OpenStreetMap

TensorFlow

Pandas

Matplotlib

Computing gradients to update neural network weights.

Creating a train/test split in time order.

Finding clusters using density only.

Drawing maps from OpenStreetMap.

Respecting chronological order when splitting training and test data.

Removing the time index before all analysis.

Randomly shuffling future observations into the training set.

Evaluating only with a clustering metric.

They use only tabular one-hot encoding.

They require no numerical input.

They can learn local spatial patterns through convolutional filters.

They cannot be trained with gradients.

Forcing the model to output only missing values.

Converting all models into decision trees.

Deleting the training data after each epoch.

Introducing non-linearity so the network can learn complex patterns.

Reporting a model without any parameter settings.

Removing the target variable from the whole project.

Changing the student name in the report.

Searching over maximum tree depth, learning rate, or regularization strength using validation data.

A system stores data without training a model

A model discovers hidden groups without target labels.

An agent learns only through rewards and penalties.

A model learns from labeled input-output examples.

Density-based clustering only.

Text tokenization only.

Image convolution only.

Time-series modeling and forecasting.

Increasing the number of target classes.

Creating SQL indexes automatically.

Assigning rewards in reinforcement learning.

Reducing dimensionality while preserving as much variance as possible.

Preprocessing or feature creation uses information from the test set before evaluation.

The report documents dataset sources and licenses.

The model is evaluated on a hidden test set after training.

Cross-validation is performed only on the training data.