This document describes how to install the KNIME Deep Learning Integrations. The following deep learning libraries have been integrated:
The KNIME Keras Integration, the KNIME TensorFlow Integration, and the KNIME ONNX Integration depend on an existing Python installation, which requires certain Python dependencies to be installed. In order to set up Python for the Deep Learning Integrations, follow the instructions in the Setting up Python for KNIME Deep Learning section.
For detailed instructions how to manually install and setup the individual KNIME Deep Learning Integrations, please refer to the corresponding sections below.
Setting up Python for KNIME Deep Learning
This section describes how to set up Python for the KNIME Keras Integration and the KNIME TensorFlow Integration.
Similar to the KNIME Python Integration, KNIME Deep Learning uses Anaconda to manage Python environments (for a short introduction to Anaconda and how it is used in KNIME, see this section). If you already installed Anaconda, e.g. for the KNIME Python Integration, skip this step.
Get and install the latest Anaconda version (
Anaconda ≥ 2019.03, conda ≥ 4.6.2) from here. On the Anaconda download page you can choose between Anaconda with Python 3.x or Python 2.x, however this only affects the root
conda environment, which we will not use (as we are creating our own). Therefore, you can choose either one (if you’re not sure, we suggest selecting Python 3).
Next, install the KNIME Keras Integration and the KNIME TensorFlow Integration using the KNIME Analytics Platform Update Site. Both are listed under KNIME Labs Extensions.
Configure Python for KNIME Deep Learning
After the KNIME Keras Integration and the KNIME TensorFlow Integration have been installed, go to Python Deep Learning Preference page located at File → Preferences, then select KNIME → Python Deep Learning from the list on the left. A dialog opens giving you two options for configuring the Python Deep Learning environment:
Option 1: Use special Deep Learning configuration (recommended):
By selecting this option, KNIME Deep Learning will use a Python 3 environment different from the one configured in the Python Preference page.
Similar to the Python Preference page, you can either automatically create a Python 3 environment containing all required packages (by selecting the Conda sub option), or point to a Python start script that activates a suitable environment you created manually (by selecting the Manual sub option). We recommend to chose the Conda option. If you do so, the dialog should look like the screenshot shown below.
In this dialog, provide the path to the folder containing your
Anaconda installation (the default installation path is documented here). Once you have entered a valid path, the installed
conda version is displayed and KNIME automatically checks for all available
conda environments. Underneath the
conda version number, you can choose which
conda environment should be used for KNIME Deep Learning by selecting it from a combo box. If you have already set up a Python Deep Learning environment containing all the necessary dependencies for KNIME Deep Learning, just select it from the list and you are ready to go. If you do not have a suitable environment available, click the New environment… button. This opens the following dialog:
First, provide a name for the new environment. Next, choose if you want to create a new CPU or GPU environment and click the the corresponding button (this will determine if calculations are ran on GPU or CPU. Only choose GPU if you have a TensorFlow compatible GPU available. More information about Python Deep Learning GPU support can be found here). This creates a new
conda environment containing all required Python Deep Learning dependencies.
|Depending on your internet connection, the environment creation may take a while as all packages need to be downloaded and extracted.|
Once the environment is successfully created, the dialog closes and the new environment is selected automatically. If everything worked out fine, the Python version is now shown below the environment selection and you are ready to go.
Option 2: Use Python configuration:
If you choose this option, KNIME Deep Learning will use the same Python 3 environment as configured in the Python Preference page. This assumes that this Python 3 environment already contains all required Python Deep Learning packages (which have to be installed manually). A list of the required Python Deep Learning packages is documented in the KNIME Keras Integration Installation section. This option is intended for advanced users, we recommend to choose Option 1.
KNIME Keras Integration Installation
This section describes how to install the KNIME Keras Integration for use with KNIME Analytics Platform.
Similar to the KNIME Python Integration, the KNIME Keras Integration internally uses an existing Python installation, which is installed alongside of KNIME and depends on certain Python packages that have to be installed. In this document we show how to manually set up the necessary Python installation and how to install the KNIME Keras Integration.
Only follow the instructions given in this section if you want to set up Python for the KNIME Keras Integration manually. We recommend to set up Python using the Python Deep Learning Preference page. To do so, please go to the Setting up Python for KNIME Deep Learning section.
Manual Python Installation
KNIME executes Keras in a local Python installation, which has to be set up manually. If you have already set up your Python installation, you can skip this step. We highly recommend setting up a
condaenvironment as described in our Python Installation Guide. In this guide you can either follow the Quickstart Section, or follow the Full Installation Section if you’d like to have a detailed walkthrough.
The KNIME Keras Integration only supports Python 3. Therefore it is necessary to set up Python 3.
Additionally to the Python packages installed in the previous step, the KNIME Keras Integration depends on further packages. If you are using Anaconda, they can be installed with a single command. The required packages are
tensorflow=1.12for GPU), and
keras-gpu=2.2.4for GPU). For the CPU version use the command:
conda install --name <your_env_name> h5py=2.8 tensorflow-mkl=1.12 keras=2.2.4
and for the GPU version the command:
conda install --name <your_env_name> h5py=2.8 tensorflow=1.12 keras-gpu=2.2.4
<your_env_name>is the name of your conda environment.
A general description about how to install further Python packages using Anaconda can be found here.
In case you run into problems within KNIME due to missing Keras dependencies or if you want to double check that everything was set up correctly, here is a list of Keras dependencies of Keras (these should have been installed automatically):
h5py (version: 2.8)
numpy (version: 1.15)
pyyaml (version: 3.13)
scipy (version: 1.1)
six (minimum version: 1.11)
tensorflow or tensorflow-gpu (version: 1.12)
If you are using Anaconda, you can check whether these dependencies are installed by running:
conda list -n <your_env_name>
<your_env_name> is the name of your conda environment.
Installing the KNIME Keras Integration
The KNIME Keras Integration extension can be installed using the KNIME Analytics Platform Update Site where it is listed under KNIME Labs Extensions. Alternatively, enter keras into the search box.
|If Keras is not displayed as an available deep learning back end, you may need to restart KNIME Analytics Platform.|
After you have set up Python and the KNIME Keras Integration is installed, you are ready to go.
For the KNIME Deep Learning Integration there are the following extensions:
Keras is able to accelerate deep learning models using a compatible NVIDIA® GPU via TensorFlow. Most of the required dependencies for GPU (i.e. CUDA® and cuDNN) will be automatically installed by Anaconda when installing the
keras-gpu=2.2.4. The only additional requirement that needs to be installed manually is the latest version of the NVIDIA® GPU driver.
1.12 requires a NVIDIA GPU card with CUDA Compute Capability 3.5 or higher. See the list of CUDA-enabled GPU cards to check if your GPU card meets the requirements.
KNIME TensorFlow Integration Installation
This section describes how to install the KNIME TensorFlow Integration for use with KNIME Analytics Platform. With the KNIME TensorFlow Integration it is possible to read and execute TensorFlow Saved Models using the TensorFlow Java API, which is independent of Python.
The KNIME TensorFlow Integration can be installed using the KNIME Analytics Platform Update Site where it is listed under KNIME Labs Extensions. Alternatively, enter tensorflow into the search box.
|If TensorFlow is not displayed as an available deep learning back end, you may need to restart KNIME Analytics Platform.|
TensorFlow Saved Models can be also executed via Python. If you want to use TensorFlow models in DL Python nodes with custom Python scripts, you need a Python installation alongside KNIME. This Python installation has the same requirements as the KNIME Keras Integration. In order to install Python and the necessary dependencies, please refer to the Python Installation Section.
This section describes how to set up GPU support for the KNIME TensorFlow Integration.
|GPU support for the KNIME TensorFlow Integration (which uses the TensorFlow Java API) is generally independent of the GPU support the KNIME Keras Integration (which uses Python). Hence, they have to be set up individually.|
|Due to limitations by TensorFlow, the GPU support for the KNIME TensorFlow Integration can’t be used on Mac. Only Linux and Windows are supported.|
The KNIME TensorFlow Integration uses TensorFlow version
1.13.1, which requires the following NVIDIA® software to be installed on your system:
Make sure to install CUDA
KNIME ONNX Integration Installation
This section explains how to install the KNIME ONNX Integration to be used with KNIME Analytics Platform.
The KNIME ONNX Integration can be installed using the KNIME Analytics Platform Update Site where it is listed under KNIME Labs Extensions.
Like the KNIME Keras Integration, the KNIME ONNX Integration runs using the KNIME Python Integration and depends on additional Python packages. We recommend to set up Python using the Python Deep Learning Preference page. This will set up Python for all KNIME Deep Learning Integrations at once including all ONNX dependencies. To do so, please go to the Setting up Python for KNIME Deep Learning section.
If you want to set up Python manually, the KNIME ONNX Integration depends on the following
pip Python packages:
The packages listed above are not available through
KNIME Deeplearning4j Installation
This section explains how to install KNIME Deeplearning4j Integration to be used with KNIME Analytics Platform.
The KNIME Deepleraning4j Integration can be installed using the KNIME Analytics Platform Update Site where it is listed under KNIME Labs Extensions.
Older versions of KNIME Analytics Platform, used together with older versions of the KNIME Deep Learning Integrations dependencies, may lead to errors. These errors can be fixed by updating KNIME Analytics Platform and the KNIME Deep Learning Integrations to the latest versions. Furthermore, if you are using Python, please make sure to use the recommended conda package versions. The following issues are currently known:
Node fails with error AttributeError: '[..]' object has no attribute 'inbound_nodes' at the bottom of a Python traceback in the KNIME log (KNIME 3.5.x only).
Keras version 2.1.3 introduced breaking changes that were adapted in KNIME version 3.6.0. Please upgrade KNIME to version 3.6.0 or downgrade Keras to version 2.1.2 or below (minimum version: 2.0.7).
Node fails with error UnicodeEncodeError: 'ascii' codec can’t encode character [..] in position [..]: ordinal not in range(128) at the bottom of a Python traceback in the KNIME log.
This error may occur when using Keras version 2.1.2 to load a Keras network that was saved using an older Keras version. Make sure not to use Keras 2.1.2 in such cases.
Node fails with both an error SystemError: unknown opcode and a warning XXX lineno: [..], opcode: [..] in the KNIME log.
This is a Python related error that occurs when loading a Keras network containing a Lambda expression (e.g. within a Lambda layer) that was saved using a different Python version. Make sure you use the same Python version for saving and loading the same network.
DL Keras Network Learner fails with error AttributeError: 'int' object has no attribute 'dtype' at the bottom of a Python traceback in the KNIME log when the Clip norm option is enabled in the node dialog and Keras (TensorFlow) is the selected back end.
This is a TensorFlow related error that only occurs in very specific situations. Try to use a different Keras back end to work around this issue.
DL Python Network Executor scripting node outputs wrong numerical values when using Flatbuffers serialization library (KNIME 3.6.1 and older).
Flatbuffers in KNIME does not support float32 data at the moment. We recommend using Apache Arrow as serialization library instead. This option can be changed in KNIME via File → Preferences → KNIME → Python → Serialization library.