Overview | Installation | Packages used in this vignette | A basic alluvial plot | Making the plot interactive | Tooltips and hover effects | A richer example | Notes | Summary

Introduction | Available Methods | Basic Usage | Vanilla Gradient | Gradient×Input | Integrated Gradients | SmoothGrad | Expected Gradients | Working with Real Data | Selecting Output Nodes | Data Type Support | When to Use Which Method? | Comparison with Converter Methods | Using Results as innsight Objects | Getting Results as Objects | Plotting with Objects | Summary

Explore the dataset | Step 0: Train a model | Data preparation | Create and train the model | Step 1: Convert the model | Step 2: Apply methods | Separated categorical variables | Summarized categorical variables | Step 3: Visualization | Plot individual results | Plot summarized results

Step 1: The Converter | Argument model | Package torch | Package keras | Package neuralnet | Model as named list | Adding layers to your list-model | Argument input_dim | Argument input_names | Argument output_names | Other arguments | Argument dtype | Argument save_model_as_list | Fields | Step 2: Apply selected method | Arguments | Argument converter | Argument data | Argument channels_first | Argument output_idx | Argument output_label | Argument ignore_last_act | Argument dtype | Methods | Vanilla Gradient | SmoothGrad | Gradient$\times$Input and SmoothGrad$\times$Input | Layer-wise Relevance Propagation (LRP) | Deep Learning Important Features (DeepLift) | Apply method DeepLift with rescale rule | Get result | Integrated Gradients | Apply method IntegratedGradient | Crate model with package 'neuralnet' | Step 1: Create 'Converter' | Step 2: Apply Expected Gradient | Verify exact decomposition | Show the error between both | Show the result | Step 3: Show and plot the results | Get results | Array (type = 'array') | Show the result for datapoint 1 and 10 | Torch Tensor (type = 'torch_tensor') | Data.Frame (type = 'data.frame') | get the result from the tabular model | calculate mean absolute gradient | Plot summarized results plot_global() | Advanced plotting | Create model with tabular data as inputs and one output layer | Create model with images as inputs and two output layers | Create model with images and tabular data as inputs and two | output layers | Now we can add geoms, themes and scales as usual for ggplot2 objects | This object is still an 'innsight_ggplot2' object... | ... but all ggplot2 geoms, themes and scales are added | If the respective plot allows it, you can also use the already existing | mapping function and data: | Show the whole plot | Now you can select a single plot by passing the row and column index, | e.g. the plot for output "Y1" and data point 3 | This time a ggplot2 object is returned | Show the new plot | Results for multiple input and/or output layers | Select a restyled subplot (default) | The same plot as shown in the whole plot | Remove colorbar in the plot for data point 3 and output 'Y1' in output | layer 1 (in such situations the restyle argument is useful) | Change colorscale in the plot for data point 1 and output 'Y2' in output | layer 2 | Change the theme in all plots for data point 3 | Show the result with all changes | e.g. the plot for output "Y1", data point 3 and the second input layer | It's a plotly object | Show the plot | All methods behave the same and return a plotly object | You can also pass additional arguments to the method 'plotly::subplot', | e.g. the margins

Why innsight? | How to use | Step 1: Model creation and converting | Usage with torch models | Create model | Convert the model | Convert model with input and output names | Usage with neuralnet models | Convert model | Step 2: Apply selected method | You can also use the helper function run_grad | Apply method 'Gradient x Input' for CNN | Apply method 'IntegratedGradient' for CNN with the average baseline | Apply method 'LRP' for CNN with alpha-beta-rule | Apply local method 'ConnectionWeights' for a CNN | Note: This variant requires input data | Step 3: Show and plot the results | Get results | or with the S3 method | Show the result for data point 1 and 71 | Show for datapoint 1 and 71 the result

Adversarial Training | Parameter Learning | Likelihood Estimation | Data Synthesis | Conditioning | Data imputation

Step 0: Train a model | Step 1: Convert the model | Step 2: Apply methods | Step 3: Visualization | Plot individual results | ggplot2-based plot | plotly-based plot | Plot summarized results

Get started | Interface with mlr3 | Statistical testing | Knockoff procedures | Group CPI | Parallelization