---
title: "Migrating from Catalyst"
output:
  rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{Migrating from Catalyst}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE,eval = FALSE,echo = T)
```


<center>

<img src="images/catalyst.png" alt="_" style="width: 600px;"/>

</center>


## Intro

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 another regular train loop.
Break the cycle - use the Catalyst!

## Catalyst with fastai

Specify loaders from catalyst dict:

```{r}
library(fastai)
library(magrittr)

loaders = loaders()

data = Data_Loaders(loaders['train'], loaders['valid'])$cuda()

nn = nn()
model = nn$Sequential() +
  nn$Flatten() +
  nn$Linear(28L * 28L, 10L)
```

Output:

```
Sequential(
  (0): Flatten()
  (1): Linear(in_features=784, out_features=10, bias=True)
)
```

## Fit

```{r}
metrics = list(accuracy,top_k_accuracy)
learn = Learner(data, model, loss_func = nn$functional$cross_entropy, opt_func = Adam,
                metrics = metrics)

learn %>% fit_one_cycle(1, 0.02)
```

```
epoch     train_loss  valid_loss  accuracy  top_k_accuracy  time    
0         0.269411    0.336529    0.910200  0.993700        00:08   
```