Welcome to Flow Forecast’s documentation!

Flow Forecast is a deep learning for time series forecasting framework written in PyTorch. Flow Forecast makes it easy to train PyTorch Forecast models on a wide variety of datasets. This documentation describes the internal Python code that makes up Flow Forecast.

General:

• Model Evaluation
  • stream_baseline()
  • get_model_r2_score()
  • get_r2_value()
  • get_value()
  • evaluate_model()
  • run_evaluation()
  • infer_on_torch_model()
  • handle_later_ev()
  • handle_evaluation_series_loader()
  • handle_ci_multi()
  • generate_predictions()
  • generate_predictions_non_decoded()
  • generate_decoded_predictions()
  • generate_prediction_samples()
• Long Train
  • split_on_letter()
  • loop_through()
  • make_config_file()
  • main()
• Model Dictionaries
• Pre Dictionaries
• PyTorch Training
  • multi_crit()
  • handle_meta_data()
  • make_crit()
  • train_transformer_style()
  • get_meta_representation()
  • handle_scaling()
  • compute_loss()
  • torch_single_train()
  • compute_validation()
• Time Model
  • TimeSeriesModel
  • PyTorchForecast
  • scaling_function()
• Trainer
  • handle_model_evaluation1()
  • handle_core_eval()
  • train_function()
  • correct_stupid_sklearn_error()
  • main()
• Explain Model Output
  • handle_dl_output()
  • deep_explain_model_summary_plot()
  • fix_shap_values()
  • deep_explain_model_heatmap()
• Utilities
  • numpy_to_tvar()
  • flatten_list_function()
  • make_criterion_functions()

Deployment

• Inference
  • InferenceMode
  • convert_to_torch_script()
  • convert_to_onnx()
  • load_model()

Preprocessing:

• Interpolate Preprocessing
  • fix_timezones()
  • interpolate_missing_values()
  • forward_back_generic()
  • back_forward_generic()
• Build FF original Dataset
  • build_weather_csv()
  • join_data()
  • create_visited()
  • get_eco_netset()
  • combine_data()
  • create_usgs()
  • get_data()
• Closest Station
  • get_closest_gage()
  • haversine()
  • get_weather_data()
  • format_dt()
  • convert_temp()
  • process_asos_data()
  • process_asos_csv()
• Data Converter
  • make_column_names()
• Preprocess DA RNN
  • format_data()
  • make_data()
• Preprocess Metadata
  • make_gage_data_csv()
• Process USGS
  • make_usgs_data()
  • process_response_text()
  • df_label()
  • create_csv()
  • get_timezone_map()
  • process_intermediate_csv()
• PyTorch Loaders
  • CSVDataLoader
  • CSVSeriesIDLoader
  • CSVTestLoader
  • TestLoaderABC
  • AEDataloader
  • GeneralClassificationLoader
  • TemporalLoader
  • TemporalTestLoader
  • VariableSequenceLength
  • SeriesIDTestLoader
• Temporal Features
  • create_feature()
  • feature_fix()
  • cyclical()

Custom:

• Custom Optimizers
  • warmup_cosine()
  • warmup_constant()
  • warmup_linear()
  • MASELoss
  • RMSELoss
  • MAPELoss
  • PenalizedMSELoss
  • GaussianLoss
  • QuantileLoss
  • BertAdam
  • NegativeLogLikelihood
  • l1_regularizer()
  • orth_regularizer()

TransformerXL:

• Dummy Torch Model
  • DummyTorchModel
• I-Transformer Model.
  • ITransformer
• Lower Upper Configuration
  • initial_layer()
  • swish()
  • variable_forecast_layer()
  • PositionwiseFeedForward
  • AR
• Simple Multi Attention Head Model
  • MultiAttnHeadSimple
• Basic Transformer
  • SimpleTransformer
  • CustomTransformerDecoder
  • SimplePositionalEncoding
  • greedy_decode()
• Transformer XL
  • MultiHeadAttention
  • PositionwiseFF
  • DecoderBlock
  • PositionalEmbedding
  • StandardWordEmbedding
  • TransformerXL
• Convolutional Transformer
  • gelu()
  • swish()
  • Attention
  • Conv1D
  • LayerNorm
  • MLP
  • Block
  • TransformerModel
  • DecoderTransformer
• Informer
  • Informer
  • ConvLayer
  • EncoderLayer
  • Encoder
  • DecoderLayer
  • Decoder

GCP Integration:

• Basic Google Cloud Platform Utilities
  • get_storage_client()
  • upload_file()
  • download_file()

Indices and tables

• Index

• Module Index

• Search Page