented
1 year ago CUDA is a gpu library, specifically for NVidia GPUs. Tensorflow will try and use that if possible, otherwise it should fall back to CPU mode. Would it be possible for you to try running an NNBC strat on a VPS with an Nvidia GPU?
Regarding the nseq strats, not sure what to say - I've been running them in dry-mode on an iMac for the last 3 days without issue. The only difference between them and other strats in the same family is the training criteria. Could you send me the output that they produce please?
Thanks,
Phil
On Sun, Jan 29, 2023 at 12:14 PM tom s @.***> wrote:
I've come across a few more issues, some of them minor and some of them I think potentially a bit harder
I can't get any of the _nesq strategies to run. I'm running oracle VPS's with ubuntu 22.04 and all of the PCA classes run except the _nesq classes. I don't know why but they won't start up, I've tried on several VPS's as I'm going through testing strats and each time they fail on startup, PCA_nesq and Anomaly_nesq.
NNBC strats have terminal memory bleed. I first ran one on an Oracle ARM 4 core with 24g of RAM and it crashed that, started out with 650 odd mb and gradually used all 24g then crashed. I started trialling other strats and then returned to it, same thing happened. Last night I started NNBC_fbb on a 2 core 1gb instance with 2gb of swap file and within hours it had filled the swap and then crashed, this morning I increased the swap to 4gb and it's steadily filling that up, which is frustrating because it makes amazing trades when it runs.
I'm no expert on memory bleed but I watched the Freqai guys fix some issues when they were putting that together and tracking down the source isn't so easy.
I get this error on startup
Could not load dynamic library 'libcuda.so.1'; dlerror: libcuda.so.1: cannot open shared object file: No such file or directory``` which looks to be a GPU driver error but I've looked through the code and none of the NNBC imports specify a GPU. It is throwing out some zero division parameter errors but I don't think this is the source of the memory bleed. On Sun, Jan 29, 2023 at 5:15 PM Nateemma ***@***.***> wrote: > On Sat, Jan 28, 2023 at 7:39 PM tom s ***@***.***> wrote: > > > Turned out the Torch installer for Darts wouldn't work for linux so I had > > to install using the pip3 command from their website and that cleared up > > all the sklearn errors. > > > > I still get a bunch of tensorflow file lock errors but I'm convinced that > > is a tensorflow issue so I've asked them to test it with freqtrade. > > > > I’ve noticed issues with tensor flow and cpu/gpu interactions with multiple > threads, especially with hyper opt. Backtesting is ok for me though > > > > PCA_dwt is throwing out this sklearn error, but only occasionally and I > > don't know if it's affecting performance, I've only just gotten it > running. > > > > > > > /home/ubuntu/freqtrade/.env/lib/python3.10/site-packages/sklearn/model_selection/_split.py:700: > > UserWarning: The least populated class in y has only 4 members, which is > > less than n_splits=5. > > > > OK, that’s because the tree-based classifiers require at least 5 positive > labels (buys/sells). You can likely ‘fix’ this by decreasing the entry and > exit thresholds in the json file > > Cheers > > Phil > > > > On Sun, Jan 29, 2023 at 12:47 PM Nateemma ***@***.***> wrote: > > > > > never mind, the import remains unchanged. > > > > > > FYI, there is a script user_data/strategies/scripts/install_packages.sh > > > that will install all packages for you. It is perhaps Mac-specific, but > > it > > > might work for other platforms > > > > > > Cheers, > > > > > > Phil > > > > > > On Sat, Jan 28, 2023 at 3:26 PM Phil Price ***@***.***> wrote: > > > > > > > Ah, I'm already using scikit-learn so maybe that's why I don't see > > these > > > > issues... > > > > > > > > I will update the code to explicitly import scikit-learn and upload > > (when > > > > I finish my current edit) > > > > > > > > Cheers, > > > > > > > > Phil > > > > > > > > On Sat, Jan 28, 2023 at 3:20 PM tom s ***@***.***> wrote: > > > > > > > >> I think this may be one issue, not sure how to best resolve as I > have > > 5 > > > >> PCA's running but it looks like all the sklearn references are being > > > >> affected by a brownout schedule, it's not clear from reading this > > > whether > > > >> all sklearn file references need to be replaced with scikit-learn > but > > I > > > >> suspect so > > > >> > > > >> https://pypi.org/project/sklearn/ > > > >> > > > >> On Sun, Jan 29, 2023 at 10:17 AM Nateemma ***@***.***> wrote: > > > >> > > > >> > I don’t see these issues, but I do have issues with conflicts > > between > > > >> > freqtrade, tensorflow and PyTorch. > > > >> > Are you using pip or conda to install the packages? I have noticed > > > that > > > >> the > > > >> > freqtrade environment doesn’t always find conda packages (and I > > > >> sometimes > > > >> > have to use conda to get Mac-specific packages) > > > >> > > > > >> > > > > >> > Cheers > > > >> > > > > >> > Phil > > > >> > > > > >> > On Wed, Jan 25, 2023 at 12:43 PM SolosoftNL ***@***.***> > > > >> > wrote: > > > >> > > > > >> > > update: still get errors in log, i have installed scipy > correctly, > > > see > > > >> > > below > > > >> > > 2023-01-25 20:38:11,162 - freqtrade.configuration.configuration > - > > > >> INFO - > > > >> > > Using pairlist from configuration. > > > >> > > 2023-01-25 20:38:11,391 - freqtrade.resolvers.iresolver - > WARNING > > - > > > >> Could > > > >> > > not import /freqtrade/user_data/strategies/DWT_LongShort.py due > to > > > 'No > > > >> > > module named 'scipy'' > > > >> > > > > > >> > > Requirement already satisfied: scipy in > > > >> > > /home/pi/.local/lib/python3.9/site-packages (1.8.1) > > > >> > > Requirement already satisfied: numpy<1.25.0,>=1.17.3 in > > > >> > > /usr/lib/python3/dist-packages (from scipy) (1.19.5) > > > >> > > > > > >> > > what could possibly the problem? > > > >> > > > > > >> > > — > > > >> > > Reply to this email directly, view it on GitHub > > > >> > > < > > > >> > > https://github.com/nateemma/strategies/issues/6#issuecomment-1404200399 > > > >> > >, > > > >> > > or unsubscribe > > > >> > > < > > > >> > > > > >> > > > > > > https://github.com/notifications/unsubscribe-auth/ABD4X54ETUVUK5DNIBFEOEDWUGF6XANCNFSM6AAAAAAUGI4SGI > > > >> > > > > > >> > > . > > > >> > > You are receiving this because you commented.Message ID: > > > >> > > ***@***.***> > > > >> > > > > > >> > > > > >> > — > > > >> > Reply to this email directly, view it on GitHub > > > >> > < > > > >> > > https://github.com/nateemma/strategies/issues/6#issuecomment-1407490200 > > > >, > > > >> > or unsubscribe > > > >> > < > > > >> > > > > > > https://github.com/notifications/unsubscribe-auth/AXXUNNMYGKBPPD5SLZWXUG3WUWEHJANCNFSM6AAAAAAUGI4SGI > > > >> > > > > >> > . > > > >> > You are receiving this because you commented.Message ID: > > > >> > ***@***.***> > > > >> > > > > >> > > > >> — > > > >> Reply to this email directly, view it on GitHub > > > >> < > > > > https://github.com/nateemma/strategies/issues/6#issuecomment-1407509456 > > >, > > > >> or unsubscribe > > > >> < > > > > > > https://github.com/notifications/unsubscribe-auth/ABD4X57JZL4PYUJP6TQIEVLWUWSTVANCNFSM6AAAAAAUGI4SGI > > > > > > > >> . > > > >> You are receiving this because you commented.Message ID: > > > >> ***@***.***> > > > >> > > > > > > > > > > — > > > Reply to this email directly, view it on GitHub > > > < > https://github.com/nateemma/strategies/issues/6#issuecomment-1407513559 > > >, > > > or unsubscribe > > > < > > > https://github.com/notifications/unsubscribe-auth/AXXUNNKCE6MGASONW26RMUTWUWVZLANCNFSM6AAAAAAUGI4SGI > > > > > > . > > > You are receiving this because you commented.Message ID: > > > ***@***.***> > > > > > > > — > > Reply to this email directly, view it on GitHub > > < https://github.com/nateemma/strategies/issues/6#issuecomment-1407551784 > >, > > or unsubscribe > > < > https://github.com/notifications/unsubscribe-auth/ABD4X5YIGT57XKNDHEDUMSLWUXRAPANCNFSM6AAAAAAUGI4SGI > > > > . > > You are receiving this because you commented.Message ID: > > ***@***.***> > > > > — > Reply to this email directly, view it on GitHub > <https://github.com/nateemma/strategies/issues/6#issuecomment-1407557227 >, > or unsubscribe > < https://github.com/notifications/unsubscribe-auth/AXXUNNI5BHSLDSNE34JLNX3WUXVHBANCNFSM6AAAAAAUGI4SGI > > . > You are receiving this because you commented.Message ID: > ***@***.***> > — Reply to this email directly, view it on GitHub <https://github.com/nateemma/strategies/issues/6#issuecomment-1407759530>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABD4X54UD457U64N6AZBNNDWU3FTHANCNFSM6AAAAAAUGI4SGI> . You are receiving this because you commented.Message ID: ***@***.***>
nateemma
commented
1 year ago FYI, I added some memory tracing instrumentation to the NNBC (and NNPredict) strategies, and uploaded to github. Will run overnight on my iMac...
Cheers,
Phil
On Sun, Jan 29, 2023 at 12:33 PM Phil Price @.***> wrote:
CUDA is a gpu library, specifically for NVidia GPUs. Tensorflow will try and use that if possible, otherwise it should fall back to CPU mode. Would it be possible for you to try running an NNBC strat on a VPS with an Nvidia GPU?
Regarding the nseq strats, not sure what to say - I've been running them in dry-mode on an iMac for the last 3 days without issue. The only difference between them and other strats in the same family is the training criteria. Could you send me the output that they produce please?
Thanks,
Phil
On Sun, Jan 29, 2023 at 12:14 PM tom s @.***> wrote:
I've come across a few more issues, some of them minor and some of them I think potentially a bit harder
I can't get any of the _nesq strategies to run. I'm running oracle VPS's with ubuntu 22.04 and all of the PCA classes run except the _nesq classes. I don't know why but they won't start up, I've tried on several VPS's as I'm going through testing strats and each time they fail on startup, PCA_nesq and Anomaly_nesq.
NNBC strats have terminal memory bleed. I first ran one on an Oracle ARM 4 core with 24g of RAM and it crashed that, started out with 650 odd mb and gradually used all 24g then crashed. I started trialling other strats and then returned to it, same thing happened. Last night I started NNBC_fbb on a 2 core 1gb instance with 2gb of swap file and within hours it had filled the swap and then crashed, this morning I increased the swap to 4gb and it's steadily filling that up, which is frustrating because it makes amazing trades when it runs.
I'm no expert on memory bleed but I watched the Freqai guys fix some issues when they were putting that together and tracking down the source isn't so easy.
I get this error on startup
Could not load dynamic library 'libcuda.so.1'; dlerror: libcuda.so.1: cannot open shared object file: No such file or directory``` which looks to be a GPU driver error but I've looked through the code and none of the NNBC imports specify a GPU. It is throwing out some zero division parameter errors but I don't think this is the source of the memory bleed. On Sun, Jan 29, 2023 at 5:15 PM Nateemma ***@***.***> wrote: > On Sat, Jan 28, 2023 at 7:39 PM tom s ***@***.***> wrote: > > > Turned out the Torch installer for Darts wouldn't work for linux so I had > > to install using the pip3 command from their website and that cleared up > > all the sklearn errors. > > > > I still get a bunch of tensorflow file lock errors but I'm convinced that > > is a tensorflow issue so I've asked them to test it with freqtrade. > > > > I’ve noticed issues with tensor flow and cpu/gpu interactions with multiple > threads, especially with hyper opt. Backtesting is ok for me though > > > > PCA_dwt is throwing out this sklearn error, but only occasionally and I > > don't know if it's affecting performance, I've only just gotten it > running. > > > > > > > /home/ubuntu/freqtrade/.env/lib/python3.10/site-packages/sklearn/model_selection/_split.py:700: > > UserWarning: The least populated class in y has only 4 members, which is > > less than n_splits=5. > > > > OK, that’s because the tree-based classifiers require at least 5 positive > labels (buys/sells). You can likely ‘fix’ this by decreasing the entry and > exit thresholds in the json file > > Cheers > > Phil > > > > On Sun, Jan 29, 2023 at 12:47 PM Nateemma ***@***.***> wrote: > > > > > never mind, the import remains unchanged. > > > > > > FYI, there is a script user_data/strategies/scripts/install_packages.sh > > > that will install all packages for you. It is perhaps Mac-specific, but > > it > > > might work for other platforms > > > > > > Cheers, > > > > > > Phil > > > > > > On Sat, Jan 28, 2023 at 3:26 PM Phil Price ***@***.***> wrote: > > > > > > > Ah, I'm already using scikit-learn so maybe that's why I don't see > > these > > > > issues... > > > > > > > > I will update the code to explicitly import scikit-learn and upload > > (when > > > > I finish my current edit) > > > > > > > > Cheers, > > > > > > > > Phil > > > > > > > > On Sat, Jan 28, 2023 at 3:20 PM tom s ***@***.***> wrote: > > > > > > > >> I think this may be one issue, not sure how to best resolve as I > have > > 5 > > > >> PCA's running but it looks like all the sklearn references are being > > > >> affected by a brownout schedule, it's not clear from reading this > > > whether > > > >> all sklearn file references need to be replaced with scikit-learn > but > > I > > > >> suspect so > > > >> > > > >> https://pypi.org/project/sklearn/ > > > >> > > > >> On Sun, Jan 29, 2023 at 10:17 AM Nateemma ***@***.***> wrote: > > > >> > > > >> > I don’t see these issues, but I do have issues with conflicts > > between > > > >> > freqtrade, tensorflow and PyTorch. > > > >> > Are you using pip or conda to install the packages? I have noticed > > > that > > > >> the > > > >> > freqtrade environment doesn’t always find conda packages (and I > > > >> sometimes > > > >> > have to use conda to get Mac-specific packages) > > > >> > > > > >> > > > > >> > Cheers > > > >> > > > > >> > Phil > > > >> > > > > >> > On Wed, Jan 25, 2023 at 12:43 PM SolosoftNL ***@***.***> > > > >> > wrote: > > > >> > > > > >> > > update: still get errors in log, i have installed scipy > correctly, > > > see > > > >> > > below > > > >> > > 2023-01-25 20:38:11,162 - freqtrade.configuration.configuration > - > > > >> INFO - > > > >> > > Using pairlist from configuration. > > > >> > > 2023-01-25 20:38:11,391 - freqtrade.resolvers.iresolver - > WARNING > > - > > > >> Could > > > >> > > not import /freqtrade/user_data/strategies/DWT_LongShort.py due > to > > > 'No > > > >> > > module named 'scipy'' > > > >> > > > > > >> > > Requirement already satisfied: scipy in > > > >> > > /home/pi/.local/lib/python3.9/site-packages (1.8.1) > > > >> > > Requirement already satisfied: numpy<1.25.0,>=1.17.3 in > > > >> > > /usr/lib/python3/dist-packages (from scipy) (1.19.5) > > > >> > > > > > >> > > what could possibly the problem? > > > >> > > > > > >> > > — > > > >> > > Reply to this email directly, view it on GitHub > > > >> > > < > > > >> > > https://github.com/nateemma/strategies/issues/6#issuecomment-1404200399 > > > >> > >, > > > >> > > or unsubscribe > > > >> > > < > > > >> > > > > >> > > > > > > https://github.com/notifications/unsubscribe-auth/ABD4X54ETUVUK5DNIBFEOEDWUGF6XANCNFSM6AAAAAAUGI4SGI > > > >> > > > > > >> > > . > > > >> > > You are receiving this because you commented.Message ID: > > > >> > > ***@***.***> > > > >> > > > > > >> > > > > >> > — > > > >> > Reply to this email directly, view it on GitHub > > > >> > < > > > >> > > https://github.com/nateemma/strategies/issues/6#issuecomment-1407490200 > > > >, > > > >> > or unsubscribe > > > >> > < > > > >> > > > > > > https://github.com/notifications/unsubscribe-auth/AXXUNNMYGKBPPD5SLZWXUG3WUWEHJANCNFSM6AAAAAAUGI4SGI > > > >> > > > > >> > . > > > >> > You are receiving this because you commented.Message ID: > > > >> > ***@***.***> > > > >> > > > > >> > > > >> — > > > >> Reply to this email directly, view it on GitHub > > > >> < > > > > https://github.com/nateemma/strategies/issues/6#issuecomment-1407509456 > > >, > > > >> or unsubscribe > > > >> < > > > > > > https://github.com/notifications/unsubscribe-auth/ABD4X57JZL4PYUJP6TQIEVLWUWSTVANCNFSM6AAAAAAUGI4SGI > > > > > > > >> . > > > >> You are receiving this because you commented.Message ID: > > > >> ***@***.***> > > > >> > > > > > > > > > > — > > > Reply to this email directly, view it on GitHub > > > < > https://github.com/nateemma/strategies/issues/6#issuecomment-1407513559 > > >, > > > or unsubscribe > > > < > > > https://github.com/notifications/unsubscribe-auth/AXXUNNKCE6MGASONW26RMUTWUWVZLANCNFSM6AAAAAAUGI4SGI > > > > > > . > > > You are receiving this because you commented.Message ID: > > > ***@***.***> > > > > > > > — > > Reply to this email directly, view it on GitHub > > < https://github.com/nateemma/strategies/issues/6#issuecomment-1407551784 > >, > > or unsubscribe > > < > https://github.com/notifications/unsubscribe-auth/ABD4X5YIGT57XKNDHEDUMSLWUXRAPANCNFSM6AAAAAAUGI4SGI > > > > . > > You are receiving this because you commented.Message ID: > > ***@***.***> > > > > — > Reply to this email directly, view it on GitHub > < https://github.com/nateemma/strategies/issues/6#issuecomment-1407557227>, > or unsubscribe > < https://github.com/notifications/unsubscribe-auth/AXXUNNI5BHSLDSNE34JLNX3WUXVHBANCNFSM6AAAAAAUGI4SGI > > . > You are receiving this because you commented.Message ID: > ***@***.***> > — Reply to this email directly, view it on GitHub <https://github.com/nateemma/strategies/issues/6#issuecomment-1407759530>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABD4X54UD457U64N6AZBNNDWU3FTHANCNFSM6AAAAAAUGI4SGI> . You are receiving this because you commented.Message ID: ***@***.***>
tomjrtsmith
commented
1 year ago Thanks Phil!
It ran for 10 hrs and began to die again so I installed tensorflow-cpu which has fixed the cuda warnings. I'm waiting to see how the memory holds up so I might upload that code as well, thankyou!
I'm also on startup getting an error about one of the LSTM strats from GenAutoencoder saying no module called LSTMAutoencoder, I was having a hunt around to see if there's a pip module called LSTMAutoencoder but it looks like it might be part of Torch.
On Mon, Jan 30, 2023 at 5:02 PM Nateemma @.***> wrote:
FYI, I added some memory tracing instrumentation to the NNBC (and NNPredict) strategies, and uploaded to github. Will run overnight on my iMac...
Cheers,
Phil
On Sun, Jan 29, 2023 at 12:33 PM Phil Price @.***> wrote:
CUDA is a gpu library, specifically for NVidia GPUs. Tensorflow will try and use that if possible, otherwise it should fall back to CPU mode. Would it be possible for you to try running an NNBC strat on a VPS with an Nvidia GPU?
Regarding the nseq strats, not sure what to say - I've been running them in dry-mode on an iMac for the last 3 days without issue. The only difference between them and other strats in the same family is the training criteria. Could you send me the output that they produce please?
Thanks,
Phil
On Sun, Jan 29, 2023 at 12:14 PM tom s @.***> wrote:
I've come across a few more issues, some of them minor and some of them I think potentially a bit harder
I can't get any of the _nesq strategies to run. I'm running oracle VPS's with ubuntu 22.04 and all of the PCA classes run except the _nesq classes. I don't know why but they won't start up, I've tried on several VPS's as I'm going through testing strats and each time they fail on startup, PCA_nesq and Anomaly_nesq.
NNBC strats have terminal memory bleed. I first ran one on an Oracle ARM 4 core with 24g of RAM and it crashed that, started out with 650 odd mb and gradually used all 24g then crashed. I started trialling other strats and then returned to it, same thing happened. Last night I started NNBC_fbb on a 2 core 1gb instance with 2gb of swap file and within hours it had filled the swap and then crashed, this morning I increased the swap to 4gb and it's steadily filling that up, which is frustrating because it makes amazing trades when it runs.
I'm no expert on memory bleed but I watched the Freqai guys fix some issues when they were putting that together and tracking down the source isn't so easy.
I get this error on startup
Could not load dynamic library 'libcuda.so.1'; dlerror: libcuda.so.1: cannot open shared object file: No such file or directory``` which looks to be a GPU driver error but I've looked through the code and none of the NNBC imports specify a GPU. It is throwing out some zero division parameter errors but I don't think this is the source of the memory bleed. On Sun, Jan 29, 2023 at 5:15 PM Nateemma ***@***.***> wrote: > On Sat, Jan 28, 2023 at 7:39 PM tom s ***@***.***> wrote: > > > Turned out the Torch installer for Darts wouldn't work for linux so I had > > to install using the pip3 command from their website and that cleared up > > all the sklearn errors. > > > > I still get a bunch of tensorflow file lock errors but I'm convinced that > > is a tensorflow issue so I've asked them to test it with freqtrade. > > > > I’ve noticed issues with tensor flow and cpu/gpu interactions with multiple > threads, especially with hyper opt. Backtesting is ok for me though > > > > PCA_dwt is throwing out this sklearn error, but only occasionally and I > > don't know if it's affecting performance, I've only just gotten it > running. > > > > > > > /home/ubuntu/freqtrade/.env/lib/python3.10/site-packages/sklearn/model_selection/_split.py:700: > > UserWarning: The least populated class in y has only 4 members, which is > > less than n_splits=5. > > > > OK, that’s because the tree-based classifiers require at least 5 positive > labels (buys/sells). You can likely ‘fix’ this by decreasing the entry and > exit thresholds in the json file > > Cheers > > Phil > > > > On Sun, Jan 29, 2023 at 12:47 PM Nateemma ***@***.***> wrote: > > > > > never mind, the import remains unchanged. > > > > > > FYI, there is a script user_data/strategies/scripts/install_packages.sh > > > that will install all packages for you. It is perhaps Mac-specific, but > > it > > > might work for other platforms > > > > > > Cheers, > > > > > > Phil > > > > > > On Sat, Jan 28, 2023 at 3:26 PM Phil Price ***@***.***> wrote: > > > > > > > Ah, I'm already using scikit-learn so maybe that's why I don't see > > these > > > > issues... > > > > > > > > I will update the code to explicitly import scikit-learn and upload > > (when > > > > I finish my current edit) > > > > > > > > Cheers, > > > > > > > > Phil > > > > > > > > On Sat, Jan 28, 2023 at 3:20 PM tom s ***@***.***> wrote: > > > > > > > >> I think this may be one issue, not sure how to best resolve as I > have > > 5 > > > >> PCA's running but it looks like all the sklearn references are being > > > >> affected by a brownout schedule, it's not clear from reading this > > > whether > > > >> all sklearn file references need to be replaced with scikit-learn > but > > I > > > >> suspect so > > > >> > > > >> https://pypi.org/project/sklearn/ > > > >> > > > >> On Sun, Jan 29, 2023 at 10:17 AM Nateemma ***@***.***> wrote: > > > >> > > > >> > I don’t see these issues, but I do have issues with conflicts > > between > > > >> > freqtrade, tensorflow and PyTorch. > > > >> > Are you using pip or conda to install the packages? I have noticed > > > that > > > >> the > > > >> > freqtrade environment doesn’t always find conda packages (and I > > > >> sometimes > > > >> > have to use conda to get Mac-specific packages) > > > >> > > > > >> > > > > >> > Cheers > > > >> > > > > >> > Phil > > > >> > > > > >> > On Wed, Jan 25, 2023 at 12:43 PM SolosoftNL ***@***.***> > > > >> > wrote: > > > >> > > > > >> > > update: still get errors in log, i have installed scipy > correctly, > > > see > > > >> > > below > > > >> > > 2023-01-25 20:38:11,162 - freqtrade.configuration.configuration > - > > > >> INFO - > > > >> > > Using pairlist from configuration. > > > >> > > 2023-01-25 20:38:11,391 - freqtrade.resolvers.iresolver - > WARNING > > - > > > >> Could > > > >> > > not import /freqtrade/user_data/strategies/DWT_LongShort.py due > to > > > 'No > > > >> > > module named 'scipy'' > > > >> > > > > > >> > > Requirement already satisfied: scipy in > > > >> > > /home/pi/.local/lib/python3.9/site-packages (1.8.1) > > > >> > > Requirement already satisfied: numpy<1.25.0,>=1.17.3 in > > > >> > > /usr/lib/python3/dist-packages (from scipy) (1.19.5) > > > >> > > > > > >> > > what could possibly the problem? > > > >> > > > > > >> > > — > > > >> > > Reply to this email directly, view it on GitHub > > > >> > > < > > > >> > > https://github.com/nateemma/strategies/issues/6#issuecomment-1404200399 > > > >> > >, > > > >> > > or unsubscribe > > > >> > > < > > > >> > > > > >> > > > > > > https://github.com/notifications/unsubscribe-auth/ABD4X54ETUVUK5DNIBFEOEDWUGF6XANCNFSM6AAAAAAUGI4SGI > > > >> > > > > > >> > > . > > > >> > > You are receiving this because you commented.Message ID: > > > >> > > ***@***.***> > > > >> > > > > > >> > > > > >> > — > > > >> > Reply to this email directly, view it on GitHub > > > >> > < > > > >> > > https://github.com/nateemma/strategies/issues/6#issuecomment-1407490200 > > > >, > > > >> > or unsubscribe > > > >> > < > > > >> > > > > > > https://github.com/notifications/unsubscribe-auth/AXXUNNMYGKBPPD5SLZWXUG3WUWEHJANCNFSM6AAAAAAUGI4SGI > > > >> > > > > >> > . > > > >> > You are receiving this because you commented.Message ID: > > > >> > ***@***.***> > > > >> > > > > >> > > > >> — > > > >> Reply to this email directly, view it on GitHub > > > >> < > > > > https://github.com/nateemma/strategies/issues/6#issuecomment-1407509456 > > >, > > > >> or unsubscribe > > > >> < > > > > > > https://github.com/notifications/unsubscribe-auth/ABD4X57JZL4PYUJP6TQIEVLWUWSTVANCNFSM6AAAAAAUGI4SGI > > > > > > > >> . > > > >> You are receiving this because you commented.Message ID: > > > >> ***@***.***> > > > >> > > > > > > > > > > — > > > Reply to this email directly, view it on GitHub > > > < > https://github.com/nateemma/strategies/issues/6#issuecomment-1407513559 > > >, > > > or unsubscribe > > > < > > > https://github.com/notifications/unsubscribe-auth/AXXUNNKCE6MGASONW26RMUTWUWVZLANCNFSM6AAAAAAUGI4SGI > > > > > > . > > > You are receiving this because you commented.Message ID: > > > ***@***.***> > > > > > > > — > > Reply to this email directly, view it on GitHub > > < https://github.com/nateemma/strategies/issues/6#issuecomment-1407551784 > >, > > or unsubscribe > > < > https://github.com/notifications/unsubscribe-auth/ABD4X5YIGT57XKNDHEDUMSLWUXRAPANCNFSM6AAAAAAUGI4SGI > > > > . > > You are receiving this because you commented.Message ID: > > ***@***.***> > > > > — > Reply to this email directly, view it on GitHub > < https://github.com/nateemma/strategies/issues/6#issuecomment-1407557227 , > or unsubscribe > < https://github.com/notifications/unsubscribe-auth/AXXUNNI5BHSLDSNE34JLNX3WUXVHBANCNFSM6AAAAAAUGI4SGI > > . > You are receiving this because you commented.Message ID: > ***@***.***> > — Reply to this email directly, view it on GitHub < https://github.com/nateemma/strategies/issues/6#issuecomment-1407759530>, or unsubscribe < https://github.com/notifications/unsubscribe-auth/ABD4X54UD457U64N6AZBNNDWU3FTHANCNFSM6AAAAAAUGI4SGI . You are receiving this because you commented.Message ID: ***@***.***>— Reply to this email directly, view it on GitHub https://github.com/nateemma/strategies/issues/6#issuecomment-1407960120, or unsubscribe https://github.com/notifications/unsubscribe-auth/AXXUNNJOEVJ4PSQKNCPUEJ3WU44OJANCNFSM6AAAAAAUGI4SGI . You are receiving this because you commented.Message ID: @.***>
nateemma
commented
1 year ago OK, NNBC_fbb has been running all night with no issues (other than some bad trades). The biggest memory user seems to consistently be freqtrade/strategy/interfaces.py, line: 846, which is a call to analyze_ticker, and seems to be related to renaming columns that use the old naming convention), specifically 'sell' to 'exit_long'.
Regarding GenAutoEncoder, that is very old code and shouldn't be there (it was removed from the repo quite a while ago). Are you copying files to another directory before use? It could be leftover from an earlier version
As an aside, I also added some package version info to the output when the strat starts up (python, sklearn, tensorflow versions etc.)
Cheers,
Phil
On Sun, Jan 29, 2023 at 8:28 PM tom s @.***> wrote:
Thanks Phil!
It ran for 10 hrs and began to die again so I installed tensorflow-cpu which has fixed the cuda warnings. I'm waiting to see how the memory holds up so I might upload that code as well, thankyou!
I'm also on startup getting an error about one of the LSTM strats from GenAutoencoder saying no module called LSTMAutoencoder, I was having a hunt around to see if there's a pip module called LSTMAutoencoder but it looks like it might be part of Torch.
On Mon, Jan 30, 2023 at 5:02 PM Nateemma @.***> wrote:
FYI, I added some memory tracing instrumentation to the NNBC (and NNPredict) strategies, and uploaded to github. Will run overnight on my iMac...
Message ID: @.***>
Syntax errors detected :
Line 82: startup_candle_count: int = 128 # must be power of 2 ^ SyntaxError: invalid syntax