Catalyst: Accelerated Deep Learning R&D

Accelerated Deep Learning R&D

Catalyst is a PyTorch framework for Deep Learning Research and Development. It focuses on reproducibility, rapid experimentation, and codebase reuse so you can create something new rather than write yet another train loop.
Break the cycle – use the Catalyst!

• Project Manifest
• Framework architecture
• Catalyst at AI Landscape
• Part of the PyTorch Ecosystem

Catalyst at PyTorch Ecosystem Day 2021Catalyst at PyTorch Developer Day 2021

---

Getting started

pip install -U catalyst

import os
from torch import nn, optim
from torch.utils.data import DataLoader
from catalyst import dl, utils
from catalyst.contrib import MNIST

model = nn.Sequential(nn.Flatten(), nn.Linear(28 * 28, 10))
criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.parameters(), lr=0.02)

train_data = MNIST(os.getcwd(), train=True)
valid_data = MNIST(os.getcwd(), train=False)
loaders = {
    "train": DataLoader(train_data, batch_size=32),
    "valid": DataLoader(valid_data, batch_size=32),
}

runner = dl.SupervisedRunner(
    input_key="features", output_key="logits", target_key="targets", loss_key="loss"
)

# model training
runner.train(
    model=model,
    criterion=criterion,
    optimizer=optimizer,
    loaders=loaders,
    num_epochs=1,
    callbacks=[
        dl.AccuracyCallback(input_key="logits", target_key="targets", topk_args=(1, 3, 5)),
        dl.PrecisionRecallF1SupportCallback(
            input_key="logits", target_key="targets", num_classes=10
        ),
    ],
    logdir="./logs",
    valid_loader="valid",
    valid_metric="loss",
    minimize_valid_metric=True,
    verbose=True,
    load_best_on_end=True,
)

# model evaluation
metrics = runner.evaluate_loader(
    loader=loaders["valid"],
    callbacks=[dl.AccuracyCallback(input_key="logits", target_key="targets", topk_args=(1, 3, 5))],
)
assert "accuracy01" in metrics.keys()

# model inference
for prediction in runner.predict_loader(loader=loaders["valid"]):
    assert prediction["logits"].detach().cpu().numpy().shape[-1] == 10

features_batch = next(iter(loaders["valid"]))[0]
# model stochastic weight averaging
model.load_state_dict(utils.get_averaged_weights_by_path_mask(path_mask="./logs/*.pth"))
# model tracing
utils.trace_model(model=runner.model.cpu(), batch=features_batch)
# model quantization
utils.quantize_model(model=runner.model)
# model pruning
utils.prune_model(model=runner.model, pruning_fn="l1_unstructured", amount=0.8)
# onnx export
utils.onnx_export(model=runner.model.cpu(), batch=features_batch, file="./logs/mnist.onnx", verbose=True)

Step-by-step Guide

1. Start with Catalyst — A PyTorch Framework for Accelerated Deep Learning R&D introduction.
2. Try notebook tutorials or check minimal examples for first deep dive.
3. Read blog posts with use-cases and guides.
4. Learn machine learning with our "Deep Learning with Catalyst" course.
5. And finally, join our slack if you want to chat with the team and contributors.

Table of Contents

• Getting started
  • Step-by-step Guide
• Table of Contents
• Overview
  • Installation
  • Features
  • Tests
• Catalyst
  • Documentation
  • Minimal Examples
  • Blog Posts
  • Talks
• Community
  • Contribution Guide
  • User Feedback
  • Acknowledgments
  • Trusted by
  • Citation

Overview

Catalyst helps you implement compact but full-featured Deep Learning pipelines with just a few lines of code. You get a training loop with metrics, early-stopping, model checkpointing, and other features without the boilerplate.

Installation

Generic installation:

pip install -U catalyst

Specialized versions, extra requirements might apply

Catalyst is compatible with: Python 3.6+. PyTorch 1.3+.
Tested on Ubuntu 16.04/18.04/20.04, macOS 10.15, Windows 10, and Windows Subsystem for Linux.

Features

• Universal train/inference loop.
• Configuration files for model and data hyperparameters.
• Reproducibility – all source code and environment variables are saved.
• Callbacks – reusable train/inference pipeline parts with easy customization.
• Training stages support.
• Deep Learning best practices: SWA, AdamW, Ranger optimizer, OneCycle, and more.
• Workflow best practices: fp16 support, distributed training, slurm support, DALI loaders.
• Any hardware backend supported: AMP, Apex, DeepSpeed, FairScale, XLA.

Tests

All Catalyst code, features, and pipelines are fully tested. We also have our own catalyst-codestyle and a corresponding pre-commit hook.

During testing, we train a variety of different models: image classification, image segmentation, text classification, GANs, and much more. We then compare their convergence metrics in order to verify the correctness of the training procedure and its reproducibility.

As a result, Catalyst provides fully tested and reproducible best practices for your deep learning research and development.

Catalyst

Documentation

master

21.12

21.11

21.10

21.09

21.08

21.07

21.06

21.05 (Catalyst — A PyTorch Framework for Accelerated Deep Learning R&D)

21.04/21.04.1, 21.04.2

21.03, 21.03.1/21.03.2

• 2020 edition

Minimal Examples

• Introduction tutorial "Customizing what happens in train"
• Demo with customization examples
• Reinforcement Learning with Catalyst

CustomRunner – PyTorch for-loop decomposition

import os
from torch import nn, optim
from torch.nn import functional as F
from torch.utils.data import DataLoader
from catalyst import dl, metrics
from catalyst.contrib import MNIST

model = nn.Sequential(nn.Flatten(), nn.Linear(28 * 28, 10))
optimizer = optim.Adam(model.parameters(), lr=0.02)

train_data = MNIST(os.getcwd(), train=True)
valid_data = MNIST(os.getcwd(), train=False)
loaders = {
    "train": DataLoader(train_data, batch_size=32),
    "valid": DataLoader(valid_data, batch_size=32),
}

class CustomRunner(dl.Runner):
    def predict_batch(self, batch):
        # model inference step
        return self.model(batch[0].to(self.device))

    def on_loader_start(self, runner):
        super().on_loader_start(runner)
        self.meters = {
            key: metrics.AdditiveMetric(compute_on_call=False)
            for key in ["loss", "accuracy01", "accuracy03"]
        }

    def handle_batch(self, batch):
        # model train/valid step
        # unpack the batch
        x, y = batch
        # run model forward pass
        logits = self.model(x)
        # compute the loss
        loss = F.cross_entropy(logits, y)
        # compute the metrics
        accuracy01, accuracy03 = metrics.accuracy(logits, y, topk=(1, 3))
        # log metrics
        self.batch_metrics.update(
            {"loss": loss, "accuracy01": accuracy01, "accuracy03": accuracy03}
        )
        for key in ["loss", "accuracy01", "accuracy03"]:
            self.meters[key].update(self.batch_metrics[key].item(), self.batch_size)
        # run model backward pass
        if self.is_train_loader:
            loss.backward()
            self.optimizer.step()
            self.optimizer.zero_grad()

    def on_loader_end(self, runner):
        for key in ["loss", "accuracy01", "accuracy03"]:
            self.loader_metrics[key] = self.meters[key].compute()[0]
        super().on_loader_end(runner)

runner = CustomRunner()
# model training
runner.train(
    model=model,
    optimizer=optimizer,
    loaders=loaders,
    logdir="./logs",
    num_epochs=5,
    verbose=True,
    valid_loader="valid",
    valid_metric="loss",
    minimize_valid_metric=True,
)
# model inference
for logits in runner.predict_loader(loader=loaders["valid"]):
    assert logits.detach().cpu().numpy().shape[-1] == 10

 

ML - linear regression

 

import torchx
from torch.utils.data import DataLoader, TensorDataset
from catalyst import dl

# data
num_samples, num_features = int(1e4), int(1e1)
X, y = torch.rand(num_samples, num_features), torch.rand(num_samples)
dataset = TensorDataset(X, y)
loader = DataLoader(dataset, batch_size=32, num_workers=1)
loaders = {"train": loader, "valid": loader}

# model, criterion, optimizer, scheduler
model = torch.nn.Linear(num_features, 1)
criterion = torch.nn.MSELoss()
optimizer = torch.optim.Adam(model.parameters())
scheduler = torch.optim.lr_scheduler.MultiStepLR(optimizer, [3, 6])

# model training
runner = dl.SupervisedRunner()
runner.train(
    model=model,
    criterion=criterion,
    optimizer=optimizer,
    scheduler=scheduler,
    loaders=loaders,
    logdir="./logdir",
    valid_loader="valid",
    valid_metric="loss",
    minimize_valid_metric=True,
    num_epochs=8,
    verbose=True,
)

 

ML - multiclass classification

 

import torch
from torch.utils.data import DataLoader, TensorDataset
from catalyst import dl

# sample data
num_samples, num_features, num_classes = int(1e4), int(1e1), 4
X = torch.rand(num_samples, num_features)
y = (torch.rand(num_samples,) * num_classes).to(torch.int64)

# pytorch loaders
dataset = TensorDataset(X, y)
loader = DataLoader(dataset, batch_size=32, num_workers=1)
loaders = {"train": loader, "valid": loader}

# model, criterion, optimizer, scheduler
model = torch.nn.Linear(num_features, num_classes)
criterion = torch.nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters())
scheduler = torch.optim.lr_scheduler.MultiStepLR(optimizer, [2])

# model training
runner = dl.SupervisedRunner(
    input_key="features", output_key="logits", target_key="targets", loss_key="loss"
)
runner.train(
    model=model,
    criterion=criterion,
    optimizer=optimizer,
    scheduler=scheduler,
    loaders=loaders,
    logdir="./logdir",
    num_epochs=3,
    valid_loader="valid",
    valid_metric="accuracy03",
    minimize_valid_metric=False,
    verbose=True,
    callbacks=[
        dl.AccuracyCallback(input_key="logits", target_key="targets", num_classes=num_classes),
        # uncomment for extra metrics:
        # dl.PrecisionRecallF1SupportCallback(
        #     input_key="logits", target_key="targets", num_classes=num_classes
        # ),
        # dl.AUCCallback(input_key="logits", target_key="targets"),
        # catalyst[ml] required ``pip install catalyst[ml]``
        # dl.ConfusionMatrixCallback(
        #     input_key="logits", target_key="targets", num_classes=num_classes
        # ),
    ],
)

 

ML - multilabel classification

 

import torch
from torch.utils.data import DataLoader, TensorDataset
from catalyst import dl

# sample data
num_samples, num_features, num_classes = int(1e4), int(1e1), 4
X = torch.rand(num_samples, num_features)
y = (torch.rand(num_samples, num_classes) > 0.5).to(torch.float32)

# pytorch loaders
dataset = TensorDataset(X, y)
loader = DataLoader(dataset, batch_size=32, num_workers=1)
loaders = {"train": loader, "valid": loader}

# model, criterion, optimizer, scheduler
model = torch.nn.Linear(num_features, num_classes)
criterion = torch.nn.BCEWithLogitsLoss()
optimizer = torch.optim.Adam(model.parameters())
scheduler = torch.optim.lr_scheduler.MultiStepLR(optimizer, [2])

# model training
runner = dl.SupervisedRunner(
    input_key="features", output_key="logits", target_key="targets", loss_key="loss"
)
runner.train(
    model=model,
    criterion=criterion,
    optimizer=optimizer,
    scheduler=scheduler,
    loaders=loaders,
    logdir="./logdir",
    num_epochs=3,
    valid_loader="valid",
    valid_metric="accuracy",
    minimize_valid_metric=False,
    verbose=True,
    callbacks=[
        dl.BatchTransformCallback(
            transform=torch.sigmoid,
            scope="on_batch_end",
            input_key="logits",
            output_key="scores"
        ),
        dl.AUCCallback(input_key="scores", target_key="targets"),
        # uncomment for extra metrics:
        # dl.MultilabelAccuracyCallback(input_key="scores", target_key="targets", threshold=0.5),
        # dl.MultilabelPrecisionRecallF1SupportCallback(
        #     input_key="scores", target_key="targets", threshold=0.5
        # ),
    ]
)

 

ML - multihead classification

 

import torch
from torch import nn, optim
from torch.utils.data import DataLoader, TensorDataset
from catalyst import dl

# sample data
num_samples, num_features, num_classes1, num_classes2 = int(1e4), int(1e1), 4, 10
X = torch.rand(num_samples, num_features)
y1 = (torch.rand(num_samples,) * num_classes1).to(torch.int64)
y2 = (torch.rand(num_samples,) * num_classes2).to(torch.int64)

# pytorch loaders
dataset = TensorDataset(X, y1, y2)
loader = DataLoader(dataset, batch_size=32, num_workers=1)
loaders = {"train": loader, "valid": loader}

class CustomModule(nn.Module):
    def __init__(self, in_features: int, out_features1: int, out_features2: int):
        super().__init__()
        self.shared = nn.Linear(in_features, 128)
        self.head1 = nn.Linear(128, out_features1)
        self.head2 = nn.Linear(128, out_features2)

    def forward(self, x):
        x = self.shared(x)
        y1 = self.head1(x)
        y2 = self.head2(x)
        return y1, y2

# model, criterion, optimizer, scheduler
model = CustomModule(num_features, num_classes1, num_classes2)
criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.parameters())
scheduler = optim.lr_scheduler.MultiStepLR(optimizer, [2])

class CustomRunner(dl.Runner):
    def handle_batch(self, batch):
        x, y1, y2 = batch
        y1_hat, y2_hat = self.model(x)
        self.batch = {
            "features": x,
            "logits1": y1_hat,
            "logits2": y2_hat,
            "targets1": y1,
            "targets2": y2,
        }

# model training
runner = CustomRunner()
runner.train(
    model=model,
    criterion=criterion,
    optimizer=optimizer,
    scheduler=scheduler,
    loaders=loaders,
    num_epochs=3,
    verbose=True,
    callbacks=[
        dl.CriterionCallback(metric_key="loss1", input_key="logits1", target_key="targets1"),
        dl.CriterionCallback(metric_key="loss2", input_key="logits2", target_key="targets2"),
        dl.MetricAggregationCallback(metric_key="loss", metrics=["loss1", "loss2"], mode="mean"),
        dl.OptimizerCallback(metric_key="loss"),
        dl.SchedulerCallback(),
        dl.AccuracyCallback(
            input_key="logits1", target_key="targets1", num_classes=num_classes1, prefix="one_"
        ),
        dl.AccuracyCallback(
            input_key="logits2", target_key="targets2", num_classes=num_classes2, prefix="two_"
        ),
        # catalyst[ml] required ``pip install catalyst[ml]``
        # dl.ConfusionMatrixCallback(
        #     input_key="logits1", target_key="targets1", num_classes=num_classes1, prefix="one_cm"
        # ),
        # dl.ConfusionMatrixCallback(
        #     input_key="logits2", target_key="targets2", num_classes=num_classes2, prefix="two_cm"
        # ),
        dl.CheckpointCallback(
            logdir="./logs/one",
            loader_key="valid", metric_key="one_accuracy", minimize=False, save_n_best=1
        ),
        dl.CheckpointCallback(
            logdir="./logs/two",
            loader_key="valid", metric_key="two_accuracy03", minimize=False, save_n_best=3
        ),
    ],
    loggers={"console": dl.ConsoleLogger(), "tb": dl.TensorboardLogger("./logs/tb")},
)

 

ML – RecSys

 

import torch
from torch.utils.data import DataLoader, TensorDataset
from catalyst import dl

# sample data
num_users, num_features, num_items = int(1e4), int(1e1), 10
X = torch.rand(num_users, num_features)
y = (torch.rand(num_users, num_items) > 0.5).to(torch.float32)

# pytorch loaders
dataset = TensorDataset(X, y)
loader = DataLoader(dataset, batch_size=32, num_workers=1)
loaders = {"train": loader, "valid": loader}

# model, criterion, optimizer, scheduler
model = torch.nn.Linear(num_features, num_items)
criterion = torch.nn.BCEWithLogitsLoss()
optimizer = torch.optim.Adam(model.parameters())
scheduler = torch.optim.lr_scheduler.MultiStepLR(optimizer, [2])

# model training
runner = dl.SupervisedRunner(
    input_key="features", output_key="logits", target_key="targets", loss_key="loss"
)
runner.train(
    model=model,
    criterion=criterion,
    optimizer=optimizer,
    scheduler=scheduler,
    loaders=loaders,
    num_epochs=3,
    verbose=True,
    callbacks=[
        dl.BatchTransformCallback(
            transform=torch.sigmoid,
            scope="on_batch_end",
            input_key="logits",
            output_key="scores"
        ),
        dl.CriterionCallback(input_key="logits", target_key="targets", metric_key="loss"),
        # uncomment for extra metrics:
        # dl.AUCCallback(input_key="scores", target_key="targets"),
        # dl.HitrateCallback(input_key="scores", target_key="targets", topk_args=(1, 3, 5)),
        # dl.MRRCallback(input_key="scores", target_key="targets", topk_args=(1, 3, 5)),
        # dl.MAPCallback(input_key="scores", target_key="targets", topk_args=(1, 3, 5)),
        # dl.NDCGCallback(input_key="scores", target_key="targets", topk_args=(1, 3, 5)),
        dl.OptimizerCallback(metric_key="loss"),
        dl.SchedulerCallback(),
        dl.CheckpointCallback(
            logdir="./logs", loader_key="valid", metric_key="loss", minimize=True
        ),
    ]
)

 

CV - MNIST classification

 

import os
from torch import nn, optim
from torch.utils.data import DataLoader
from catalyst import dl
from catalyst.contrib import MNIST

model = nn.Sequential(nn.Flatten(), nn.Linear(28 * 28, 10))
criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.parameters(), lr=0.02)

train_data = MNIST(os.getcwd(), train=True)
valid_data = MNIST(os.getcwd(), train=False)
loaders = {
    "train": DataLoader(train_data, batch_size=32),
    "valid": DataLoader(valid_data, batch_size=32),
}

runner = dl.SupervisedRunner()
# model training
runner.train(
    model=model,
    criterion=criterion,
    optimizer=optimizer,
    loaders=loaders,
    num_epochs=1,
    logdir="./logs",
    valid_loader="valid",
    valid_metric="loss",
    minimize_valid_metric=True,
    verbose=True,
# uncomment for extra metrics:
#     callbacks=[
#         dl.AccuracyCallback(input_key="logits", target_key="targets", num_classes=10),
#         dl.PrecisionRecallF1SupportCallback(
#             input_key="logits", target_key="targets", num_classes=10
#         ),
#         dl.AUCCallback(input_key="logits", target_key="targets"),
#         # catalyst[ml] required ``pip install catalyst[ml]``
#         dl.ConfusionMatrixCallback(
#             input_key="logits", target_key="targets", num_classes=num_classes
#         ),
#     ]
)

 

CV - MNIST segmentation

 

import os
import torch
from torch import nn
from torch.utils.data import DataLoader
from catalyst import dl
from catalyst.contrib import IoULoss, MNIST


model = nn.Sequential(
    nn.Conv2d(1, 1, 3, 1, 1), nn.ReLU(),
    nn.Conv2d(1, 1, 3, 1, 1), nn.Sigmoid(),
)
criterion = IoULoss()
optimizer = torch.optim.Adam(model.parameters(), lr=0.02)

train_data = MNIST(os.getcwd(), train=True)
valid_data = MNIST(os.getcwd(), train=False)
loaders = {
    "train": DataLoader(train_data, batch_size=32),
    "valid": DataLoader(valid_data, batch_size=32),
}

class CustomRunner(dl.SupervisedRunner):
    def handle_batch(self, batch):
        x = batch[self._input_key]
        x_noise = (x + torch.rand_like(x)).clamp_(0, 1)
        x_ = self.model(x_noise)
        self.batch = {self._input_key: x, self._output_key: x_, self._target_key: x}

runner = CustomRunner(
    input_key="features", output_key="scores", target_key="targets", loss_key="loss"
)
# model training
runner.train(
    model=model,
    criterion=criterion,
    optimizer=optimizer,
    loaders=loaders,
    num_epochs=1,
    callbacks=[
        dl.IOUCallback(input_key="scores", target_key="targets"),
        dl.DiceCallback(input_key="scores", target_key="targets"),
        dl.TrevskyCallback(input_key="scores", target_key="targets", alpha=0.2),
    ],
    logdir="./logdir",
    valid_loader="valid",
    valid_metric="loss",
    minimize_valid_metric=True,
    verbose=True,
)

 

CV - MNIST model distillation

 

import os
import torch
from torch import nn, optim
from torch.nn import functional as F
from torch.utils.data import DataLoader
from catalyst import dl
from catalyst.contrib import MNIST

# [!] teacher model should be already pretrained
teacher = nn.Sequential(nn.Flatten(), nn.Linear(28 * 28, 10))
student = nn.Sequential(nn.Flatten(), nn.Linear(28 * 28, 10))
criterion = {"cls": nn.CrossEntropyLoss(), "kl": nn.KLDivLoss(reduction="batchmean")}
optimizer = optim.Adam(student.parameters(), lr=0.02)

train_data = MNIST(os.getcwd(), train=True)
valid_data = MNIST(os.getcwd(), train=False)
loaders = {
    "train": DataLoader(train_data, batch_size=32),
    "valid": DataLoader(valid_data, batch_size=32),
}

class DistilRunner(dl.Runner):
    def handle_batch(self, batch):
        x, y = batch

        self.model["teacher"].eval()  # let's manually set teacher model to eval mode
        with torch.no_grad():
            t_logits = self.model["teacher"](x)

        s_logits = self.model["student"](x)
        self.batch = {
            "t_logits": t_logits, "s_logits": s_logits, "targets": y,
            "s_logprobs": F.log_softmax(s_logits, dim=-1), "t_probs": F.softmax(t_logits, dim=-1)
        }

runner = DistilRunner()
callbacks = [
    dl.AccuracyCallback(input_key="t_logits", target_key="targets", num_classes=2, prefix="teacher_"),
    dl.AccuracyCallback(input_key="s_logits", target_key="targets", num_classes=2, prefix="student_"),
    dl.CriterionCallback(input_key="s_logits", target_key="targets", metric_key="cls_loss", criterion_key="cls"),
    dl.CriterionCallback(input_key="s_logprobs", target_key="t_probs", metric_key="kl_div_loss", criterion_key="kl"),
    dl.MetricAggregationCallback(metric_key="loss", metrics=["kl_div_loss", "cls_loss"], mode="mean"),
    dl.OptimizerCallback(metric_key="loss", model_key="student"),
    dl.CheckpointCallback(logdir="./logs", loader_key="valid", metric_key="loss", minimize=True, save_n_best=3),
]
# model training
runner.train(
    model={"teacher": teacher, "student": student},
    criterion=criterion,
    optimizer=optimizer,
    loaders=loaders,
    num_epochs=1,
    logdir="./logs",
    verbose=True,
    callbacks=callbacks,
)

 

CV - MNIST metric learning

 

import os
from torch.optim import Adam
from torch.utils.data import DataLoader
from catalyst import dl
from catalyst.data import BatchBalanceClassSampler
from catalyst.contrib import data, datasets, models, nn


# 1. train and valid loaders
train_dataset = datasets.MnistMLDataset(root=os.getcwd())
sampler = BatchBalanceClassSampler(
    labels=train_dataset.get_labels(), num_classes=5, num_samples=10, num_batches=10
)
train_loader = DataLoader(dataset=train_dataset, batch_sampler=sampler)

valid_dataset = datasets.MnistQGDataset(root=os.getcwd(), gallery_fraq=0.2)
valid_loader = DataLoader(dataset=valid_dataset, batch_size=1024)

# 2. model and optimizer
model = models.MnistSimpleNet(out_features=16)
optimizer = Adam(model.parameters(), lr=0.001)

# 3. criterion with triplets sampling
sampler_inbatch = data.HardTripletsSampler(norm_required=False)
criterion = nn.TripletMarginLossWithSampler(margin=0.5, sampler_inbatch=sampler_inbatch)

# 4. training with catalyst Runner
class CustomRunner(dl.SupervisedRunner):
    def handle_batch(self, batch) -> None:
        if self.is_train_loader:
            images, targets = batch["features"].float(), batch["targets"].long()
            features = self.model(images)
            self.batch = {"embeddings": features, "targets": targets,}
        else:
            images, targets, is_query = \
                batch["features"].float(), batch["targets"].long(), batch["is_query"].bool()
            features = self.model(images)
            self.batch = {"embeddings": features, "targets": targets, "is_query": is_query}

callbacks = [
    dl.ControlFlowCallback(
        dl.CriterionCallback(input_key="embeddings", target_key="targets", metric_key="loss"),
        loaders="train",
    ),
    dl.ControlFlowCallback(
        dl.CMCScoreCallback(
            embeddings_key="embeddings",
            labels_key="targets",
            is_query_key="is_query",
            topk_args=[1],
        ),
        loaders="valid",
    ),
    dl.PeriodicLoaderCallback(
        valid_loader_key="valid", valid_metric_key="cmc01", minimize=False, valid=2
    ),
]

runner = CustomRunner(input_key="features", output_key="embeddings")
runner.train(
    model=model,
    criterion=criterion,
    optimizer=optimizer,
    callbacks=callbacks,
    loaders={"train": train_loader, "valid": valid_loader},
    verbose=False,
    logdir="./logs",
    valid_loader="valid",
    valid_metric="cmc01",
    minimize_valid_metric=False,
    num_epochs=10,
)

 

CV - MNIST GAN

 

import os
import torch
from torch import nn
from torch.utils.data import DataLoader
from catalyst import dl
from catalyst.contrib import Flatten, GlobalMaxPool2d, Lambda, MNIST

latent_dim = 128
generator = nn.Sequential(
    # We want to generate 128 coefficients to reshape into a 7x7x128 map
    nn.Linear(128, 128 * 7 * 7),
    nn.LeakyReLU(0.2, inplace=True),
    Lambda(lambda x: x.view(x.size(0), 128, 7, 7)),
    nn.ConvTranspose2d(128, 128, (4, 4), stride=(2, 2), padding=1),
    nn.LeakyReLU(0.2, inplace=True),
    nn.ConvTranspose2d(128, 128, (4, 4), stride=(2, 2), padding=1),
    nn.LeakyReLU(0.2, inplace=True),
    nn.Conv2d(128, 1, (7, 7), padding=3),
    nn.Sigmoid(),
)
discriminator = nn.Sequential(
    nn.Conv2d(1, 64, (3, 3), stride=(2, 2), padding=1),
    nn.LeakyReLU(0.2, inplace=True),
    nn.Conv2d(64, 128, (3, 3), stride=(2, 2), padding=1),
    nn.LeakyReLU(0.2, inplace=True),
    GlobalMaxPool2d(),
    Flatten(),
    nn.Linear(128, 1),
)

model = {"generator": generator, "discriminator": discriminator}
criterion = {"generator": nn.BCEWithLogitsLoss(), "discriminator": nn.BCEWithLogitsLoss()}
optimizer = {
    "generator": torch.optim.Adam(generator.parameters(), lr=0.0003, betas=(0.5, 0.999)),
    "discriminator": torch.optim.Adam(discriminator.parameters(), lr=0.0003, betas=(0.5, 0.999)),
}
train_data = MNIST(os.getcwd(), train=False)
loaders = {"train": DataLoader(train_data, batch_size=32)}

class CustomRunner(dl.Runner):
    def predict_batch(self, batch):
        batch_size = 1
        # Sample random points in the latent space
        random_latent_vectors = torch.randn(batch_size, latent_dim).to(self.device)
        # Decode them to fake images
        generated_images = self.model["generator"](random_latent_vectors).detach()
        return generated_images

    def handle_batch(self, batch):
        real_images, _ = batch
        batch_size = real_images.shape[0]

        # Sample random points in the latent space
        random_latent_vectors = torch.randn(batch_size, latent_dim).to(self.device)

        # Decode them to fake images
        generated_images = self.model["generator"](random_latent_vectors).detach()
        # Combine them with real images
        combined_images = torch.cat([generated_images, real_images])

        # Assemble labels discriminating real from fake images
        labels = \
            torch.cat([torch.ones((batch_size, 1)), torch.zeros((batch_size, 1))]).to(self.device)
        # Add random noise to the labels - important trick!
        labels += 0.05 * torch.rand(labels.shape).to(self.device)

        # Discriminator forward
        combined_predictions = self.model["discriminator"](combined_images)

        # Sample random points in the latent space
        random_latent_vectors = torch.randn(batch_size, latent_dim).to(self.device)
        # Assemble labels that say "all real images"
        misleading_labels = torch.zeros((batch_size, 1)).to(self.device)

        # Generator forward
        generated_images = self.model["generator"](random_latent_vectors)
        generated_predictions = self.model["discriminator"](generated_images)

        self.batch = {
            "combined_predictions": combined_predictions,
            "labels": labels,
            "generated_predictions": generated_predictions,
            "misleading_labels": misleading_labels,
        }


runner = CustomRunner()
runner.train(
    model=model,
    criterion=criterion,
    optimizer=optimizer,
    loaders=loaders,
    callbacks=[
        dl.CriterionCallback(
            input_key="combined_predictions",
            target_key="labels",
            metric_key="loss_discriminator",
            criterion_key="discriminator",
        ),
        dl.CriterionCallback(
            input_key="generated_predictions",
            target_key="misleading_labels",
            metric_key="loss_generator",
            criterion_key="generator",
        ),
        dl.OptimizerCallback(
            model_key="generator",
            optimizer_key="generator",
            metric_key="loss_generator"
        ),
        dl.OptimizerCallback(
            model_key="discriminator",
            optimizer_key="discriminator",
            metric_key="loss_discriminator"
        ),
    ],
    valid_loader="train",
    valid_metric="loss_generator",
    minimize_valid_metric=True,
    num_epochs=20,
    verbose=True,
    logdir="./logs_gan",
)

# visualization (matplotlib required):
# import matplotlib.pyplot as plt
# %matplotlib inline
# plt.imshow(runner.predict_batch(None)[0, 0].cpu().numpy())

 

CV - MNIST VAE

 

import os
import torch
from torch import nn, optim
from torch.nn import functional as F
from torch.utils.data import DataLoader
from catalyst import dl, metrics
from catalyst.contrib import MNIST

LOG_SCALE_MAX = 2
LOG_SCALE_MIN = -10

def normal_sample(loc, log_scale):
    scale = torch.exp(0.5 * log_scale)
    return loc + scale * torch.randn_like(scale)

class VAE(nn.Module):
    def __init__(self, in_features, hid_features):
        super().__init__()
        self.hid_features = hid_features
        self.encoder = nn.Linear(in_features, hid_features * 2)
        self.decoder = nn.Sequential(nn.Linear(hid_features, in_features), nn.Sigmoid())

    def forward(self, x, deterministic=False):
        z = self.encoder(x)
        bs, z_dim = z.shape

        loc, log_scale = z[:, : z_dim // 2], z[:, z_dim // 2 :]
        log_scale = torch.clamp(log_scale, LOG_SCALE_MIN, LOG_SCALE_MAX)

        z_ = loc if deterministic else normal_sample(loc, log_scale)
        z_ = z_.view(bs, -1)
        x_ = self.decoder(z_)

        return x_, loc, log_scale

class CustomRunner(dl.IRunner):
    def __init__(self, logdir, device):
        super().__init__()
        self._logdir = logdir
        self._device = device

    def get_engine(self):
        return dl.DeviceEngine(self._device)

    def get_loggers(self):
        return {
            "console": dl.ConsoleLogger(),
            "csv": dl.CSVLogger(logdir=self._logdir),
            "tensorboard": dl.TensorboardLogger(logdir=self._logdir),
        }

    @property
    def stages(self):
        return ["train"]

    def get_stage_len(self, stage: str) -> int:
        return 3

    def get_loaders(self, stage: str):
        train_data = MNIST(os.getcwd(), train=True)
        valid_data = MNIST(os.getcwd(), train=False)
        loaders = {
            "train": DataLoader(train_data, batch_size=32),
            "valid": DataLoader(valid_data, batch_size=32),
        }
        return loaders

    def get_model(self, stage: str):
        model = self.model if self.model is not None else VAE(28 * 28, 64)
        return model

    def get_optimizer(self, stage: str, model):
        return optim.Adam(model.parameters(), lr=0.02)

    def get_callbacks(self, stage: str):
        return {
            "optimizer": dl.OptimizerCallback(metric_key="loss"),
            "checkpoint": dl.CheckpointCallback(
                self._logdir, loader_key="valid", metric_key="loss", minimize=True
            ),
        }

    def on_loader_start(self, runner):
        super().on_loader_start(runner)
        self.meters = {
            key: metrics.AdditiveMetric(compute_on_call=False)
            for key in ["loss_ae", "loss_kld", "loss"]
        }

    def handle_batch(self, batch):
        x, _ = batch
        x = x.view(x.size(0), -1)
        x_, loc, log_scale = self.model(x, deterministic=not self.is_train_loader)

        loss_ae = F.mse_loss(x_, x)
        loss_kld = (-0.5 * torch.sum(1 + log_scale - loc.pow(2) - log_scale.exp(), dim=1)).mean()
        loss = loss_ae + loss_kld * 0.01

        self.batch_metrics = {"loss_ae": loss_ae, "loss_kld": loss_kld, "loss": loss}
        for key in ["loss_ae", "loss_kld", "loss"]:
            self.meters[key].update(self.batch_metrics[key].item(), self.batch_size)

    def on_loader_end(self, runner):
        for key in ["loss_ae", "loss_kld", "loss"]:
            self.loader_metrics[key] = self.meters[key].compute()[0]
        super().on_loader_end(runner)

    def predict_batch(self, batch):
        random_latent_vectors = torch.randn(1, self.model.hid_features).to(self.device)
        generated_images = self.model.decoder(random_latent_vectors).detach()
        return generated_images

runner = CustomRunner("./logs", "cpu")
runner.run()
# visualization (matplotlib required):
# import matplotlib.pyplot as plt
# %matplotlib inline
# plt.imshow(runner.predict_batch(None)[0].cpu().numpy().reshape(28, 28))

 

CV - MNIST multistage finetuning

 

import os
from torch import nn, optim
from torch.utils.data import DataLoader
from catalyst import dl, utils
from catalyst.contrib import MNIST


class CustomRunner(dl.IRunner):
    def __init__(self, logdir, device):
        super().__init__()
        self._logdir = logdir
        self._device = device

    def get_engine(self):
        return dl.DeviceEngine(self._device)

    def get_loggers(self):
        return {
            "console": dl.ConsoleLogger(),
            "csv": dl.CSVLogger(logdir=self._logdir),
            "tensorboard": dl.TensorboardLogger(logdir=self._logdir),
        }

    @property
    def stages(self):
        return ["train_freezed", "train_unfreezed"]

    def get_stage_len(self, stage: str) -> int:
        return 3

    def get_loaders(self, stage: str):
        train_data = MNIST(os.getcwd(), train=True)
        valid_data = MNIST(os.getcwd(), train=False)
        loaders = {
            "train": DataLoader(train_data, batch_size=32),
            "valid": DataLoader(valid_data, batch_size=32),
        }
        return loaders

    def get_model(self, stage: str):
        model = (
            self.model
            if self.model is not None
            else nn.Sequential(nn.Flatten(), nn.Linear(784, 128), nn.ReLU(), nn.Linear(128, 10))
        )
        if stage == "train_freezed":
            # freeze layer
            utils.set_requires_grad(model[1], False)
        else:
            utils.set_requires_grad(model, True)
        return model

    def get_criterion(self, stage: str):
        return nn.CrossEntropyLoss()

    def get_optimizer(self, stage: str, model):
        if stage == "train_freezed":
            return optim.Adam(model.parameters(), lr=1e-3)
        else:
            return optim.SGD(model.parameters(), lr=1e-1)

    def get_scheduler(self, stage: str, optimizer):
        return None

    def get_callbacks(self, stage: str):
        return {
            "criterion": dl.CriterionCallback(
                metric_key="loss", input_key="logits", target_key="targets"
            ),
            "optimizer": dl.OptimizerCallback(metric_key="loss"),
            # "scheduler": dl.SchedulerCallback(loader_key="valid", metric_key="loss"),
            # "accuracy": dl.AccuracyCallback(
            #     input_key="logits", target_key="targets", topk_args=(1, 3, 5)
            # ),
            # "classification": dl.PrecisionRecallF1SupportCallback(
            #     input_key="logits", target_key="targets", num_classes=10
            # ),
            # "confusion_matrix": dl.ConfusionMatrixCallback(
            #     input_key="logits", target_key="targets", num_classes=10
            # ),
            "checkpoint": dl.CheckpointCallback(
                self._logdir, loader_key="valid", metric_key="loss", minimize=True, save_n_best=3
            ),
        }

    def handle_batch(self, batch):
        x, y = batch
        logits = self.model(x)

        self.batch = {
            "features": x,
            "targets": y,
            "logits": logits,
        }

runner = CustomRunner("./logs", "cpu")
runner.run()

 

CV - MNIST multistage finetuning (distributed)

 

import os
from torch import nn, optim
from torch.utils.data import DataLoader
from catalyst import dl, utils
from catalyst.contrib import MNIST


class CustomRunner(dl.IRunner):
    def __init__(self, logdir):
        super().__init__()
        self._logdir = logdir

    def get_engine(self):
        return dl.DistributedDataParallelEngine()

    def get_loggers(self):
        return {
            "console": dl.ConsoleLogger(),
            "csv": dl.CSVLogger(logdir=self._logdir),
            "tensorboard": dl.TensorboardLogger(logdir=self._logdir),
        }

    @property
    def stages(self):
        return ["train_freezed", "train_unfreezed"]

    def get_stage_len(self, stage: str) -> int:
        return 3

    def get_loaders(self, stage: str):
        train_data = MNIST(os.getcwd(), train=True)
        valid_data = MNIST(os.getcwd(), train=False)
        loaders = {
            "train": DataLoader(train_data, batch_size=32),
            "valid": DataLoader(valid_data, batch_size=32),
        }
        return loaders

    def get_model(self, stage: str):
        model = nn.Sequential(nn.Flatten(), nn.Linear(784, 128), nn.ReLU(), nn.Linear(128, 10))
        if stage == "train_freezed":  # freeze layer
            utils.set_requires_grad(model[1], False)
        else:
            utils.set_requires_grad(model, True)
        return model

    def get_criterion(self, stage: str):
        return nn.CrossEntropyLoss()

    def get_optimizer(self, stage: str, model):
        if stage == "train_freezed":
            return optim.Adam(model.parameters(), lr=1e-3)
        else:
            return optim.SGD(model.parameters(), lr=1e-1)

    def get_callbacks(self, stage: str):
        return {
            "criterion": dl.CriterionCallback(
                metric_key="loss", input_key="logits", target_key="targets"
            ),
            "optimizer": dl.OptimizerCallback(metric_key="loss"),
            "accuracy": dl.AccuracyCallback(
                input_key="logits", target_key="targets", topk_args=(1, 3, 5)
            ),
            "classification": dl.PrecisionRecallF1SupportCallback(
                input_key="logits", target_key="targets", num_classes=10
            ),
            # catalyst[ml] required ``pip install catalyst[ml]``
            # "confusion_matrix": dl.ConfusionMatrixCallback(
            #     input_key="logits", target_key="targets", num_classes=10
            # ),
            "checkpoint": dl.CheckpointCallback(
                self._logdir,
                loader_key="valid",
                metric_key="loss",
                minimize=True,
                save_n_best=3,
                # here is the main trick:
                load_on_stage_start={
                    "model": "best",
                    "global_epoch_step": "last",
                    "global_batch_step": "last",
                    "global_sample_step": "last",
                },
            ),
            "verbose": dl.TqdmCallback(),
        }

    def handle_batch(self, batch):
        x, y = batch
        logits = self.model(x)

        self.batch = {
            "features": x,
            "targets": y,
            "logits": logits,
        }


if __name__ == "__main__":
    runner = CustomRunner("./logs")
    runner.run()

 

AutoML - hyperparameters optimization with Optuna

 

import os
import optuna
import torch
from torch import nn
from torch.utils.data import DataLoader
from catalyst import dl
from catalyst.contrib import MNIST


def objective(trial):
    lr = trial.suggest_loguniform("lr", 1e-3, 1e-1)
    num_hidden = int(trial.suggest_loguniform("num_hidden", 32, 128))

    train_data = MNIST(os.getcwd(), train=True)
    valid_data = MNIST(os.getcwd(), train=False)
    loaders = {
        "train": DataLoader(train_data, batch_size=32),
        "valid": DataLoader(valid_data, batch_size=32),
    }
    model = nn.Sequential(
        nn.Flatten(), nn.Linear(784, num_hidden), nn.ReLU(), nn.Linear(num_hidden, 10)
    )
    optimizer = torch.optim.Adam(model.parameters(), lr=lr)
    criterion = nn.CrossEntropyLoss()

    runner = dl.SupervisedRunner(input_key="features", output_key="logits", target_key="targets")
    runner.train(
        model=model,
        criterion=criterion,
        optimizer=optimizer,
        loaders=loaders,
        callbacks={
            "accuracy": dl.AccuracyCallback(
                input_key="logits", target_key="targets", num_classes=10
            ),
            # catalyst[optuna] required ``pip install catalyst[optuna]``
            "optuna": dl.OptunaPruningCallback(
                loader_key="valid", metric_key="accuracy01", minimize=False, trial=trial
            ),
        },
        num_epochs=3,
    )
    score = trial.best_score
    return score

study = optuna.create_study(
    direction="maximize",
    pruner=optuna.pruners.MedianPruner(
        n_startup_trials=1, n_warmup_steps=0, interval_steps=1
    ),
)
study.optimize(objective, n_trials=3, timeout=300)
print(study.best_value, study.best_params)

 

Config API - minimal example

 

import torch
from torch.utils.data import TensorDataset
from catalyst import dl

NUM_SAMPLES, NUM_FEATURES, NUM_CLASSES = int(1e4), int(1e1), 4
LOGDIR = "./logs"

class CustomConfigRunner(dl.SupervisedConfigRunner):
    def get_datasets(self, stage: str):
        # sample data
        X = torch.rand(NUM_SAMPLES, NUM_FEATURES)
        y = (torch.rand(NUM_SAMPLES,) * NUM_CLASSES).to(torch.int64)

        # pytorch dataset
        dataset = TensorDataset(X, y)
        datasets = {"train": dataset, "valid": dataset}
        return datasets


runner = CustomConfigRunner(
    input_key="features",
    output_key="logits",
    target_key="targets",
    loss_key="loss",
    config={
        "args": {
            "logdir": LOGDIR,
            "valid_loader": "valid",
            "valid_metric": "accuracy01",
            "minimize_valid_metric": False,
            "verbose": False,
        },
        "model": {
            "_target_": "Linear",
            "in_features": NUM_FEATURES,
            "out_features": NUM_CLASSES,
        },
        "engine": {"_target_": "DeviceEngine"},
        "loggers": {
            "console": {"_target_": "ConsoleLogger"},
            "csv": {"_target_": "CSVLogger", "logdir": LOGDIR},
            "tensorboard": {"_target_": "TensorboardLogger", "logdir": LOGDIR},
        },
        "stages": {
            "stage1": {
                "num_epochs": 10,
                "criterion": {"_target_": "CrossEntropyLoss"},
                "optimizer": {"_target_": "Adam", "lr": 1e-3},
                "scheduler": {"_target_": "MultiStepLR", "milestones": [2]},
                "loaders": {"batch_size": 32, "num_workers": 1},
                "callbacks": {
                    "accuracy": {
                        "_target_": "AccuracyCallback",
                        "input_key": "logits",
                        "target_key": "targets",
                        "num_classes": NUM_CLASSES,
                    },
                    "classification": {
                        "_target_": "PrecisionRecallF1SupportCallback",
                        "input_key": "logits",
                        "target_key": "targets",
                        "num_classes": NUM_CLASSES,
                    },
                    "criterion": {
                        "_target_": "CriterionCallback",
                        "input_key": "logits",
                        "target_key": "targets",
                        "metric_key": "loss",
                    },
                    "optimizer": {"_target_": "OptimizerCallback", "metric_key": "loss"},
                    "scheduler": {"_target_": "SchedulerCallback"},
                    "checkpointer": {
                        "_target_": "CheckpointCallback",
                        "logdir": LOGDIR,
                        "loader_key": "valid",
                        "metric_key": "accuracy01",
                        "minimize": False,
                        "save_n_best": 3,
                    },
                },
            },
        },
    },
)
runner.run()

 

• More Catalyst examples.
• Config/Hydra API – full example.

Blog Posts

Talks

Community

Accelerated with Catalyst

Research Papers

 

• Hierarchical Attention for Sentiment Classification with Visualization
• Pediatric Bone Age Assessment
• Implementation of the paper "Tell Me Where to Look: Guided Attention Inference Network"
• Implementation of the paper "Filter Response Normalization Layer: Eliminating Batch Dependence in the Training of Deep Neural Networks"
• Implementation of the paper "Utterance-level Aggregation For Speaker Recognition In The Wild"
• Implementation of the paper "Looking to Listen at the Cocktail Party: A Speaker-Independent Audio-Visual Model for Speech Separation"
• Implementation of the paper "ESRGAN: Enhanced Super-Resolution Generative Adversarial Networks"

 

Blog Posts

 

• Solving the Cocktail Party Problem using PyTorch
• Beyond fashion: Deep Learning with Catalyst (Config API)
• Tutorial from Notebook API to Config API (RU)

 

Competitions

 

• Kaggle Quick, Draw! Doodle Recognition Challenge - 11th place
• Catalyst.RL - NeurIPS 2018: AI for Prosthetics Challenge – 3rd place
• Kaggle Google Landmark 2019 - 30th place
• iMet Collection 2019 - FGVC6 - 24th place
• ID R&D Anti-spoofing Challenge - 14th place
• NeurIPS 2019: Recursion Cellular Image Classification - 4th place
• MICCAI 2019: Automatic Structure Segmentation for Radiotherapy Planning Challenge 2019
  • 3rd place solution for Task 3: Organ-at-risk segmentation from chest CT scans
  • and 4th place solution for Task 4: Gross Target Volume segmentation of lung cancer
• Kaggle Seversteal steel detection - 5th place
• RSNA Intracranial Hemorrhage Detection - 5th place
• APTOS 2019 Blindness Detection – 7th place
• Catalyst.RL - NeurIPS 2019: Learn to Move - Walk Around – 2nd place
• xView2 Damage Assessment Challenge - 3rd place

 

Toolkits

 

• Catalyst.RL – A Distributed Framework for Reproducible RL Research by Scitator
• Catalyst.Classification - Comprehensive classification pipeline with Pseudo-Labeling by Bagxi and Pdanilov
• Catalyst.Segmentation - Segmentation pipelines - binary, semantic and instance, by Bagxi
• Catalyst.Detection - Anchor-free detection pipeline by Avi2011class and TezRomacH
• Catalyst.GAN - Reproducible GANs pipelines by Asmekal
• Catalyst.Neuro - Brain image analysis project, in collaboration with TReNDS Center
• MLComp – Distributed DAG framework for machine learning with UI by Lightforever
• Pytorch toolbelt - PyTorch extensions for fast R&D prototyping and Kaggle farming by BloodAxe
• Helper functions - An assorted collection of helper functions by Ternaus
• BERT Distillation with Catalyst by elephantmipt

 

Other

 

• CamVid Segmentation Example - Example of semantic segmentation for CamVid dataset
• Notebook API tutorial for segmentation in Understanding Clouds from Satellite Images Competition
• Catalyst.RL - NeurIPS 2019: Learn to Move - Walk Around – starter kit
• Catalyst.RL - NeurIPS 2019: Animal-AI Olympics - starter kit
• Inria Segmentation Example - An example of training segmentation model for Inria Sattelite Segmentation Challenge
• iglovikov_segmentation - Semantic segmentation pipeline using Catalyst
• Logging Catalyst Runs to Comet - An example of how to log metrics, hyperparameters and more from Catalyst runs to Comet

 

See other projects at the GitHub dependency graph.

If your project implements a paper, a notable use-case/tutorial, or a Kaggle competition solution, or if your code simply presents interesting results and uses Catalyst, we would be happy to add your project to the list above! Do not hesitate to send us a PR with a brief description of the project similar to the above.

Contribution Guide

We appreciate all contributions. If you are planning to contribute back bug-fixes, there is no need to run that by us; just send a PR. If you plan to contribute new features, new utility functions, or extensions, please open an issue first and discuss it with us.

• Please see the Contribution Guide for more information.
• By participating in this project, you agree to abide by its Code of Conduct.

User Feedback

We've created feedback@catalyst-team.com as an additional channel for user feedback.

• If you like the project and want to thank us, this is the right place.
• If you would like to start a collaboration between your team and Catalyst team to improve Deep Learning R&D, you are always welcome.
• If you don't like Github Issues and prefer email, feel free to email us.
• Finally, if you do not like something, please, share it with us, and we can see how to improve it.

We appreciate any type of feedback. Thank you!

Acknowledgments

Since the beginning of the Сatalyst development, a lot of people have influenced it in a lot of different ways.

Catalyst.Team

• Dmytro Doroshenko (ditwoo)
• Eugene Kachan (bagxi)
• Nikita Balagansky (elephantmipt)
• Sergey Kolesnikov (scitator)

Catalyst.Contributors

• Aleksey Grinchuk (alexgrinch)
• Aleksey Shabanov (AlekseySh)
• Alex Gaziev (gazay)
• Andrey Zharkov (asmekal)
• Artem Zolkin (arquestro)
• David Kuryakin (dkuryakin)
• Evgeny Semyonov (lightforever)
• Eugene Khvedchenya (bloodaxe)
• Ivan Stepanenko
• Julia Shenshina (julia-shenshina)
• Nguyen Xuan Bac (ngxbac)
• Roman Tezikov (TezRomacH)
• Valentin Khrulkov (khrulkovv)
• Vladimir Iglovikov (ternaus)
• Vsevolod Poletaev (hexfaker)
• Yury Kashnitsky (yorko)

Trusted by

• Awecom
• Researchers at the Center for Translational Research in Neuroimaging and Data Science (TReNDS)
• Deep Learning School
• Researchers at Emory University
• Evil Martians
• Researchers at the Georgia Institute of Technology
• Researchers at Georgia State University
• Helios
• HPCD Lab
• iFarm
• Kinoplan
• Researchers at the Moscow Institute of Physics and Technology
• Neuromation
• Poteha Labs
• Provectus
• Researchers at the Skolkovo Institute of Science and Technology
• SoftConstruct
• Researchers at Tinkoff
• Researchers at Yandex.Research

Citation

Please use this bibtex if you want to cite this repository in your publications:

@misc{catalyst,
    author = {Kolesnikov, Sergey},
    title = {Catalyst - Accelerated deep learning R&D},
    year = {2018},
    publisher = {GitHub},
    journal = {GitHub repository},
    howpublished = {\url{https://github.com/catalyst-team/catalyst}},
}

---

Author: catalyst-team
Source Code: https://github.com/catalyst-team/catalyst
License: Apache-2.0 License

#python #deep-learning #pytorch 

What is GEEK

Buddha Community

CSharp REPL: A Command Line C# REPL with Syntax Highlighting

C# REPL

A cross-platform command line REPL for the rapid experimentation and exploration of C#. It supports intellisense, installing NuGet packages, and referencing local .NET projects and assemblies.

 

(click to view animation)

C# REPL provides the following features:

• Syntax highlighting via ANSI escape sequences
• Intellisense with fly-out documentation
• Nuget package installation
• Reference local assemblies, solutions, and projects
• Navigate to source via Source Link
• IL disassembly (both Debug and Release mode)
• Fast and flicker-free rendering. A "diff" algorithm is used to only render what's changed.

Installation

C# REPL is a .NET 6 global tool, and runs on Windows 10, Mac OS, and Linux. It can be installed via:

dotnet tool install -g csharprepl

If you're running on Mac OS Catalina (10.15) or later, make sure you follow any additional directions printed to the screen. You may need to update your PATH variable in order to use .NET global tools.

After installation is complete, run csharprepl to begin. C# REPL can be updated via dotnet tool update -g csharprepl.

Usage:

Run csharprepl from the command line to begin an interactive session. The default colorscheme uses the color palette defined by your terminal, but these colors can be changed using a theme.json file provided as a command line argument.

Evaluating Code

Type some C# into the prompt and press Enter to run it. The result, if any, will be printed:

> Console.WriteLine("Hello World")
Hello World

> DateTime.Now.AddDays(8)
[6/7/2021 5:13:00 PM]

To evaluate multiple lines of code, use Shift+Enter to insert a newline:

> var x = 5;
  var y = 8;
  x * y
40

Additionally, if the statement is not a "complete statement" a newline will automatically be inserted when Enter is pressed. For example, in the below code, the first line is not a syntactically complete statement, so when we press enter we'll go down to a new line:

> if (x == 5)
  | // caret position, after we press Enter on Line 1

Finally, pressing Ctrl+Enter will show a "detailed view" of the result. For example, for the DateTime.Now expression below, on the first line we pressed Enter, and on the second line we pressed Ctrl+Enter to view more detailed output:

> DateTime.Now // Pressing Enter shows a reasonable representation
[5/30/2021 5:13:00 PM]

> DateTime.Now // Pressing Ctrl+Enter shows a detailed representation
[5/30/2021 5:13:00 PM] {
  Date: [5/30/2021 12:00:00 AM],
  Day: 30,
  DayOfWeek: Sunday,
  DayOfYear: 150,
  Hour: 17,
  InternalKind: 9223372036854775808,
  InternalTicks: 637579915804530992,
  Kind: Local,
  Millisecond: 453,
  Minute: 13,
  Month: 5,
  Second: 0,
  Ticks: 637579915804530992,
  TimeOfDay: [17:13:00.4530992],
  Year: 2021,
  _dateData: 9860951952659306800
}

A note on semicolons: C# expressions do not require semicolons, but statements do. If a statement is missing a required semicolon, a newline will be added instead of trying to run the syntatically incomplete statement; simply type the semicolon to complete the statement.

> var now = DateTime.Now; // assignment statement, semicolon required

> DateTime.Now.AddDays(8) // expression, we don't need a semicolon
[6/7/2021 5:03:05 PM]

Keyboard Shortcuts

• Basic Usage
  • Ctrl+C - Cancel current line
  • Ctrl+L - Clear screen
  • Enter - Evaluate the current line if it's a syntactically complete statement; otherwise add a newline
  • Control+Enter - Evaluate the current line, and return a more detailed representation of the result
  • Shift+Enter - Insert a new line (this does not currently work on Linux or Mac OS; Hopefully this will work in .NET 7)
  • Ctrl+Shift+C - Copy current line to clipboard
  • Ctrl+V, Shift+Insert, and Ctrl+Shift+V - Paste text to prompt. Automatically trims leading indent
• Code Actions
  • F1 - Opens the MSDN documentation for the class/method under the caret (example)
  • F9 - Shows the IL (intermediate language) for the current statement in Debug mode.
  • Ctrl+F9 - Shows the IL for the current statement with Release mode optimizations.
  • F12 - Opens the source code in the browser for the class/method under the caret, if the assembly supports Source Link.
• Autocompletion
  • Ctrl+Space - Open autocomplete menu. If there's a single option, pressing Ctrl+Space again will select the option
  • Enter, Right Arrow, Tab - Select active autocompletion option
  • Escape - closes autocomplete menu
• Text Navigation
  • Home and End - Navigate to beginning of a single line and end of a single line, respectively
  • Ctrl+Home and Ctrl+End - Navigate to beginning of line and end across multiple lines in a multiline prompt, respectively
  • Arrows - Navigate characters within text
  • Ctrl+Arrows - Navigate words within text
  • Ctrl+Backspace - Delete previous word
  • Ctrl+Delete - Delete next word

Adding References

Use the #r command to add assembly or nuget references.

• For assembly references, run #r "AssemblyName" or #r "path/to/assembly.dll"
• For project references, run #r "path/to/project.csproj". Solution files (.sln) can also be referenced.
• For nuget references, run #r "nuget: PackageName" to install the latest version of a package, or #r "nuget: PackageName, 13.0.5" to install a specific version (13.0.5 in this case).

To run ASP.NET applications inside the REPL, start the csharprepl application with the --framework parameter, specifying the Microsoft.AspNetCore.App shared framework. Then, use the above #r command to reference the application DLL. See the Command Line Configuration section below for more details.

csharprepl --framework  Microsoft.AspNetCore.App

Command Line Configuration

The C# REPL supports multiple configuration flags to control startup, behavior, and appearance:

csharprepl [OPTIONS] [response-file.rsp] [script-file.csx] [-- <additional-arguments>]

Supported options are:

• OPTIONS:
  • -r <dll> or --reference <dll>: Reference an assembly, project file, or nuget package. Can be specified multiple times. Uses the same syntax as #r statements inside the REPL. For example, csharprepl -r "nuget:Newtonsoft.Json" "path/to/myproj.csproj"
    • When an assembly or project is referenced, assemblies in the containing directory will be added to the assembly search path. This means that you don't need to manually add references to all of your assembly's dependencies (e.g. other references and nuget packages). Referencing the main entry assembly is enough.
  • -u <namespace> or --using <namespace>: Add a using statement. Can be specified multiple times.
  • -f <framework> or --framework <framework>: Reference a shared framework. The available shared frameworks depends on the local .NET installation, and can be useful when running an ASP.NET application from the REPL. Example frameworks are:
    • Microsoft.NETCore.App (default)
    • Microsoft.AspNetCore.All
    • Microsoft.AspNetCore.App
    • Microsoft.WindowsDesktop.App
  • -t <theme.json> or --theme <theme.json>: Read a theme file for syntax highlighting. This theme file associates C# syntax classifications with colors. The color values can be full RGB, or ANSI color names (defined in your terminal's theme). The NO_COLOR standard is supported.
  • --trace: Produce a trace file in the current directory that logs CSharpRepl internals. Useful for CSharpRepl bug reports.
  • -v or --version: Show version number and exit.
  • -h or --help: Show help and exit.
• response-file.rsp: A filepath of an .rsp file, containing any of the above command line options.
• script-file.csx: A filepath of a .csx file, containing lines of C# to evaluate before starting the REPL. Arguments to this script can be passed as <additional-arguments>, after a double hyphen (--), and will be available in a global args variable.

If you have dotnet-suggest enabled, all options can be tab-completed, including values provided to --framework and .NET namespaces provided to --using.

Integrating with other software

C# REPL is a standalone software application, but it can be useful to integrate it with other developer tools:

Windows Terminal

To add C# REPL as a menu entry in Windows Terminal, add the following profile to Windows Terminal's settings.json configuration file (under the JSON property profiles.list):

{
    "name": "C# REPL",
    "commandline": "csharprepl"
},

To get the exact colors shown in the screenshots in this README, install the Windows Terminal Dracula theme.

Visual Studio Code

To use the C# REPL with Visual Studio Code, simply run the csharprepl command in the Visual Studio Code terminal. To send commands to the REPL, use the built-in Terminal: Run Selected Text In Active Terminal command from the Command Palette (workbench.action.terminal.runSelectedText).

Windows OS

To add the C# REPL to the Windows Start Menu for quick access, you can run the following PowerShell command, which will start C# REPL in Windows Terminal:

$shell = New-Object -ComObject WScript.Shell
$shortcut = $shell.CreateShortcut("$env:appdata\Microsoft\Windows\Start Menu\Programs\csharprepl.lnk")
$shortcut.TargetPath = "wt.exe"
$shortcut.Arguments = "-w 0 nt csharprepl.exe"
$shortcut.Save()

You may also wish to add a shorter alias for C# REPL, which can be done by creating a .cmd file somewhere on your path. For example, put the following contents in C:\Users\username\.dotnet\tools\csr.cmd:

wt -w 0 nt csharprepl

This will allow you to launch C# REPL by running csr from anywhere that accepts Windows commands, like the Window Run dialog.

Comparison with other REPLs

This project is far from being the first REPL for C#. Here are some other projects; if this project doesn't suit you, another one might!

Visual Studio's C# Interactive pane is full-featured (it has syntax highlighting and intellisense) and is part of Visual Studio. This deep integration with Visual Studio is both a benefit from a workflow perspective, and a drawback as it's not cross-platform. As far as I know, the C# Interactive pane does not support NuGet packages or navigating to documentation/source code. Subjectively, it does not follow typical command line keybindings, so can feel a bit foreign.

csi.exe ships with C# and is a command line REPL. It's great because it's a cross platform REPL that comes out of the box, but it doesn't support syntax highlighting or autocompletion.

dotnet script allows you to run C# scripts from the command line. It has a REPL built-in, but the predominant focus seems to be as a script runner. It's a great tool, though, and has a strong community following.

dotnet interactive is a tool from Microsoft that creates a Jupyter notebook for C#, runnable through Visual Studio Code. It also provides a general framework useful for running REPLs.

Download Details:
Author: waf
Source Code: https://github.com/waf/CSharpRepl
License: MPL-2.0 License

#dotnet  #aspdotnet  #csharp 

Top Deep Learning Development Services | Hire Deep Learning Developer

View more: https://www.inexture.com/services/deep-learning-development/

We at Inexture, strategically work on every project we are associated with. We propose a robust set of AI, ML, and DL consulting services. Our virtuoso team of data scientists and developers meticulously work on every project and add a personalized touch to it. Because we keep our clientele aware of everything being done associated with their project so there’s a sense of transparency being maintained. Leverage our services for your next AI project for end-to-end optimum services.

#deep learning development #deep learning framework #deep learning expert #deep learning ai #deep learning services

Top 10 Deep Learning Sessions To Look Forward To At DVDC 2020

The Deep Learning DevCon 2020, DLDC 2020, has exciting talks and sessions around the latest developments in the field of deep learning, that will not only be interesting for professionals of this field but also for the enthusiasts who are willing to make a career in the field of deep learning. The two-day conference scheduled for 29th and 30th October will host paper presentations, tech talks, workshops that will uncover some interesting developments as well as the latest research and advancement of this area. Further to this, with deep learning gaining massive traction, this conference will highlight some fascinating use cases across the world.

Here are ten interesting talks and sessions of DLDC 2020 that one should definitely attend:

Also Read: Why Deep Learning DevCon Comes At The Right Time

---

Adversarial Robustness in Deep Learning

By Dipanjan Sarkar

**About: **Adversarial Robustness in Deep Learning is a session presented by Dipanjan Sarkar, a Data Science Lead at Applied Materials, as well as a Google Developer Expert in Machine Learning. In this session, he will focus on the adversarial robustness in the field of deep learning, where he talks about its importance, different types of adversarial attacks, and will showcase some ways to train the neural networks with adversarial realisation. Considering abstract deep learning has brought us tremendous achievements in the fields of computer vision and natural language processing, this talk will be really interesting for people working in this area. With this session, the attendees will have a comprehensive understanding of adversarial perturbations in the field of deep learning and ways to deal with them with common recipes.

Read an interview with Dipanjan Sarkar.

Imbalance Handling with Combination of Deep Variational Autoencoder and NEATER

By Divye Singh

**About: **Imbalance Handling with Combination of Deep Variational Autoencoder and NEATER is a paper presentation by Divye Singh, who has a masters in technology degree in Mathematical Modeling and Simulation and has the interest to research in the field of artificial intelligence, learning-based systems, machine learning, etc. In this paper presentation, he will talk about the common problem of class imbalance in medical diagnosis and anomaly detection, and how the problem can be solved with a deep learning framework. The talk focuses on the paper, where he has proposed a synergistic over-sampling method generating informative synthetic minority class data by filtering the noise from the over-sampled examples. Further, he will also showcase the experimental results on several real-life imbalanced datasets to prove the effectiveness of the proposed method for binary classification problems.

Default Rate Prediction Models for Self-Employment in Korea using Ridge, Random Forest & Deep Neural Network

By Dongsuk Hong

About: This is a paper presentation given by Dongsuk Hong, who is a PhD in Computer Science, and works in the big data centre of Korea Credit Information Services. This talk will introduce the attendees with machine learning and deep learning models for predicting self-employment default rates using credit information. He will talk about the study, where the DNN model is implemented for two purposes — a sub-model for the selection of credit information variables; and works for cascading to the final model that predicts default rates. Hong’s main research area is data analysis of credit information, where she is particularly interested in evaluating the performance of prediction models based on machine learning and deep learning. This talk will be interesting for the deep learning practitioners who are willing to make a career in this field.

---

#opinions #attend dldc 2020 #deep learning #deep learning sessions #deep learning talks #dldc 2020 #top deep learning sessions at dldc 2020 #top deep learning talks at dldc 2020

Few Shot Learning — A Case Study (2)

In the previous blog, we looked into the fact why Few Shot Learning is essential and what are the applications of it. In this article, I will be explaining the Relation Network for Few-Shot Classification (especially for image classification) in the simplest way possible. Moreover, I will be analyzing the Relation Network in terms of:

1. Effectiveness of different architectures such as Residual and Inception Networks
2. Effects of transfer learning via using pre-trained classifier on ImageNet dataset

Moreover, effectiveness will be evaluated on the accuracy, time required for training, and the number of required training parameters.

Please watch the GitHub repository to check out the implementations and keep updated with further experiments.

Introduction to Few-Shot Classification

In few shot classification, our objective is to design a method which can identify any object images by analyzing few sample images of the same class. Let’s the take one example to understand this. Suppose Bob has a client project to design a 5 class classifier, where 5 classes can be anything and these 5 classes can even change with time. As discussed in previous blog, collecting the huge amount of data is very tedious task. Hence, in such cases, Bob will rely upon few shot classification methods where his client can give few set of example images for each classes and after that his system can perform classification young these examples with or without the need of additional training.

In general, in few shot classification four terminologies (N way, K shot, support set, and query set) are used.

1. N way: It means that there will be total N classes which we will be using for training/testing, like 5 classes in above example.
2. K shot: Here, K means we have only K example images available for each classes during training/testing.
3. Support set: It represents a collection of all available K examples images from each classes. Therefore, in support set we have total N*K images.
4. Query set: This set will have all the images for which we want to predict the respective classes.

At this point, someone new to this concept will have doubt regarding the need of support and query set. So, let’s understand it intuitively. Whenever humans sees any object for the first time, we get the rough idea about that object. Now, in future if we see the same object second time then we will compare it with the image stored in memory from the when we see it for the first time. This applied to all of our surroundings things whether we see, read, or hear. Similarly, to recognise new images from query set, we will provide our model a set of examples i.e., support set to compare.

And this is the basic concept behind Relation Network as well. In next sections, I will be giving the rough idea behind Relation Network and I will be performing different experiments on 102-flower dataset.

About Relation Network

The Core idea behind Relation Network is to learn the generalized image representations for each classes using support set such that we can compare lower dimensional representation of query images with each of the class representations. And based on this comparison decide the class of each query images. Relation Network has two modules which allows us to perform above two tasks:

1. Embedding module: This module will extract the required underlying representations from each input images irrespective of the their classes.
2. Relation Module: This module will score the relation of embedding of query image with each class embedding.

Training/Testing procedure:

We can define the whole procedure in just 5 steps.

1. Use the support set and get underlying representations of each images using embedding module.
2. Take the average of between each class images and get the single underlying representation for each class.
3. Then get the embedding for each query images and concatenate them with each class’ embedding.
4. Use the relation module to get the scores. And class with highest score will be the label of respective query image.
5. [Only during training] Use MSE loss functions to train both (embedding + relation) modules.

Few things to know during the training is that we will use only images from the set of selective class, and during the testing, we will be using images from unseen classes. For example, from the 102-flower dataset, we will use 50% classes for training, and rest will be used for validation and testing. Moreover, in each episode, we will randomly select 5 classes to create the support and query set and follow the above 5 steps.

That is all need to know about the implementation point of view. Although the whole process is simple and easy to understand, I’ll recommend reading the published research paper, Learning to Compare: Relation Network for Few-Shot Learning, for better understanding.

#deep-learning #few-shot-learning #computer-vision #machine-learning #deep learning #deep learning

Deep Reinforcement Learning for Video Games Made Easy

In this post, we will investigate how easily we can train a Deep Q-Network (DQN) agent (Mnih et al., 2015) for Atari 2600 games using the Google reinforcement learning library Dopamine. While many RL libraries exist, this library is specifically designed with four essential features in mind:

• Easy experimentation
• Flexible development
• Compact and reliable
• Reproducible

_We believe these principles makes __Dopamine _one of the best RL learning environment available today. Additionally, we even got the library to work on Windows, which we think is quite a feat!

In my view, the visualization of any trained RL agent is an absolute must in reinforcement learning! Therefore, we will (of course) include this for our own trained agent at the very end!

We will go through all the pieces of code required (which is** minimal compared to other libraries**), but you can also find all scripts needed in the following Github repo.

1. Brief Introduction to Reinforcement Learning and Deep Q-Learning

The general premise of deep reinforcement learning is to

“derive efficient representations of the environment from high-dimensional sensory inputs, and use these to generalize past experience to new situations.”

• Mnih et al. (2015)

As stated earlier, we will implement the DQN model by Deepmind, which only uses raw pixels and game score as input. The raw pixels are processed using convolutional neural networks similar to image classification. The primary difference lies in the objective function, which for the DQN agent is called the optimal action-value function

where_ rₜ is the maximum sum of rewards at time t discounted by γ, obtained using a behavior policy π = P(a_∣_s)_ for each observation-action pair.

There are relatively many details to Deep Q-Learning, such as Experience Replay (Lin, 1993) and an _iterative update rule. _Thus, we refer the reader to the original paper for an excellent walk-through of the mathematical details.

One key benefit of DQN compared to previous approaches at the time (2015) was the ability to outperform existing methods for Atari 2600 games using the same set of hyperparameters and only pixel values and game score as input, clearly a tremendous achievement.

2. Installation

This post does not include instructions for installing Tensorflow, but we do want to stress that you can use both the CPU and GPU versions.

Nevertheless, assuming you are using Python 3.7.x, these are the libraries you need to install (which can all be installed via pip):

tensorflow-gpu=1.15   (or tensorflow==1.15  for CPU version)
cmake
dopamine-rl
atari-py
matplotlib
pygame
seaborn
pandas

#reinforcement-learning #q-learning #games #machine-learning #deep-learning #deep learning